synthetic_transfer_learning.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<style>.container { width:95% !important; }</style>"
      ],
      "text/plain": [
       "<IPython.core.display.HTML object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "from IPython.core.display import display, HTML\n",
    "display(HTML(\"<style>.container { width:95% !important; }</style>\"))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Stimulus-driven predictive model of brain activity.\n",
    "What we will need:\n",
    "- \"imagenet_trained_model_weights.pkl\"\n",
    "- The betas and experiment stimuli\n",
    "- A brain mask"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "import src.numpy_utility as pnu\n",
    "from src.file_utility import save_stuff, flatten_dict, embed_dict, zip_dict\n",
    "from src.config import *\n",
    "\n",
    "%matplotlib inline"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "#device: 4\n",
      "device#: 0\n",
      "device name: NVIDIA A100-PCIE-40GB\n",
      "\n",
      "torch: 1.10.0+cu113\n",
      "cuda:  11.3\n",
      "cudnn: 8200\n",
      "dtype: torch.float32\n"
     ]
    }
   ],
   "source": [
    "import torch\n",
    "print ('#device:', torch.cuda.device_count())\n",
    "print ('device#:', torch.cuda.current_device())\n",
    "print ('device name:', torch.cuda.get_device_name(torch.cuda.current_device()))\n",
    "\n",
    "torch.manual_seed(time.time())\n",
    "device = torch.device(\"cuda:2\") #cuda\n",
    "torch.backends.cudnn.enabled=True\n",
    "\n",
    "print ('\\ntorch:', torch.__version__)\n",
    "print ('cuda: ', torch.version.cuda)\n",
    "print ('cudnn:', torch.backends.cudnn.version())\n",
    "print ('dtype:', torch.get_default_dtype())\n",
    "#torch.set_default_dtype(torch.float64)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 37,
   "metadata": {},
   "outputs": [],
   "source": [
    "timestamp = time.strftime('%b-%d-%Y_%H%M', time.localtime()) # 'Aug-10-2020_1638' #\n",
    "\n",
    "model_name = 'gnet8r_pred'\n",
    "model_dir = '%s_%s' % (model_name, timestamp)\n",
    "\n",
    "output_dir = root_dir+'output/multisubject/%s_%s/' % (model_name, timestamp) \n",
    "input_dir = root_dir + 'output/multisubject/gnet8r_mpf_evc_Feb-05-2022_2105/' # 'gnet8r_mpf_evc_Feb-04-2022_1844/'\n",
    "\n",
    "stim_dir = root_dir+'../../data/nsd/stims/'\n",
    "voxel_dir = root_dir+'../../data/nsd/voxels/'\n",
    "\n",
    "exp_design_file = root_dir+\"../../data/nsd/nsd_expdesign.mat\"\n",
    "\n",
    "\n",
    "if not os.path.exists(output_dir):\n",
    "    os.makedirs(output_dir)   "
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Load the fake voxel data"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "dict_keys(['Bs', 'Rs', 'brain_nii_shape', 'pred', 'val_cc', 'voxel_index', 'voxel_mask', 'voxel_roi'])\n"
     ]
    }
   ],
   "source": [
    "## load \n",
    "pred_set = h5py.File(input_dir + 'V1_4_pred.h5py', 'r')\n",
    "checkpoint = embed_dict({k:np.copy(v) for k,v in pred_set.items()})\n",
    "pred_set.close()\n",
    "print (checkpoint.keys())"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [],
   "source": [
    "voxel_pred = {int(s): v for s,v in checkpoint['pred'].items()}\n",
    "voxel_roi  = {int(s): v for s,v in checkpoint['voxel_roi'].items()}\n",
    "voxel_mask = {int(s): v for s,v in checkpoint['voxel_mask'].items()}\n",
    "voxel_idx  = {int(s): v for s,v in checkpoint['voxel_index'].items()}\n",
    "val_cc     = {int(s): v for s,v in checkpoint['val_cc'].items()}\n",
    "Bs = {int(s): v for s,v in checkpoint['Bs'].items()}\n",
    "Rs = {int(s): v for s,v in checkpoint['Rs'].items()}\n",
    "subjects = list(voxel_pred.keys()) "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [],
   "source": [
    "from src.roi import roi_map, iterate_roi\n",
    "\n",
    "block_group_names = ['V1', 'V2', 'V3', 'hV4']\n",
    "block_group = [[1,2], [3,4], [5,6], [7]]"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Load the stimuli images\n",
    "From the subjectwise image preparation"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "--------  subject 1  -------\n",
      "block size: (10000, 3, 227, 227) , dtype: uint8 , value range: 0 255\n",
      "--------  subject 2  -------\n",
      "block size: (10000, 3, 227, 227) , dtype: uint8 , value range: 0 255\n",
      "--------  subject 3  -------\n",
      "block size: (10000, 3, 227, 227) , dtype: uint8 , value range: 0 255\n",
      "--------  subject 4  -------\n",
      "block size: (10000, 3, 227, 227) , dtype: uint8 , value range: 0 255\n",
      "--------  subject 5  -------\n",
      "block size: (10000, 3, 227, 227) , dtype: uint8 , value range: 0 255\n",
      "--------  subject 6  -------\n",
      "block size: (10000, 3, 227, 227) , dtype: uint8 , value range: 0 255\n",
      "--------  subject 7  -------\n",
      "block size: (10000, 3, 227, 227) , dtype: uint8 , value range: 0 255\n",
      "--------  subject 8  -------\n",
      "block size: (10000, 3, 227, 227) , dtype: uint8 , value range: 0 255\n"
     ]
    }
   ],
   "source": [
    "exp_design = loadmat(exp_design_file)\n",
    "ordering = exp_design['masterordering'].flatten() - 1 # zero-indexed ordering of indices (matlab-like to python-like)\n",
    "\n",
    "image_data = {}\n",
    "for s in subjects: \n",
    "    image_data_set = h5py.File(stim_dir + \"S%d_stimuli_227.h5py\"%s, 'r')\n",
    "    image_data[s] = np.copy(image_data_set['stimuli'])\n",
    "    image_data_set.close()\n",
    "    print ('--------  subject %d  -------' % s)\n",
    "    print ('block size:', image_data[s].shape, ', dtype:', image_data[s].dtype, ', value range:',\\\n",
    "           np.min(image_data[s][0]), np.max(image_data[s][0]))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Examples"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 288x576 with 24 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "n = 1003\n",
    "plt.figure(figsize=(4,1*len(subjects)))\n",
    "for k,s in enumerate(subjects): \n",
    "    for i in range(3):\n",
    "        plt.subplot(len(subjects), 3, 3*k+i+1)\n",
    "        plt.imshow(image_data[s][n+i].transpose((1,2,0)), cmap='gray', interpolation='None')\n",
    "        plt.gca().get_xaxis().set_visible(False)\n",
    "        plt.gca().get_yaxis().set_visible(False)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Resample, ordering and split"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [],
   "source": [
    "def ordering_split(voxel, ordering):\n",
    "    data_size, nv = voxel.shape \n",
    "    print (\"Total number of voxels = %d\" % nv)\n",
    "    ordering_data = ordering[:data_size]\n",
    "    shared_mask = ordering_data<1000  # the first 1000 indices are the shared indices\n",
    "\n",
    "    val_voxel_data = voxel[shared_mask]    \n",
    "    val_stim_ordering  = ordering_data[shared_mask]\n",
    " \n",
    "    trn_voxel_data = voxel[~shared_mask]\n",
    "    trn_stim_ordering  = ordering_data[~shared_mask]\n",
    "    return trn_stim_ordering, trn_voxel_data, val_stim_ordering, val_voxel_data\n",
    "\n",
    "def zscore(a):\n",
    "    m = np.mean(a, axis=0, keepdims=True)\n",
    "    s = np.std(a, axis=0, keepdims=True)\n",
    "    return (a - m)/s"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### No-noise model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {},
   "outputs": [],
   "source": [
    "noise_model = 'no_noise'\n",
    "n_repeats = 1\n",
    "\n",
    "voxel_data = {}\n",
    "for s in voxel_pred.keys():\n",
    "    noise    = np.random.normal(0,1,size=(voxel_pred[s].shape[0] * n_repeats, voxel_pred[s].shape[1])).astype(np.float32)\n",
    "    signal   = np.repeat(zscore(voxel_pred[s]), repeats=n_repeats, axis=0)\n",
    "\n",
    "    ordering = np.repeat(np.arange(len(voxel_pred[s])), repeats=n_repeats, axis=0)\n",
    "    voxel_data[s] = signal"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Correlated noise synthetic model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [],
   "source": [
    "noise_model = 'corr_noise'\n",
    "n_repeats = 3\n",
    "\n",
    "voxel_data = {}\n",
    "for s in voxel_pred.keys():\n",
    "    noise    = np.random.normal(0,1,size=(voxel_pred[s].shape[0] * n_repeats, voxel_pred[s].shape[1])).astype(np.float32)\n",
    "    signal   = np.repeat(zscore(voxel_pred[s]), repeats=n_repeats, axis=0)\n",
    "    \n",
    "    ordering = np.repeat(np.arange(len(voxel_pred[s])), repeats=n_repeats, axis=0)\n",
    "\n",
    "    #voxel_data[s] = signal *  B.reshape((1,-1) + noise * np.sqrt(1 -  B.reshape((1,-1)**2)\n",
    "    voxel_data[s] = signal * B[s].reshape((1,-1)) + np.dot(noise, R[s].T) "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(10000, 4627)\n"
     ]
    }
   ],
   "source": [
    "print (voxel_data[subjects[0]].shape)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[<matplotlib.lines.Line2D at 0x7f19a8d80ba8>]"
      ]
     },
     "execution_count": 18,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAZQAAAEUCAYAAAAMdcB4AAAAOXRFWHRTb2Z0d2FyZQBNYXRwbG90bGliIHZlcnNpb24zLjMuNCwgaHR0cHM6Ly9tYXRwbG90bGliLm9yZy8QVMy6AAAACXBIWXMAAAsTAAALEwEAmpwYAABDp0lEQVR4nO2deXgV1dnAf0kIhEAIu6wKUXIEBAQEN0BFrEtb22qrLa17bdVat1Y/+9Vq1Gpt/erS1mrrUrfiUq1ardWqKC4ICKgswglLkH0nhAQCWe73x8xNbm7m3jsz98xdkvf3PHkmd84yZ87MnPe857znPTmhUAhBEARBSJbcdBdAEARBaBuIQBEEQRCMIAJFEARBMIIIFEEQBMEIIlAEQRAEI4hAEQRBEIzQId0FyDSUUn8FSrXWJwZ8nTOAf8cIHqW1XpJk/im5D0EQhDAiUCJQSl0CXArMSsHlRgIh4CKgISpsXTIZp/g+BEEQABEoACil8oBfAmUpvOwRwFqt9ROmMkzTfQiCIAAyh4JSqgBYCNwKPAVsSNGlRwLLvCRQSoWUUmUxwtJ1H4IgCIAIFIACoBtwrtb6AqDeKZJS6lil1FtKqT3233+VUhP9XFAplQMMB76wfxcopZLVFl3dhyAIQlCIQIEqYJjW+vlYEZRSp2DNRxQDvwJ+DRwMvK+UmuzjmiVAIXCoUmoxsBeoUUo9q5Tq7SM/cHEfgiAIQdLu51C01o1AY6xwpVQu8BAwDzhBa91gn/8T8BnwB2Csfa5fgstVa62rsYa7AI4B7gIqgEnANcBIpdRErfU+pVQhluCJpDBS6Gitt7u5D0EQhKBp9wLFBWOxNIoHgR5KqciwV4FrlVIDtdYbgE0J8roDuAlYDdwG/F1rXW6HvaKUWoklvC4B/gTcANwSlcf19l+YHM93JAiCEAAiUBJzqH282/5z4mCsSfBLE+T1KYC9xsRpncljWILkJPv4JPBhRPhbWBPuT7opuCAIQioRgZKYPPv4K2BOjDjLAbTWjyRzIa11nVJqF9DV/r0aS5sBwNaOVmut307mOoIgCEEgAiUxa+xjdXRDrpSaAPQE9nnJUCl1O/B9YIzWek/E+Z5AH6w5FUEQhKxCrLwSMx9rbuQqpVTX8EmlVDfgeeBveDfRXQsMBX4Ydf5m+zjDX1EFQRDSh2goCbCHoa4CngMWKqUeAWqx5ksOAb6vtfYqUB630/9OKTUMWAxMA84C/qK1fj9GWWQCXhCEjEU0FBdorV8AvgKsx5pLuR1r3ceZWutnfORXB5wGPIolRO4DRgDXAVeYKbUgCEJqyQmFQukugyAIgtAGEA1FEARBMEK7nENZsGBBJ2AC1mR7tOt4QRAEwZk8oD/wyfjx4/dHB7ZLgYIlTD5IdyEEQRCylMm0XHQNtF+Bsglg0KBB5Ofne05cUVHB0KFDjRcq25B6sJB6sJB6sGjL9VBXV8f69eshhpup9ipQGgDy8/N9CZRwWkHqIYzUg4XUg0U7qAfHqQKZlBcEQRCMIAJFEARBMIIIFEEQBMEIIlAEQRAEI4hAEQRBEIwgAkUQBEEwggiUdsznK3cwvWwmO6taLXgVBEHwjAiUdsw78zcAsGpDVZpLIghCW0AEiiAIgmAE4yvllVJDgd8DJ9qnXgN+prXeliDdPCwfW9G8qLX+drL5C4IgCMFiVKAopXoB7wIdgd/a+V8PjFZKTdRaH4iRLgdrg6mXgRejgr9MNn9BEAQheExrKNcBg4BRWutlAEqpucBbwAXAwzHSDQG6AK9orZ8OIH9BEAQhYEzPoXwXeC/c2ANord8GtB0Wi5H2cVmcOMnkLwiCIASMMYGilOoBlAALHIIXAuPjJG8hUJRSXQznLzgguz8LgmASkxrKQPu4wSFsE1CslCqOkfYIYA9wj1JqD1CtlFqllIrUOpLJv92zfmsNn5ZvT3cxBEFow5icQymyj3sdwvbZxy7AbofwkXb67sD59vFq4BmlVL7W+qkk83ekoqLCbdRWlJeX+06bDspmrLeO0wc1nauuqQZg48aNlOft8pVvttVDUEg9WEg9WLTXejApUHLsY7yBlFhhfwXytNYPhE8opZ4FlgB3K6VmJJm/I0OHDvW1EU55eTmlpaWe06UXS6BElrtoYS1Qy4ABAygt7eOYavmXlQzoXUi3Lh1bhWVnPZhH6sFC6sGiLddDXV1d3I64ySGvavvY2SEsfM5xSbbW+qFIYWKf2wc8BRyEZVLsO3/BP7f9bSG3/W1huoshCEIWYFKgrLWP/R3CBgCVWusaj3lutY9dA8pfcMHG7U6jjIIgCC0xJlC01pVABTDOIXgsMN8pnVJqoFJqqVLqZofgw+1jhd/8BUEQhNRgeh3Ki8A0pVRYEKCUmgYo4FmnBFrrDViT8JcqpbpFpDsYuBB4V2u92W/+giAIQmowvVL+d1hWWu8opX4PFAA3YK0deRpAKVUCHAfM1lqvttP9BHgJmK2UehjLoutKoN4Oc52/4B5ZhyIIgkmMaii2g8YpwOfAbcA1WP65TtdahzfdmII12T4lIt3LwDeBGiwfXT8DPgaOi1oV7yZ/QRAEIQ0Y9zastdbAGXHCHwcedzj/CvBKsvm3VeZ9sZX7nl/C7396DP17Faa7OIIgCK2Q/VCyhDlLLYO3NZv2pLkkgiAIzohAacfk5CSOk0r+OauCS+96P93FEATBJ8aHvIRgacsT6S+8698VjiAI6Uc0FEEQ2hy/efJTPly0OXFEwSiioWQZJoep2rK2I7RvFq/exeLVu5g0ul+6i9KuEA2lDdPYGOKx1zRbdorrlGxi7tKtvP1Jy10aNmyr4fHXy2lsbPu9gEUrd/CvD79MHLGd858561igM2tLCtFQsgwvWsXqjVW8PX8DqzeKz8xs4v5/LAFg2oSBTef+75lFbNm5j1MnDqJ/77ZtNn7X058DcOakQ9JckszmqTdWADCjbGqaS9KMaCgGeWPOOjbtyDxtICfTzLkE/wTwKOcv38aS1TuN5vnRos2sWOd6ayLPLP+ykjlLtgDwweebWLVBOk2ZgAgUQxyoa+DJN1Zw62NOOxQLQuZyz7OLufPJz4zm+cA/v+CWR4P7Fm7720L+8MJSAB58aRm/elh8w2YCIlAMs29/Q7qLILRl2v4UipDFiEBpx8hIWPaQike1t7Y+BVcR2jIiUEyT5h7kgboGlla03B8+FGMmX8yGs4dUPKpGHy/E2s3V7NhdG0BphGxEBEob47HXNHc88SmbZJfFtkmGaZU3PjSPn947O93FEDIEEShZzoG6BjZHWJat32btgrx3f/PwRbZZee3bX8+2yn3pLkbKqNlX59jL31m1n03b97K7+kDzSdEqhQxG1qEYJpTiL/7P//yCecu28fgvT6Bjfl5zObK44bnlkQWs31aTUfb1QXLtHz6mel99q/u98p6Pmv7v17NzqoslCJ4RDcUQqdMCWkqKRaus9QP1De4kSDastA5rWWEO1DWw/0DbtZ6r3ieT4ULbQASKIWJNfAeNVzH2wnvZ59H3mvs/5qI7Z6W7GGkl87sBgiACJQvxrglFCru59kZd2URl5ByCIAgZiwgUQ3gd8rr5kflcfveHBi5sHdKlIQFcfd9sfvHgPBobQ0wvm8kL765uET5n6Vaml80U89IkiHy7zrvtXe6e8XnMuAfqGpheNrPpb8vO5AwcppfN5I8vLEkYb+eeeqaXzTTmsLB6X13T/zt21zK9bGaTuxWvTC+byb3PLTZSLtPsrbXq7c2569JdlKQRgZImVq6vYneNn553S8GRYzc1rcVJ85lIYRfEVM+2ylq+3FJNgz0/E+0p9v3PNgGwdku1+Yu3QxoaQ3xaviNmeLS3hgoD20Z/vCSxZrthh/U+z15sZh+SnVX7m/4Pvzvvf+4/70+WbUu6TEFQWW3d53/nbUgQM/MRgWKYdFtXJdKUYpVvZ9V+ppfNZOV6/w79vGpJqzdUMb1sJtsr/WkuG7fXML1sJhuiJvGDZuX6KtG4UkzYluSzFbEFqQn+96F5PPPWykCv0ZYRgWKI1K30COZKYW+zb32SfC8px2UZ31lgXWvRKn+NRLjX/LHPYRC/vD1/PQCLDXvoDYx093IM0NDQmJLrrNlczasfrU3JtcK0gcfThAiUrCPq7ctxPh35knrRHEy93A++9EXTUFetvcgy2hzYFGn7IDO0IYh+3iZ79Tc+OI/pZTON5ReXDK1fITYiULKEWCNZMeRJVGhqaGrHcuCDzzfz0MvLAFi+1hpGe/6d1TFS+iP+vQthkpl3iEbmwYR4iEDJcsxPsiffPKdMjDVJlMwQKf+evZZla3YljpgEs2ytLx5+F9mm00HP8i8rZdvfNoC4XjFEupu08DBHy0bBfalymsyPjRUpcNKlocSaI/r7f63J3CBdxgTZ6Kbz0d/2t4VA1La/2eWCTkA0lMDYsK2G9z7dCMDmHXuZucB5svvl99dQV9/Iy++v8eleJPFX57fHanptSygEr3ywhn37W7oa8Wou2dgY4l8frGGvw2Zm7326MRCrrw3bapj1aUvtYOP2vU3POBH1Dc3PuLExxMvvr2maWzLFO/M3sGVncF6m3e6XEsvEuKGh5XseCoV49cMv2bO3zjG+Vz4t325UQ9y4vYb3Frp7vn55c+76FubRfpi/PHPMoUVDCYgbH5xHQ2OIE8cO4KaH57O3tp6p4we2ivf8zNUs+7KSxat2sm9/Pd875TBP14klK0KkvoOXSPw0hkI8985qtu/ezyVfU03n126pZtce9x/V4lU7eTZiPiZS7v31leXk5ebw1M0nuc7PDTc8MJcQcMLY/k3nXpttWQOVTR+UMP2sTzfx/MzV1B5oYFDfLjw/czW7aw5wwemlRsrX2Bji0dc0RYX5/Pbyib7ySPS+PPXGCl/5hnn/s808P3N103u+/MtKnnl7FStj7QffwrAkcf53z1gEmNMQb3xwHvUNIU4cN8BIftFU7tnPE/8pT3rY+p5nF2eMI1XRUJJg1579fFq+nU/Lt7dqEBsinDAm6tnV2CuC99e1NI1ctmZXC9f08Yj3wbnRNGIvkIxPdWTvMjzsluADceqZe3FaOW9Z/EV2DQE4wIzM0W0DsLRiV9Mq9QP2s317/oamHnq0Rmpi4V3k6vJotu7a12Qe7oelFcmZSe+vs+/brou6eusYrbE6kY4dGOI5XK1vaOT9zzYl/Lb2H2jgo0XORhHh1zScRTL3+EWF+7YiSIxrKEqpocDvgRPtU68BP9Nax/1alFKnAjcB44FGYA5wk9Z6TlS8ecAEhyxe1Fp/O7nSe+PWxxawdZe1uK17146A2TmI2x//FIjf44r/DrYOjfkB+HyZ73veyZ1FcF9/Q0Mj7y5MPDGdCdzxROvnt7e2ntc/dnaxce9zi/nLDZOTu2go9hDnNfd/3Ko8Xti+O7mhmTBtYWrkXx9+yQvvVtAhL4fjRvWLGe+pN1cwc8FGenbrxPAhPQIrz68d3rV0YFSgKKV6Ae8CHYHf2vlfD4xWSk3UWjv6GlFKnQD8B1gK/NJOdwUwSyk1WWs9z46XA4wAXgZejMom5SYiYWEC3hwYRm/RG0lVzQEqqw9w8EFd/RUq2a/Vo0DcFrHKPehJ3ZXrdzOkf1Gr86n0Y+bHL1Z0z3HP3tjvSqwFfNm2SZpfYt1/rEdcvs6/Z4dkCG96lmjrgfD8SO2BBrZV7qOxMcRBPQsDL1+6MK2hXAcMAkZprZcBKKXmAm8BFwAPx0h3H7AOOFprvddO9ySwDLgDOMWONwToAryitX7acNlTwvzl27jnWWcndTlYY/VVe+vc9zTiOodsfS4TGia/zf/NjyzgwZ9PMloWr1z7h485MWIexQ3X/XEO5506rOl33OHJWOc9LU7NDlO9iGVLTbz0/pq4caMpe3SBwRIFy9X3tdQQ0/8lmsf0HMp3gffCwgRAa/02oO2wViilegBjgOfDwsROtwWYBRwXEX2kfVxGhpJox8Z4/p9CQFUCi5fotqLG7iG1umrEiUghkqixOVDfkFE+qqLLGx53d8P2yloO1GXGxlyR70VTQ5qhLcq2yn3U2/VsclzejZjbuD398wCZTF19o2/fd6nAmECxBUMJ4NRlWIg1N+JEFaCAex3CegOROmULgaKU6uKrsG2I6AlopzYqkRDJyWlON3/5dn567+ykypSwofTQgf4gapW3k8B2ur3GxhBX3TebB178wv3FBPbtr+fq+z7mkVeXs3ZLNdf9cU7iRDYm9KIsUa6aSHV5H/jnUq66b3bKfJt5xaSGEraJdVpUsAkoVkoVRwdorRu01iu01i0MvpVSo4HjgcjW7QhgD3CPUmoPUK2UWqWUctR+sg03HVZvvVr3kZ3y9fqxJPNxxUq7emNi1+tOSRvtDBeWu9+b40BdQ2Afaov7a/rfqdKd07sdqgx5iOtE2BrtsxU72LYruX1UoolZqhSrarUR1nUNDY0Jtd7aFG0/3dAYSqhRL7T3mollyFjvQYMPApMCJTxb6qSzht9MVxqFUqor8KT9866IoJH2dboD5wMXYwmYZ5RS53ksbyC49bTrl5iNdgCNeSTrt7pfLBhZA19udrcXR6KhwviJHdL6yO7CO2Zx19OxN67KZCK1UBNzKH5yMKGYxpQthlSB2Yu3cPGds1i72fJJdutjC7ng1+/FTXPxnbPYuD25xbJuir9l5z4uvMPdVtexnvH5Ce4laExOyrvxhJGwWpVShcC/sOZVfqO1jqzhvwJ5WusHIuI/CywB7lZKzdBau+5OVFT431+9vLzc8XyIUIuw6P+3bG3dwNbWWmOiuyorY6bbs8dKt2nzZsoLWi8EW7V6Nd0K89hn57V23Vpy7a8znD/Agbq6pryra6yPauPGja1s7svLy5t6+QBzP1vB3iGtrVPq6prnfFautFyPNIYaI9K13luiak8V5eXl7N7dfB9Oz6K8vJzKypYWcU7xdu7a1ep5lK+wFuFFP49ELK1onVc0u6taWxbFeuZhtm1rtppvaLRe0d27d7eKu2q1swPN/QdaW4ZFpo0c+nTKI1H5AFauWtW0HqihoYENG2OvEm9V3zHyjDy/zX73d1VWUl5ezvqN1nu5t6a5sa7a0/x9rPlyDbVVljn+xk3N/dREzyfevc5aYL0/cz9fQW1Vl6ZFlYnu55PPVzJ8cOem35X2t7p161bKy1v3ocPpa+x727BxQ6uwqr3OTVW8+wsLkhUrYu/Z4uV9N41JgRJ2Q9rZISx8LsaSWAulVHesdSvHA49hmRA3obV+KDqN1nqfUuop4BYsk2LX+3wOHTqU/Px8t9GbKC8vp7S0FFjvGB4ZFvl/2Yz1jCvt3Sp+QUEBUEeP7t0JV2N0HktmWC7D+/frR2lppN27FefQkhJ6dOtE5/ergAMMHjyYvNxcYGtT/gD5+fl8vr4DL72/hvGqN1DLgAEDbLW/eeFaz74H07NbJ8IjmA15RZTN+JKyi8czbHA3vn/ru5w4qhsdO+ZDjfVh3PWC1QDl5uYC1rnlm1r3IYqKiigtLaV4+TLA+uD2NBQDLedLymasb9VjLRla0ipejx49KC0NW1FZ9fHr56xy55Bj16UbIp+ZcxhAcbdiIhXxWUuqeHdR86vdnL45TZ8+fQBLEO2vs+qkY0EXyma0fIdKSkqwRohbsnNPa/PUNxcd4KffPgJo6ePr9y+1Th95nfD/j9w4hcKCDk3lfOOz/Vz+reHAJvbub2TAgAGAs+v76HssLS3lty++1yre7JXw33nrmVE2lVU71wG76dG9O6WlpezL2QFsp0uXLoBlXlvUtYjwgMarn1Szs2o/j/7vCew4sIXw+xnv2wN44/Nm4XvwkEMp6JjXFH/xGuu5vfTxLiaMPozWz7z1dwswYMAASkv7NP3+aIUGaujbty+lpc2eEp54vZw37fsF6PLJPqCWQQMHEq7L8LV2Ve3H6Vl37TmQAb2dB3NyczfS0NjIHc/HdlfUoUs/SgZ2ixmeDHV1dXE74iaHvMK70jjZVA4AKrXWMfVGpVRfrDUsx2NpIj/UWrvVc8NLp30u3kgdXsb0vdBsOZTjcLYlsUwzI4neNnbRKutjjlyl/t7iuP0DwPIi64YPY6wmjtbskxoaCwg39eCEkzWdl+GqSJ9ZL7t4ptFsj7r+5yuT2zdl3/7W4/f/nefQ8LscFd6wfS/7fMxfzFnaXC9VcbbZ9nK/bp/Lm073S4whrxj14PabiUVQbYwbjAkUrXUlUAGMcwgeC8yPlVYpVQS8CRwJ3Ku1/nG0MFFKDVRKLVVK3eyQxeH20f8Ylil8tHdBWIq4yTO8wPIvLy9rpQnc+9xiZi9OvBOinzmjpO7XabokxTImuq4ywTLJVxlSWO5FK3e03gQuHLYquJ0vM9Qy2zXL1uzi4jtnxXWp48S2yn1ccPt7rN+a2v1rTK9DeRGYppQKN/AopaZhmQU/GyfdA1jC5H6t9XVOEbTWG7Am4y9VSjXpc0qpg4ELgXe11uZ2EkoHxt/++BmGrVf2xvCl9Ohry5v+b7kDZKKrBvgZZ3sLEUEmLDKNh0lB+fzM5DZXywShnQxeHnXk9/Py+2uoPdDAqlgONGPwybJt1KXBTZHplfK/w7K+ekcp9XugALgBa23K0wBKqRKsxYqztdarlVLDgfOASuAzpdQPojONWBX/E+AlYLZS6mEsi68rsdaq/MTwvaSMnATmDF4cJ4Z55NXlTBzeJ3HEcBkShMcqwRYn09IUt5OhUIhQKMQf/rHUc9rNO/Zy3R/ntNyHw6Z2fz33PLeYi7+qHFLG5tePL+Tqc0ZFldFdWq890Ygr+EzXErdCrqrmAH/4xxJPeb/+sb+92rfs3MufXvT+bLOVx/6tKSzowNEj+5KbazttDTuQ9JnnsjW7eHPeeo4f1Y/PVmzn0jOHmylsFEY1FNsB5BTgc+A24Bosv1una63DnuWmAE/ZR4AT7GN34G92WPRfOP+XgW9izeT+FvgZ8DFwXOTq/LaGF9fuYTZu38vLHyTn3iyyEQxaRrjugcaIV3uggblfxPdC7MQzb68CnDeu+mzlTpas3sWz76zylOcXaypd75MSTfRCzlTj9jnPXLCRL9ZUus43FIIdtl8rr+/Sc8lsHZ0hSqAXDauhMcT9trAOC3g/ncpI/m/GIuZ9sY17n1scqNZi3Nuw1loDZ8QJfxx4POL3Q0Ar66046V8BXvFfwuSpawjxxOvmTPMi92JPhjfnrjfqLM9xLZ4L/NyG23JHGwuEcev6vaGhkRlvreLMSYdQbHuITgUz3opt5hmJ31eg0cd6NicDh8+SnJj3dH2HVtZPp8AvpmXN7ggHsYtW7mBH1X6WxRC6K9fvdvXOhk3/P1y0mXGqNwfSvHAxEbLBlg8WrqzhzQWV6S5GC0KhEE/8x4WQiyUZHIY6Wpxx3GLYH8mMh9/3vPMwy0Mvu1NQP1u5k//MWcf23bVce+6oxAlSjN+qqTO0wv/P/xRXNd5ofmKPvNo85xi9QDb687r5EXdOLcPp5izdylXf8VdCSJ39hWyw5YPGOC1iCHhjjvN+F16488lPm/7/bEVzr/Ht+Rta9IS84jjnEYPIu9zqwSFdjcutYk05AqyLsxESWMNh/569lsbGUFOvuMFO49RLDMcJb/k774voOGb6tqbm5GNtL52IRIJ9q0HXK5GXmr1kC5tStBlU+drYmq+XRvY1+/2JR7yN3SK/YS/kRrwkXupswXLLdPiNOescNcFkh9BiIQIlAJ6M2CrVjf26U4wlq5tXiD/6mm76X6/dzb3PuV676ZpEbVvkrpOm3KPrOB+7F96ZH7tBzcmB595Zxd//u5J5X2x1JQpWrK/iQF1D0usyUkFVzQEeeVUnjuiDv//X3TCdG3ZWNXdIqmrquOGBucbyjscDhjSuleureP/z+HMP8d6ttz7xJ/QjOx1e6myZvZYlhPO6lo8WBzNXJwLFB0GZMCbaKjiMV0sgNw4WY81NRLN+217WbE7Otj1EiNUezSD9Ut8QalpAGDn+HG+BZH19Y0zne27xazrt1Tw0me2OP03hArjoHrGXcqfDujpycWSYA3WNbKvcx0uzKpoExJylW5s7WEGUMyJPv8+6viHUqs06UBfMXIzMoQSMG+ETfiHTaeHjZOXkxOcrdxjpud/0cMx1rsaZv9y54YynacVrG9w0cH5X9HvV2pLZ7+Uf71Yw5Uhvm4Wlg6AdrjoRyyT6mvs+bvFk9Vprcn3iiL6pKZhvWr6PQXWKRUPxwYYd/ucwTJAhlpBZzaflO9gfw62H5QI++DLsSbCZmhuinXp6xY9Juin8NmqmLBn9PGKnIq/flpwn4lSQqoWhIlA8snZzNUvXmt0not2RzlXPEZLijy8Et1jOTa/ahFFCsg2FW2ujZDHZoKVr299Y9/DCu5bHpyA0KRN5OmniQXWYRKB4pGpverWTSALfqjfL3V1EE+0MURtcs5M+suMhxbOMTEiGqOR7Enz7QTTSpvKMXr8iQ14ZQiDvto+Hu3XXvqS36k1EUJ5957lchGiaF96t4HM35ptZ5DgqIOvPhHjdxXDffv9zPTW+3dGY5Z+z1qS7CBmPCBSveOwy+DUbTlSGaO0kGWufmJfJlK6hQVa6sKIKkfzcRKoIaj1BIvz7HGvGrdbi1voxWdJVl+lAhrwyhFyPDyJVDdM1939sPM9M3HskWdZEmEfHa6h+eNf7McPcvAKpMnVNaigpCWYu8OerLJJ7nnW3nipVXpl/+ddPUnKdVLPOYftuGfLKELy+3G5873hdKNj29AbBN21P5qeNL5NcX5UpuxFEt1Er16dmzReIQAkcNxPnTj0IIX0kku9u1uGYXGUej0wZpQlyWCpZM+HpZTMdz28w5PoHYK7DQshYxCpPNBu21TgusEyEKU8WfhCB4pFcj92QFevbgiWREMn23elbuxFNuoa8ojHp9ytV+PWv5cQbc9cZn3PMBtc/0YhA8UimqLWpIFsmptsz6VqT0RZoq9+yVws8k4hACZhAFju10Q9ByF4yRFHyhMlvM4hP0m+drnHply8IxJeXIAjtksiOWVVNcguWl6/dTbfC/CRLZIbonTSdNi0LaqhUNBSPZIp2kI09wmwhnZOaQnp4/ePk9zCqMuCbLZIg25ptAc15iUDxTIZIFCEwRJx459WP3HmrzlS8WGmliiD7NUFlLQIlS5FGT8gk/Ji3pptIDcDLTqZtAVnYmCF4VUOdxi+TZd3WGhmWEYQk2bUncxy9OvHfeesDy1vmULKUpRW7EkfywYeL0rcZV5tHZLWQAWyrDNCbuGgoQiS1SXhvFQShfSNzKBlCpkzJB+FdWBCE9kFQQ+YiULySIRJFplCCQ6pWaOvIpLzQgrboWl4QhNQgGoogpIhqwwvUBCHTkDmUDKEt7mIotOTPL32R7iIIQqD0LOoUSL4iULIVGfESBMEnA/t0CSRf484hlVJDgd8DJ9qnXgN+prXeZiKd3/xNkSm+vD41uJeDIAjti6wY8lJK9QLeBY4BfovV8J8JvKWU6phsOr/5C4IgCMFjWkO5DhgEjNJaLwNQSs0F3gIuAB5OMp3f/AVBEISAMT2H8l3gvXBjD6C1fhvQdliy6fzmLwiCINgENXJvTKAopXoAJYDTnqQLgfHJpPObv2kyZQ5FEATBL9kwhzLQPm5wCNsEFCulipNI5zd/QRAEIQWYFChF9nGvQ1h4swEnWzW36fzmLwiCIKQAk5Py4cGgeNqUU5jbdH7zj0lFRYWX6ABs3pXZeygIgiAkYtOmjZTnVxrP16RAqbaPnR3CwueqkkjnN/+YDB06lPz8fC9JKNhcDWTf7nSCIAhh+vcfQGlpX8/p6urq4nbETQ55rbWP/R3CBgCVWuuaJNL5zd8otQdkHxJBEAQnjAkUrXUlUAGMcwgeC8xPJp3f/E2zaJWsUBcEQXDC9DqUF4FpSqnDwyeUUtMABTxrIJ3f/AVBEISAMb1S/nfA+cA7SqnfAwXADVhrR54GUEqVAMcBs7XWq92m8xgvMHJkIYogCIIjRjUU20HjFOBz4DbgGuBl4HSt9X472hTgKfvoJZ3reEEi4kQQhGxnf0BzwTlB7dyVySxYsGAIUOHHyuulWRX8413v5saCIAiZxIyyqZ7TRFh5DR0/fvya6HDZD8UjMuQlCILgjAgUj4g8EQQh2zmkX9dA8hWB4hGRJ4IgZDt5ucG0ZCJQPCJDXoIgZDtBtWMiULwi8kQQhCwn4/dDaS/kioYiCILgiAgUj4g8EQQh6wmoHROB4pEcGfMSBCHLkSGvDEE0FEEQBGdEoHhErLwEQch2xMorQxB5IgiC4IwIFI+IPBEEQXBGBIpHZMhLEIRsJ6hmTASKRyYM75PuIgiCIGQkIlA8Uty1Y7qLIAiCkJGIQBEEQWhnyDoUQRAEwQxiNiwIgiCYQDQUQRAEIaMRgSIIgtDOELNhQRAEIaMRgSIIGUy3LvnpLoLQBhFfXoLQDvnBqcPSXQRBcI0IFEHIYDrkyScqZA/ytgpCBpMrruOELEIEiiAIQjtDrLwEQRCEjEYEiiBkMLJdghAEslJeEARByGhEoAiCILQ3AtJ8O5jOUCl1OXA1cAiwErhDa/2si3TdgNuBs4G+wEbg78CtWusDEfHOAP4dI5tRWuslyd2BIGQOyXz3k8f044PPN5srjCAkwKiGopT6OfBnYBFwDbABeEYpdW6CdDnAP4GfAK8AVwGzgF8Az0VFHwmEgAuB86L+1pm5EyGbKSpsO6vLk+lH/ujMwxPGue67o5K4gpCtBDWHYkxDUUp1B8qAGVrr79vnHgbeA+5WSr2gtW6IkfxM4GTgSq31A/a5h5RS64H/VUodr7X+yD5/BLBWa/2EqbJnMocfXMzytbvTXQxBENoQ2WA2/HWgC/Bg+ITWuhFLYxkMHBcn7Yn28fGo88/bx8i0I4FlSZQzq5g2YVC6i5AxlAwoSncRUk+WWXkdP+qgdBch7fQqMj6TkDWYFChH2ceFUecXRoU7cTswVmtdE3W+t32sh6ahseHAF/bvAqVUVj69kUN7uIwZCrQcbYme3ToBwanz6SDolfKms3f7tk44vI/hK2cO7fmLNSlQBgK7tNZ7o85vso8Hx0qotd6ptf7MIehy+zjbPpYAhcChSqnFwF6gRin1rFKqt0P6rGfogG4pvV7/XoUpvZ5Jwo1je/6gvWK6rrp2djl/1ZakfhYS1PqmhL17pVS/BFGqtdbVQBFWAx/NPvvYxUvBlFLnYVl8vau1nmufHmkfjwHuAiqASVgGACOVUhO11vui84pFRUWFlyIZZe9ep6pqzZ4d65O+1uSRRXywdI9jWNn0QZTNaL7Gj0/tSdkMd2VLNbW1tU3/dynIpaa2sUV4fX09AA0Nsabqso8NGzb4Tlu+YkXCOBs3bvSdP8CUI4p4f0nzu1VT7W6+r7q6OmGcvFxoaEwYLePIBllZU1NDeXm58XzdDBdtShB+B3ATVj3G6/C47gwppc4EHgU2Y1lzhVkN3Ab8XWsdro1XlFIrgYeAS4A/ub3O0KFDyc/3YxGUfCNfWFgI7E8Yr7S0NOnr9ezZE2j+6EsGFLF64x7H/E1cLyiOGzOIF961OgF5eXlAy9YmP78D0OAYlq0MGjQI2BE3znjVmwV6e6vzqrQUy9AyNgMHDEiYfzx69exF5LvVs0ePFr9jcdyYg1m2TseNk5uTS0MWPsfCglx2JK6CtNK1Sxf7W/dGXV1d3I64myGvSxP8vWTHqwY6O6QPn6tyU2Cl1PeAF7DeytO11mvDYVrrJVrrWyKESZjHsOZZTnJzjfbGgN4th7FuuXh8mkrSGrdDbH+5YTLfnDyk+YSL7knPbp04bGBqhwzDdO6YZySfRL3dXt06cfU5Rxi5lldGDOnueP48F3u4TB0/IPEFsqGr70CXTrGb1Z5FnRzPXz99dCBluemCsTxy4xQu/9bwQPKPJqGGorV+xGVea4GeSqlOWuvIrnf4zUmouyulfoxlFbYTmKa1/tzNhbXWdUqpXUBXl2VNO6k03lEHd2/xO79DLn+5YXJGuEa/40dHcfFv3k8YL3ptSWMCgfLgzyfRMT+X3JwcLrpzVjJF5MYfjOGup129ik0Udcln34Hgh94KCzq43jPlsm8O56GXzRhI3n/1sRR37cjLH3zZ4nwoBB3zm8tz/fTRvP7xOpZW7GoRz80Yfga8nk0cfFBX1m5JPEyXiNwYH12suaecHKtO/dKtSz6FBR2YPKY/D74U8eyzwGx4IVYxj4w6P9Y+fhIvsVLqfKxhq83AFCdhopS6XSm1WilVFHW+J9AHa04lcIq7mOl9ppOiwny6uJ1ADZCCTv6M9HoUdXQ42/yVFHftSOdOHehkQFPo28NJ8Y5PjqEvtrir0336o7PPunaiT4/OdMx3rtvIBrB3cYHvzlMmWUy7t8qMTyiGdAjKkCTWMzf1fkZjUqD8G6gFfho+oZTKBa4AvgTmxEqolBoB/BXYBpyotY7VjVoLDAV+GHX+Zvs4w1fJPXLJKX2TziOV30oyPZxMpTRK64LmBsj0/aazYRvSv4hTjzazFsnxPpK9uQSVHUocJWO56/KJTf9/75RDGXVoT9dpb7pgrON5r1WRTMN/REkPehUXNP3+03XHN+cb0DttrMuitd6hlLoLKLMFyUwsK61JwLmRq+SVUt+007xsn7oF6AS8ARytlDo6KvtFWutFWAsfLwV+p5QaBiwGpgFnAX/RWiceOzFA547iUzPdmPweCgs6sLe2Pma4rwbRYAHV4GLenJu8oYRjkQJo7SMbq44dcumQ568ykilap/w89tf5H3I85KCuDOrTbJjaIS+Xow7vzeJVO12lHxFLo4lxT041dFDPzmzd5dpotRXR84fhdVpBYrplvA24HjgW+CPW/Mk5Wut/RMW7z/4Lc4J9PA94yuHvLLDmSoDTsCzAzrLzGAFch6UJZSylg4uTzuPKs0cmjtSOCTdAXntfF51hWbuM9tADTUSPJIeqLv364fzP98dYPxxu6HvTDvWeqcuKufLsEUYWHnYrzKdfr0K6dWlZF/97/pGu0vuVJz/99kjOPy2xYUAYt4YbJ48fyBVnjfBZKotGD1LylovGJYzzzcmHNP3fp3tBi7B07KVjdJW51joE/J/9Fy/ekKjfida6RMbdCVxm/7UL7r/mWFZv2MPY0l5N53514Viqauq4/x9t37nyQT29z2F4ITzPEmvC1A/XnDuKy+7+0DHs/NOG8eQb8deIdOvakTHDrOftVKoxw3rxzNurki2mo5A5blQ/5i9vbYbslaOGW0Ipug09osSc4Hbi2CMO4sNF7rwsjxjSnQP1zqbJ0VWTm5vDpNH9+PM/v/BdNi9aV/cYFmGRnHPyoU2GEUMHFLGtsnmtVl6c9zkbfHm1H0w8DA9PtE/3zhw9suW8zfAhPVqdSzcFhkxlo3GqqfwOrV9dvx9J+MPr6JCn33yje+WRnHbM4ITpI9uC8PWT1RqCaEOi28cQzfWZ79L6LBbpnJM3MRAYOWTWnG+AE0pRWZvsILlFBIoP8vNyOPfkEl8feOR48q8ubJ64O9mFXX4sq5qgUQe7G64LNyTXpGBdxLdPGppU+q8f3+wJaMxhvfjmlCFc9FWVbLGMMeawXokjeSQVIyC5OTBpTD/OnHQI35laYp/114h2TWIbAre36mZYKF5P3zOpkyfk+Zy7SgYRKD75xuQhXPvdUXT3OFb+teOtMc8cLC3jwZ9P4isTB3HBGaX85KwRrRYhmsHdW3xxjAbVrdnszReN41tThjBheJ+kx5oTEc8ENnpY4c4fT2gVZ3xEZyA3N4dzppbE3Ecl3jBFEHuvnDpxkGPvMkSIS76mmudWDBB5lSvPHslV344/T3f99NFcGrHPSnTdfGPyEDrk5fLdaYdSWOB+RD2eNvOHa+I5Ko9Pty7xn88lX3N+53OwhM3ZJw7l9kvj+bWNjZOs6terkMljWo/w5+RYc1dJEfUsvhLHU3k2mA0Lboj6Aou7duTCM0rpkJfL8aP7cfSI9A1jTZswMKn0gw/qynemlpCTY401p5pYnc0h/YNzex9rmCwZouVXuBcdCsHJRw1smlsJnzPFcaMO4pgj4rufH1vam5PGxdamvQiRlJCgfhLNU5x94tCk35+7Lms2P87Jgcu/NYLvnlzSIk4oZM1dmeLKs0fEX4MlcyjZzYVnlDJiSPeE3oODGJboXVwQ14Lpf74/xtPwXaTFWuTQkUkmjgjYvXma10acfsxgBvf15C+1BW5fkzMnHdLi98ThfTytp/DCN45JbvHf9dNHt3juGbSu0RjhDsAJYwdQOri4lSlvqu75nKkliSP5QARKihjSv4ibLhznOJkcSRCLwPLycrnxvCNjho8Z1otrPWwFe/JRzT3UiT40KjfzH9ec01yeVJk/xrqM38vHs/s/77RhnBinp+90/ch3w+1r8t1ph7aYj7nm3FEcOcz8/My5J5cwtsS/gARL+4l87snQVGeJIia5ct3vCvrirh0pu2Q8PWwNKVzev5dN9ZUfeJvwD2ZoXQRK2sgktxJeyfSVz67qNkGcRBOxU47s5/5aHol2z9G8z0vrio93/VhhR0VtHXT6MYNjOnpMRCxXIpGcenRiqzbTjBjSgxzgNENeBrxw7OHp2Vk0E75LESgpxs0HmGmcOLZ/zLB+PtaIeG2DvzH5kMSRDBCpCT1180n07WEtFAuFWi8aO+NYa6jvWyf4tzY70qUl16H2wjunSVY/r1OfHp1baCnnnTaMmy5suYjO5Gt66MBudDE0t3LHj5oNLDrlt26+whPe3Ys68feyqRw2KPkFxV45pG9LzTSyIxAt5L92nPUe9fO4sV23wvy0CMtEiEAxxI0/sCxv3I+LOzermaa5zCibyvAhLdX6yMYmnoPJGTHUd69DWFOOjC3QWuRL8+R1Qjw2mPdHWRp17JDLjLKpcSeoE9G/d2HMOoqke1EnZpRNbTEZ760GY99spr1vTsRqkB/9xQkOsb0TVBW4ec+PHtmXGWVT3e90afPQDZM5//SW+5lkQl9VBEqSnHyUZRmVcdYtAZKsluUlvSnHiMmQDY2uG4Kci0o6Z4cMEhXXKfyoKOOSoOYK3HCC3RHqEWFJFmSjHx6GjcfpLhbVJkP7aQUD4qwThvCtKUNYv60GcH5hnr659b5f2dxIpaoj9PTNJ2V1PWUakYI8E3qzLXAoT2QZnfYjiRaQT998Uqv1O72KC3j65pPYW1vPj373gZvLGuO0YwZx6tHWmqJfXTiW2x//NJDrdMjLob4hxJjDenH0iL7M/WJrzLg/OPUwvv+VwwLrXIiGkiQ5OTnk5sZfJhT5kocXCcayDilJYFacjCZkak+HVEmU3Nwcby9+CoWPm2JF95bdknGNfRzCcxSJzOGTxc2jjeVqJDc3Jy1uSMJtQ/j/4K8HuQla9MgyBYEIlBQzsE8XHrjueM441ln1HKd6t9i3IJo/XXscD//PZMewR26cEvfaN3x/DA9dPylhGcPvvlN+pYOL6dfL+0T80IjFYaY/rq9MTDws9rsrJiaM4wWnlfrR9XXeqYfx4M8T13eqiKz3RI/ArQnqhOF9+NN1xye/tiXBkFdQMjbams/rm3moxy2m27rGLUNeaaBHgn0J4q1fiLfDYSLtJb9DLvkdEruK+cv1k6lvaHTM78bzjvTlBDLSWaIpS7dRJT1YvHpXU365cb5Wk37Q7vnpMY7OH6PrKy8v1+iOi16JJ7hNakGp2Gcj/IxNNsjDD+ne4r248uyR9CruxK2PLXSdxykTBvKvD780V6gEdO2c36pzZI2PZIZaKxqK0IquhfmOLil6FxcE5lHYD2HngeH95b00Nn68voZ7s15NPL1iumnwIzwG9E5ukaJXSga4W7uRQw6H9OvqOX+ndUXRPup6FXeip/3ex7PWjNxOISy0nbekNkDUs+tS0MGVW/t0IRqK4Io7fjSBXsXJvMhmmsk/Xnsc+/a33ImvqffqIz8vzj3/cO1x7Kmp83yNcJlv+PNcz2ndcu9Vx9IYCvGzP8bcadtT/Zx9whBGDOnOnU9+lnTZIL6wv/WS8QyMcPX+28sn0rVzPrc8Ot86EfXq/PKCsWzZae1keNdlE7nxoXkJr9+pYx63XjKeWx5dEDdenx6dufmicS2GaKO57YdHsWLdbnpHrU1yQzbNj/lBBIrgiqEue5BuSGYOJXKP7LApRKOtoniZbAynjV6wCM1CpndxAVt31dLB9oTbo6hTCxNQP2UOCjebkIXi/IomLy838I2wwgyL2s108EFhDaT5efYqLmDd1hry83Pp2jmfrgMt7fRgD9pK9HVicfgh3eOGFxXmM872NrCzar/r66eC7l2t9zOeN+4gEYHSRrjl4sTbhaYXd4393T85msrqA55ybnQxhxJNrCGvn31vNEPsRurqc0axdPXOwAVCty75VEVoPoWdMmdY0RROPfNfXTg27kZkkULvirNGsGjlTvoHPNzolfBQWrcuHdm1J/F767kvFTU9kmio9tyTSzj4oK6B+GtzgwgUw6RLo1UHd0/Tld3irmYG9unSYvgjLlFOE/0oPtHa0vgIP1dFhfkJ3blHc/NF45o0Jrfc/sOjuPr+jwFrncC0o5LbRiAWydkzmeEnEfvkRHtgiEfXzvkcN8rbs0iEiQn+4q4d+dGZhzPmsF785J6Pks8wGo8NSsf8PE6I4yopaESgGKKtmwNmA/E8Oafq8SQaLnGiT8TkcNhHWPCktusT/j5Ge9qJMlVeppO7jluv0SYIamMsU4hAyVDu+NEEln25K93FCIxfX3oUet1uI3mNHNqD4i4dmXJkP274s/MEbe/uBZx1whD+OWuNkWtmCsm6LXPbmP74G8OTNMrwT1Ciz/QE+XmnDUMNLqa+ZrPZjCMIdE96A4hAyVCGDihyNRF+7bmjqKxuPTF46tGDXG/d64dfnHck85dU+E5fMrAbJR4XhcUiNyeH6V85jKqa8Bi2k5v3HL59UkmTQBk2qJhTJgzkq8elSiOIz8+/N5ptlbWu4/vqVCfRuU3nMEoifnXhWPRaM52TZAj7ySovby1QSgdb79vXMuR9CwoRKFnOhOHO7j0uiPJEappRh/akU8P2QK+RiHj7hCQiNzeHi77qvJ94OhgXtUdJIpLtXbcl89XhQ3p4mo8Jk8ph6kx734JCFjYaol/PQg4b2I1Lvtb2X5pkGVfqrfGMRewGIXZLcfFXFccbntx1y7jS3lx59ojEET3gpk383rRDOaRfV5SD2Wyi9FecNaKFoYJfrjx7JIcfXOxpX5QgGvzzThtmPlPDnHHsYL59ov99dtKJaCiG6NAhl9suPSrdxchoehRZppXhFe6m8NLbnjZhINMmBGNFlYifTx9tPE83tz6kfxG/uczZl1mi9JNG92PS6MRu0RMx+rBeHifkg+H0YwZTkJ/Hw68uT3dRYvKDUzNf6MVCNJQA+OaUIfROwWK2TOGUCQN9ucNInsy2eAmS9mBVeM7UEsCbNwMhvYiGEgDnTC3hnKklTC+bme6ipIR4Y8OFnTqwd389U8cP5PHXNZCd2yC3VTJZLh0/uh/HG9CO3CKvZfKIQBEC5ZFfNLt0f+L1cqN5x+6lS8vgZmthIfvIdKEnQ15CykhkjeV1UeBYe3J/iO3ILxWbGMXi2CP6pvR6YZ9Nk0b7MzDI8HZJyFKMayhKqcuBq4FDgJXAHVrrZ12kOwP4d4zgUVrrJcleQ8gMYjX8J4/3tuL4mJF9GVd6gtG9Tvxw07kDOVwFa6YdTdfCfB7/5QlxvQO4IpPHvAR3ZNAzNKqhKKV+DvwZWARcA2wAnlFKnesi+UisjtOFwHlRf+sMXUPIYPwoGOkWJmDt6Z2OLWY75uelVSsTgseVJplB6qYxDUUp1R0oA2Zorb9vn3sYeA+4Wyn1gta6IWYGcASwVmv9RIDXENowMtnvnsMGdmPxqp2+3PELQixMaihfB7oAD4ZPaK0bsbSJwcBxCdKPBJYFfA2hXSC99kScfeJQ7rp8IgcflA5z7/SS6f6wIsm2N9mkQAmv6ovekHlhVHgrlFI5wHDgC/t3gVLKSXvyfQ0h/QzuazVeHfKC/kyyp8FIF7m5Oe1SmEQio4XmMTkpPxDYpbXeG3V+k32M5xWtBCgEDlVKLcbSVuqUUi8BV2qtw06jkrmGkGau+s4RVGzak2BTJf/IfILQLsmg1z6hQFFKJVpZVK21rgaKgOiGHmCffYy3a9JI+3gMcBdQAUzCmnQfqZSaqLXel+Q1hDRTWNCBkUO9O/ETBJOEt8ft2tmsC6AgyDZd242GsilB+B3ATbTarLIV8cJWA7cBf9dah1e/vaKUWgk8BFwC/CnJa7SiosK/+/XycveL9LzENUWqrmnyOtu2bqG8vMp3+r37LXuMhobGlNd5Op5xPNJVnkyrByd65If46oTujB7UQHl5Oftqrf7ounXryN2/1cg1gqqHDjkNrfIOhRoBWLFiJfkd0quuuBEolyYI/9Q+VgNOG3CEz8VsKew1Jkscgh7DEiQn2Uff13Bi6NCh5Od776WUl5dTWpp43cE153Snd/cCSgaY2ffDHesBXJUvWdzWQ2KsMn/9pNFJmd/u2VsHbCIvLzcl9x/GXD0kz++uGMiazXsoLU2dy5IwmVQPiTg8wltQ5w/2AAcYPHgwpT523IzGZD3ksL6pl3zRGaWMP7wPPbt1ioqzEWhk2LDDAjejr6uri9sRTyhQtNaPuLzWWqCnUqqT1jpyx6fwarUNLvOJvHadUmoXEJ49NH6NIJk4IrWrp7OddKzlaGsM6tuFQX1l5LctcsrEQekuQkJMWnktxBqSOjLq/Fj7+EmshEqp25VSq5VSRVHnewJ9sOZUkrqGIAhCthGpncQkg/phJq28/g3UAj8F5gIopXKBK4AvgTlx0q4FhgI/BO6NOH+zfZxh4BrtggvPKKWgY/pXj6eDrp07MO2ogZw4LnO3qxUyj2wwDjz5qPTs4eMVYwJFa71DKXUXUGY38jOBs7Gstc6NXMGulPqmneZl+9TjWHM1v1NKDQMWA9OAs4C/aK3f93qN9spXskAtDoqcnBwulh0zBSFtmHYOeRtQA/wE+BZQDpyjtX4hKt599vFlaJorOQ24E0uIXIJl+XUdcL/PawhZwjEj+zIkLRt0CYJgEqMCRWsdAv7P/osXb4jDuZ3AZfZf0tcQsoervnNEuosgCIIBZD8UQRDaNWccaznYGNgnc63j4s3zfG/aoQB0yEt/cy47NgqC0K6ZMLxPVu9wedoxgzntmMHpLgYgGoogCIJgCBEogiAIGcpJWWYCL0NegiAIGcolXzuc804rzRpP2qKhCIIgZCi5uTlZtVBZBIogCIJgBBEogiAIghFEoAiCIAhGEIEiCIIgGEEEiiAIgmAEESiCIAiCEdrrOpQ8sLaz9EsyadsSUg8WUg8WUg8WbbUeIu7L0ZY5JxQKOZ1v0yxYsGAS8EG6yyEIgpClTB4/fvyH0Sfbq4byCTAZ2AS0+025BEEQXJIH9CfGduvtUkMRBEEQzCOT8oIgCIIRRKAIgiAIRhCBIgiCIBhBBIogCIJgBBEogiAIghFEoAiCIAhGEIEiCIIgGKG9Lmz0hVJqKPB74ET71GvAz7TW29JWKAMopf4KlGqtT4w67+p+TcdLJUqpU4GbgPFAIzAHuElrPSciTnuoh6nAbcAYoAr4B1Y9VEfEafP1EEYpNRqYD9yptS6LON9u6sAPsrDRJUqpXsACoCNwP5Ywvh5YA0zUWh9IX+n8o5S6BHgEmBUpUNzer+l4qUQpdQLwLrAUeMwu0xXAAGCy1npeO6mHqcBbdrmeAAYDV9u/p2itG9tDPYRRSnUA5gFjgVvDAqU91YFfRENxz3XAIGCU1noZgFJqLtaHeAHwcBrL5hmlVB7wS6AsRhS392s6Xiq5D1gHHK213muX6UlgGXAHcArtox7uBtYCJ2it99llWgs8AJwK/MdDubO5HsL8AhjpcL491YEvZA7FPd8F3gs/eACt9duAtsOyBqVUAbAQuBV4CtjgEM3t/ZqOlxKUUj2whneeDwsTu0xbgFnAcfaptl4PBcA24OGwMLGZZR9H28c2XQ9hlFKjsIZAb3cIbhd1kAwiUFxgNz4lWOppNAuxxt+ziQKgG3Cu1voCoD4y0O39mo6XYqoABdzrENYbqG8P9aC1rtVan6a1vjMq6Ej7uLY91AM0DXX9DUtDeDoqrF3UQbKIQHHHQPvo1JPfBBQrpYpTWJ5kqQKGaa2fjxHu9n5Nx0sZWusGrfUKrfXGyPP2ZOzxwGzaQT1Eo5Q6RCl1IfAHYAnwEu2nHv4HGAZc5hDWXuogKUSguKPIPu51CAsPE3RJUVmSRmvdqLWujxPF7f2ajpdWlFJdgSftn3fRzupBKdUTa0L4b1ha7E+11rW0g3pQSo0EbgZ+rrVe7xClzdeBCUSguCPHPsYziWtL5nJu79d0vLShlCoE/oU1r3KX1noW7a8eQlhj9+cDXwBvK6XOpo3Xg22g8jjwodY61kR4m64DU4iVlzvCtvidHcLC56pSVJZU4PZ+TcdLC0qp7lj2/8djmQ//0g5qV/Wgtd4FPAeglHoBa8jrXuBrdpS2Wg/XYxkfTFJK9bbP9bCPhfa5dvUu+EU0FHestY/9HcIGAJVa65oUlido3N6v6XgpRynVF2styvHAX4Efaq3DPcN2Uw/R2BZfr2GtSQnPM7XVejgNay3IPCyLt21YE+NgCZttNLeVbbUOjCACxQVa60qgAhjnEDwWa0Vtm8Ht/ZqOl2qUUkXAm1gWTfdqrX8cIUzaRT0opQ5XSq1RSl3hEFyENeyyn7ZdDz/DWnMU+fcDO+wp+/dy2nYdGEEEinteBKYppQ4Pn1BKTcMyPX02baUKDrf3azpeKnkAS5jcr7W+Lkactl4PK4Fi4DKlVMeIMh0CfBvLg8Ie2nA9aK0XaK3fjvwDPrKDV9vnaj2UOevqwBTiesUlSqk+WGPK9Vi+dwqAG7A+yOO11vvTWLykUEqtAdZEuV5xdb+m46UKpdRwrInnSuBaotbiAGitn27r9WCX6QdYPfE5WOsvegFXYg0DTdJaL2kP9RCJUmoIlgYR6XqlXdWBH0SgeEApFV4INwXL3O914HqdpY7cwjgJFPu8q/s1HS8VKKUuAx6MF0drnWPHbbP1EFGmc7DWYRwB1ADvAL/UWpdHxGnz9RDGSaDY59tNHfhBBIogCIJgBJlDEQRBEIwgAkUQBEEwgggUQRAEwQgiUARBEAQjiEARBEEQjCACRRAEQTCCCBRBEATBCCJQBEEQBCOIQBEEQRCMIAJFEARBMML/A7U9sAlgrW+gAAAAAElFTkSuQmCC\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.plot(np.var(voxel_data[subjects[0]], axis=0))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "---"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "--------  subject 1  -------\n",
      "Total number of voxels = 4627\n",
      "--------  subject 2  -------\n",
      "Total number of voxels = 3765\n",
      "--------  subject 3  -------\n",
      "Total number of voxels = 3778\n",
      "--------  subject 4  -------\n",
      "Total number of voxels = 3027\n",
      "--------  subject 5  -------\n",
      "Total number of voxels = 3685\n",
      "--------  subject 6  -------\n",
      "Total number of voxels = 4024\n",
      "--------  subject 7  -------\n",
      "Total number of voxels = 3201\n",
      "--------  subject 8  -------\n",
      "Total number of voxels = 3490\n"
     ]
    }
   ],
   "source": [
    "from src.load_nsd import image_feature_fn, ordering_split\n",
    "trn_stim_ordering, trn_voxel_data, val_stim_ordering, val_voxel_data = {},{},{},{}\n",
    "\n",
    "stim_data = {}\n",
    "data_size, nnv = {}, {}\n",
    "for k,s in enumerate(subjects):\n",
    "    print ('--------  subject %d  -------' % s)\n",
    "    data_size[s], nnv[s] = voxel_data[s].shape      \n",
    "    stim_data[s] = image_feature_fn(image_data[s])\n",
    "    \n",
    "    trn_stim_ordering[s], trn_voxel_data[s], \\\n",
    "    val_stim_ordering[s], val_voxel_data[s] = \\\n",
    "        ordering_split(voxel_data[s], ordering, combine_trial=False)\n",
    "    \n",
    "del image_data\n",
    "del voxel_data"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[[[[0.47611815]]\n",
      "\n",
      "  [[0.45324636]]\n",
      "\n",
      "  [[0.41308543]]]]\n"
     ]
    }
   ],
   "source": [
    "trn_stim_mean = sum([np.mean(stim_data[s], axis=(0,2,3), keepdims=True) for s in subjects])/len(subjects)\n",
    "print (trn_stim_mean)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "---\n",
    "# Joint model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "metadata": {},
   "outputs": [],
   "source": [
    "from src.torch_joint_training_unpacked_sequences import *\n",
    "from src.torch_gnet import Encoder\n",
    "from src.torch_mpf import Torch_LayerwiseFWRF"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Model instanciation"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "metadata": {},
   "outputs": [],
   "source": [
    "_log_act_fn = lambda _x: T.log(1 + T.abs(_x))*T.tanh(_x)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "metadata": {},
   "outputs": [],
   "source": [
    "N = 64\n",
    "\n",
    "shared_model = Encoder(trn_stim_mean, trunk_width=N).to(device)\n",
    "rec, fmaps, h = shared_model(T.from_numpy(stim_data[subjects[0]][:20]).to(device))\n",
    "\n",
    "subject_fwrfs = {s: Torch_LayerwiseFWRF(fmaps, nv=nnv[s], pre_nl=_log_act_fn, \\\n",
    "                 post_nl=_log_act_fn, dtype=np.float32).to(device) for s in subjects}"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Load prefilters value"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<All keys matched successfully>"
      ]
     },
     "execution_count": 27,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "try:\n",
    "    from torch.hub import load_state_dict_from_url\n",
    "except ImportError:\n",
    "    from torch.utils.model_zoo import load_url as load_state_dict_from_url\n",
    "    \n",
    "state_dict = load_state_dict_from_url('https://download.pytorch.org/models/alexnet-owt-4df8aa71.pth', progress=True)\n",
    "        ### Rename dictionary keys to match new breakdown\n",
    "pre_state_dict = {}\n",
    "pre_state_dict['conv1.0.weight'] = state_dict.pop('features.0.weight')\n",
    "pre_state_dict['conv1.0.bias'] = state_dict.pop('features.0.bias')\n",
    "pre_state_dict['conv2.0.weight'] = state_dict.pop('features.3.weight')\n",
    "pre_state_dict['conv2.0.bias'] = state_dict.pop('features.3.bias')\n",
    "  \n",
    "    \n",
    "shared_model.pre.load_state_dict(pre_state_dict)\n",
    "#for s,m in subject_models.items():\n",
    "#    m.pre.load_state_dict(pre_state_dict)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Loss function, grad function, etc."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 29,
   "metadata": {},
   "outputs": [],
   "source": [
    "optimizer_net = optim.Adam([\n",
    "        #{'params': shared_model.pre.parameters()},\n",
    "        {'params': shared_model.enc.parameters()},\n",
    "    ], lr=1e-3, betas=(0.9, 0.999), eps=1e-08)\n",
    "\n",
    "subject_optimizer = {s: optim.Adam([\n",
    "        {'params': sp.parameters()}\n",
    "    ], lr=1e-3, betas=(0.9, 0.999), eps=1e-08) for s,sp in subject_fwrfs.items()}\n",
    "\n",
    "#subject_opts = {s: [optimizer_net] for s in subject_optimizer.keys()}\n",
    "#subject_opts = {s: [subject_optimizer[s]] for s in subject_optimizer.keys()}\n",
    "subject_opts = {s: [optimizer_net, subject_optimizer[s]] for s in subject_optimizer.keys()}"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 30,
   "metadata": {},
   "outputs": [],
   "source": [
    "def _model_fn(_ext, _con, _x):\n",
    "    '''model consists of an extractor (_ext) and a connection model (_con)'''\n",
    "    _y, _fm, _h = _ext(_x)\n",
    "    return _con(_fm)\n",
    "\n",
    "def _smoothness_loss_fn(_rf, n):\n",
    "    delta_x = T.sum(T.pow(T.abs(_rf[:, 1:]    - _rf[:, :-1]), n))\n",
    "    delta_y = T.sum(T.pow(T.abs(_rf[:, :, 1:] - _rf[:, :, :-1]), n))\n",
    "    return delta_x + delta_y\n",
    "\n",
    "def vox_loss_fn(r, v, nu=0.5, delta=1.):\n",
    "    #err = T.sum(huber(r, v, delta), dim=0)\n",
    "    err = T.sum((r - v)**2, dim=0)\n",
    "    # squared correlation coefficient with 'leak'\n",
    "    cr = r - T.mean(r, dim=0, keepdim=True)\n",
    "    cv = v - T.mean(v, dim=0, keepdim=True)\n",
    "    wgt = T.clamp(T.pow(T.mean(cr*cv, dim=0), 2) / ((T.mean(cr**2, dim=0)) * (T.mean(cv**2, dim=0)) + 1e-6), min=nu, max=1).detach()\n",
    " \n",
    "    weighted_err = wgt * err # error per voxel\n",
    "    loss = T.sum(weighted_err) / T.mean(wgt)\n",
    "    return err, loss\n",
    "    \n",
    "def _loss_fn(_ext, _con, _x, _v):\n",
    "    _r = _model_fn(_ext, _con, _x)\n",
    "    #_err = T.sum((_r - _v)**2, dim=0)\n",
    "    #_loss = T.sum(_err)\n",
    "    _err, _loss = vox_loss_fn(_r, _v, nu=0.1, delta=.5)\n",
    "\n",
    "    _loss += fpX(1e-1) * T.sum(T.abs(_con.w))\n",
    "    return _err, _loss\n",
    "    \n",
    "    \n",
    "def _training_fn(_ext, _con, _opts, xb, yb):\n",
    "    for _opt in _opts:\n",
    "        _opt.zero_grad()\n",
    "        _err, _loss = _loss_fn(_ext, _con, T.from_numpy(xb).to(device), T.from_numpy(yb).to(device))\n",
    "        _loss.backward()\n",
    "        _opt.step()\n",
    "    return _err\n",
    "\n",
    "def _holdout_fn(_ext, _con, xb, yb):\n",
    "    # print (xb.shape, yb.shape)\n",
    "    _err,_ = _loss_fn(_ext, _con, T.from_numpy(xb).to(device), T.from_numpy(yb).to(device))\n",
    "    return _err\n",
    "\n",
    "def _pred_fn(_ext, _con, xb):\n",
    "    return _model_fn(_ext, _con, T.from_numpy(xb).to(device))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Training\n",
    "IDEA: Use the more stable estimate from the holdout val cc to weight the voxels\n",
    "Use a contiguity cost instead of sparsity for RF weights"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 31,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "subject 1 masked 4627 of 4627\n",
      "subject 1 training/holdout 8100 900\n",
      "subject 2 masked 3765 of 3765\n",
      "subject 2 training/holdout 8100 900\n",
      "subject 3 masked 3778 of 3778\n",
      "subject 3 training/holdout 8100 900\n",
      "subject 4 masked 3027 of 3027\n",
      "subject 4 training/holdout 8100 900\n",
      "subject 5 masked 3685 of 3685\n",
      "subject 5 training/holdout 8100 900\n",
      "subject 6 masked 4024 of 4024\n",
      "subject 6 training/holdout 8100 900\n",
      "subject 7 masked 3201 of 3201\n",
      "subject 7 training/holdout 8100 900\n",
      "subject 8 masked 3490 of 3490\n",
      "subject 8 training/holdout 8100 900\n",
      "\\ : 100.0 %\n",
      "  Epoch 1 of 50 took       73.992s\n",
      "  training loss:               0.655085\n",
      "  holdout loss (batch):        0.519092\n",
      "  Subject 1: median (max) validation accuracy = 0.564 (0.925)\n",
      "  Subject 2: median (max) validation accuracy = 0.599 (0.917)\n",
      "  Subject 3: median (max) validation accuracy = 0.706 (0.950)\n",
      "  Subject 4: median (max) validation accuracy = 0.730 (0.937)\n",
      "  Subject 5: median (max) validation accuracy = 0.810 (0.957)\n",
      "  Subject 6: median (max) validation accuracy = 0.720 (0.942)\n",
      "  Subject 7: median (max) validation accuracy = 0.671 (0.916)\n",
      "  Subject 8: median (max) validation accuracy = 0.770 (0.947)\n",
      "** Saving params with joint score = 0.699 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 2 of 50 took       74.634s\n",
      "  training loss:               0.503597\n",
      "  holdout loss (batch):        0.417178\n",
      "  Subject 1: median (max) validation accuracy = 0.660 (0.945)\n",
      "  Subject 2: median (max) validation accuracy = 0.677 (0.950)\n",
      "  Subject 3: median (max) validation accuracy = 0.778 (0.955)\n",
      "  Subject 4: median (max) validation accuracy = 0.795 (0.950)\n",
      "  Subject 5: median (max) validation accuracy = 0.869 (0.967)\n",
      "  Subject 6: median (max) validation accuracy = 0.789 (0.951)\n",
      "  Subject 7: median (max) validation accuracy = 0.755 (0.938)\n",
      "  Subject 8: median (max) validation accuracy = 0.828 (0.962)\n",
      "** Saving params with joint score = 0.774 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 3 of 50 took       74.722s\n",
      "  training loss:               0.420529\n",
      "  holdout loss (batch):        0.339437\n",
      "  Subject 1: median (max) validation accuracy = 0.724 (0.957)\n",
      "  Subject 2: median (max) validation accuracy = 0.746 (0.956)\n",
      "  Subject 3: median (max) validation accuracy = 0.831 (0.961)\n",
      "  Subject 4: median (max) validation accuracy = 0.843 (0.955)\n",
      "  Subject 5: median (max) validation accuracy = 0.902 (0.972)\n",
      "  Subject 6: median (max) validation accuracy = 0.838 (0.956)\n",
      "  Subject 7: median (max) validation accuracy = 0.816 (0.947)\n",
      "  Subject 8: median (max) validation accuracy = 0.867 (0.967)\n",
      "** Saving params with joint score = 0.826 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 4 of 50 took       74.726s\n",
      "  training loss:               0.363591\n",
      "  holdout loss (batch):        0.298982\n",
      "  Subject 1: median (max) validation accuracy = 0.766 (0.962)\n",
      "  Subject 2: median (max) validation accuracy = 0.784 (0.961)\n",
      "  Subject 3: median (max) validation accuracy = 0.858 (0.966)\n",
      "  Subject 4: median (max) validation accuracy = 0.871 (0.962)\n",
      "  Subject 5: median (max) validation accuracy = 0.917 (0.976)\n",
      "  Subject 6: median (max) validation accuracy = 0.863 (0.962)\n",
      "  Subject 7: median (max) validation accuracy = 0.850 (0.955)\n",
      "  Subject 8: median (max) validation accuracy = 0.888 (0.969)\n",
      "** Saving params with joint score = 0.854 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 5 of 50 took       74.783s\n",
      "  training loss:               0.325599\n",
      "  holdout loss (batch):        0.266092\n",
      "  Subject 1: median (max) validation accuracy = 0.793 (0.967)\n",
      "  Subject 2: median (max) validation accuracy = 0.816 (0.967)\n",
      "  Subject 3: median (max) validation accuracy = 0.874 (0.969)\n",
      "  Subject 4: median (max) validation accuracy = 0.889 (0.966)\n",
      "  Subject 5: median (max) validation accuracy = 0.925 (0.978)\n",
      "  Subject 6: median (max) validation accuracy = 0.881 (0.968)\n",
      "  Subject 7: median (max) validation accuracy = 0.872 (0.961)\n",
      "  Subject 8: median (max) validation accuracy = 0.902 (0.971)\n",
      "** Saving params with joint score = 0.874 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 6 of 50 took       74.771s\n",
      "  training loss:               0.292025\n",
      "  holdout loss (batch):        0.231356\n",
      "  Subject 1: median (max) validation accuracy = 0.822 (0.968)\n",
      "  Subject 2: median (max) validation accuracy = 0.844 (0.973)\n",
      "  Subject 3: median (max) validation accuracy = 0.892 (0.972)\n",
      "  Subject 4: median (max) validation accuracy = 0.904 (0.973)\n",
      "  Subject 5: median (max) validation accuracy = 0.934 (0.980)\n",
      "  Subject 6: median (max) validation accuracy = 0.896 (0.973)\n",
      "  Subject 7: median (max) validation accuracy = 0.892 (0.967)\n",
      "  Subject 8: median (max) validation accuracy = 0.915 (0.973)\n",
      "** Saving params with joint score = 0.892 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 7 of 50 took       74.749s\n",
      "  training loss:               0.263912\n",
      "  holdout loss (batch):        0.212735\n",
      "  Subject 1: median (max) validation accuracy = 0.846 (0.971)\n",
      "  Subject 2: median (max) validation accuracy = 0.866 (0.977)\n",
      "  Subject 3: median (max) validation accuracy = 0.905 (0.974)\n",
      "  Subject 4: median (max) validation accuracy = 0.917 (0.977)\n",
      "  Subject 5: median (max) validation accuracy = 0.942 (0.981)\n",
      "  Subject 6: median (max) validation accuracy = 0.910 (0.977)\n",
      "  Subject 7: median (max) validation accuracy = 0.907 (0.971)\n",
      "  Subject 8: median (max) validation accuracy = 0.926 (0.977)\n",
      "** Saving params with joint score = 0.906 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 8 of 50 took       74.804s\n",
      "  training loss:               0.239547\n",
      "  holdout loss (batch):        0.183886\n",
      "  Subject 1: median (max) validation accuracy = 0.868 (0.976)\n",
      "  Subject 2: median (max) validation accuracy = 0.884 (0.981)\n",
      "  Subject 3: median (max) validation accuracy = 0.919 (0.977)\n",
      "  Subject 4: median (max) validation accuracy = 0.927 (0.981)\n",
      "  Subject 5: median (max) validation accuracy = 0.947 (0.983)\n",
      "  Subject 6: median (max) validation accuracy = 0.922 (0.981)\n",
      "  Subject 7: median (max) validation accuracy = 0.920 (0.974)\n",
      "  Subject 8: median (max) validation accuracy = 0.935 (0.979)\n",
      "** Saving params with joint score = 0.918 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 9 of 50 took       74.708s\n",
      "  training loss:               0.218136\n",
      "  holdout loss (batch):        0.166361\n",
      "  Subject 1: median (max) validation accuracy = 0.884 (0.979)\n",
      "  Subject 2: median (max) validation accuracy = 0.899 (0.983)\n",
      "  Subject 3: median (max) validation accuracy = 0.928 (0.980)\n",
      "  Subject 4: median (max) validation accuracy = 0.937 (0.983)\n",
      "  Subject 5: median (max) validation accuracy = 0.952 (0.985)\n",
      "  Subject 6: median (max) validation accuracy = 0.930 (0.983)\n",
      "  Subject 7: median (max) validation accuracy = 0.930 (0.977)\n",
      "  Subject 8: median (max) validation accuracy = 0.943 (0.982)\n",
      "** Saving params with joint score = 0.928 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 10 of 50 took       74.719s\n",
      "  training loss:               0.200226\n",
      "  holdout loss (batch):        0.147047\n",
      "  Subject 1: median (max) validation accuracy = 0.900 (0.980)\n",
      "  Subject 2: median (max) validation accuracy = 0.912 (0.985)\n",
      "  Subject 3: median (max) validation accuracy = 0.938 (0.983)\n",
      "  Subject 4: median (max) validation accuracy = 0.944 (0.985)\n",
      "  Subject 5: median (max) validation accuracy = 0.956 (0.987)\n",
      "  Subject 6: median (max) validation accuracy = 0.938 (0.985)\n",
      "  Subject 7: median (max) validation accuracy = 0.939 (0.980)\n",
      "  Subject 8: median (max) validation accuracy = 0.949 (0.984)\n",
      "** Saving params with joint score = 0.936 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 11 of 50 took       74.946s\n",
      "  training loss:               0.182325\n",
      "  holdout loss (batch):        0.135075\n",
      "  Subject 1: median (max) validation accuracy = 0.913 (0.981)\n",
      "  Subject 2: median (max) validation accuracy = 0.921 (0.986)\n",
      "  Subject 3: median (max) validation accuracy = 0.945 (0.984)\n",
      "  Subject 4: median (max) validation accuracy = 0.951 (0.986)\n",
      "  Subject 5: median (max) validation accuracy = 0.960 (0.987)\n",
      "  Subject 6: median (max) validation accuracy = 0.944 (0.986)\n",
      "  Subject 7: median (max) validation accuracy = 0.946 (0.981)\n",
      "  Subject 8: median (max) validation accuracy = 0.955 (0.985)\n",
      "** Saving params with joint score = 0.943 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 12 of 50 took       74.844s\n",
      "  training loss:               0.165916\n",
      "  holdout loss (batch):        0.115509\n",
      "  Subject 1: median (max) validation accuracy = 0.928 (0.983)\n",
      "  Subject 2: median (max) validation accuracy = 0.932 (0.988)\n",
      "  Subject 3: median (max) validation accuracy = 0.954 (0.987)\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "  Subject 4: median (max) validation accuracy = 0.957 (0.988)\n",
      "  Subject 5: median (max) validation accuracy = 0.964 (0.989)\n",
      "  Subject 6: median (max) validation accuracy = 0.952 (0.988)\n",
      "  Subject 7: median (max) validation accuracy = 0.954 (0.984)\n",
      "  Subject 8: median (max) validation accuracy = 0.961 (0.986)\n",
      "** Saving params with joint score = 0.951 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 13 of 50 took       74.778s\n",
      "  training loss:               0.151613\n",
      "  holdout loss (batch):        0.106966\n",
      "  Subject 1: median (max) validation accuracy = 0.937 (0.984)\n",
      "  Subject 2: median (max) validation accuracy = 0.940 (0.988)\n",
      "  Subject 3: median (max) validation accuracy = 0.960 (0.987)\n",
      "  Subject 4: median (max) validation accuracy = 0.962 (0.989)\n",
      "  Subject 5: median (max) validation accuracy = 0.966 (0.990)\n",
      "  Subject 6: median (max) validation accuracy = 0.956 (0.989)\n",
      "  Subject 7: median (max) validation accuracy = 0.959 (0.986)\n",
      "  Subject 8: median (max) validation accuracy = 0.964 (0.987)\n",
      "** Saving params with joint score = 0.956 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 14 of 50 took       74.904s\n",
      "  training loss:               0.139064\n",
      "  holdout loss (batch):        0.094158\n",
      "  Subject 1: median (max) validation accuracy = 0.946 (0.985)\n",
      "  Subject 2: median (max) validation accuracy = 0.948 (0.989)\n",
      "  Subject 3: median (max) validation accuracy = 0.965 (0.988)\n",
      "  Subject 4: median (max) validation accuracy = 0.967 (0.990)\n",
      "  Subject 5: median (max) validation accuracy = 0.969 (0.990)\n",
      "  Subject 6: median (max) validation accuracy = 0.961 (0.989)\n",
      "  Subject 7: median (max) validation accuracy = 0.965 (0.987)\n",
      "  Subject 8: median (max) validation accuracy = 0.969 (0.988)\n",
      "** Saving params with joint score = 0.961 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 15 of 50 took       74.867s\n",
      "  training loss:               0.128247\n",
      "  holdout loss (batch):        0.091541\n",
      "  Subject 1: median (max) validation accuracy = 0.951 (0.986)\n",
      "  Subject 2: median (max) validation accuracy = 0.953 (0.989)\n",
      "  Subject 3: median (max) validation accuracy = 0.969 (0.989)\n",
      "  Subject 4: median (max) validation accuracy = 0.969 (0.989)\n",
      "  Subject 5: median (max) validation accuracy = 0.970 (0.990)\n",
      "  Subject 6: median (max) validation accuracy = 0.964 (0.990)\n",
      "  Subject 7: median (max) validation accuracy = 0.968 (0.987)\n",
      "  Subject 8: median (max) validation accuracy = 0.971 (0.988)\n",
      "** Saving params with joint score = 0.964 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 16 of 50 took       74.922s\n",
      "  training loss:               0.119816\n",
      "  holdout loss (batch):        0.078186\n",
      "  Subject 1: median (max) validation accuracy = 0.958 (0.986)\n",
      "  Subject 2: median (max) validation accuracy = 0.959 (0.990)\n",
      "  Subject 3: median (max) validation accuracy = 0.972 (0.989)\n",
      "  Subject 4: median (max) validation accuracy = 0.973 (0.991)\n",
      "  Subject 5: median (max) validation accuracy = 0.972 (0.991)\n",
      "  Subject 6: median (max) validation accuracy = 0.969 (0.991)\n",
      "  Subject 7: median (max) validation accuracy = 0.972 (0.989)\n",
      "  Subject 8: median (max) validation accuracy = 0.974 (0.990)\n",
      "** Saving params with joint score = 0.968 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 17 of 50 took       74.906s\n",
      "  training loss:               0.111906\n",
      "  holdout loss (batch):        0.072907\n",
      "  Subject 1: median (max) validation accuracy = 0.963 (0.988)\n",
      "  Subject 2: median (max) validation accuracy = 0.963 (0.990)\n",
      "  Subject 3: median (max) validation accuracy = 0.975 (0.990)\n",
      "  Subject 4: median (max) validation accuracy = 0.976 (0.991)\n",
      "  Subject 5: median (max) validation accuracy = 0.974 (0.992)\n",
      "  Subject 6: median (max) validation accuracy = 0.972 (0.992)\n",
      "  Subject 7: median (max) validation accuracy = 0.974 (0.990)\n",
      "  Subject 8: median (max) validation accuracy = 0.976 (0.990)\n",
      "** Saving params with joint score = 0.972 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 18 of 50 took       74.908s\n",
      "  training loss:               0.106629\n",
      "  holdout loss (batch):        0.068993\n",
      "  Subject 1: median (max) validation accuracy = 0.966 (0.988)\n",
      "  Subject 2: median (max) validation accuracy = 0.966 (0.990)\n",
      "  Subject 3: median (max) validation accuracy = 0.977 (0.991)\n",
      "  Subject 4: median (max) validation accuracy = 0.977 (0.992)\n",
      "  Subject 5: median (max) validation accuracy = 0.975 (0.992)\n",
      "  Subject 6: median (max) validation accuracy = 0.974 (0.992)\n",
      "  Subject 7: median (max) validation accuracy = 0.976 (0.991)\n",
      "  Subject 8: median (max) validation accuracy = 0.978 (0.991)\n",
      "** Saving params with joint score = 0.973 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 19 of 50 took       74.937s\n",
      "  training loss:               0.102278\n",
      "  holdout loss (batch):        0.065342\n",
      "  Subject 1: median (max) validation accuracy = 0.968 (0.989)\n",
      "  Subject 2: median (max) validation accuracy = 0.968 (0.991)\n",
      "  Subject 3: median (max) validation accuracy = 0.978 (0.991)\n",
      "  Subject 4: median (max) validation accuracy = 0.978 (0.992)\n",
      "  Subject 5: median (max) validation accuracy = 0.976 (0.992)\n",
      "  Subject 6: median (max) validation accuracy = 0.975 (0.993)\n",
      "  Subject 7: median (max) validation accuracy = 0.978 (0.991)\n",
      "  Subject 8: median (max) validation accuracy = 0.979 (0.991)\n",
      "** Saving params with joint score = 0.975 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 20 of 50 took       74.824s\n",
      "  training loss:               0.098503\n",
      "  holdout loss (batch):        0.062889\n",
      "  Subject 1: median (max) validation accuracy = 0.971 (0.990)\n",
      "  Subject 2: median (max) validation accuracy = 0.970 (0.991)\n",
      "  Subject 3: median (max) validation accuracy = 0.980 (0.992)\n",
      "  Subject 4: median (max) validation accuracy = 0.980 (0.992)\n",
      "  Subject 5: median (max) validation accuracy = 0.977 (0.993)\n",
      "  Subject 6: median (max) validation accuracy = 0.977 (0.993)\n",
      "  Subject 7: median (max) validation accuracy = 0.979 (0.991)\n",
      "  Subject 8: median (max) validation accuracy = 0.980 (0.991)\n",
      "** Saving params with joint score = 0.976 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 21 of 50 took       74.824s\n",
      "  training loss:               0.095286\n",
      "  holdout loss (batch):        0.058538\n",
      "  Subject 1: median (max) validation accuracy = 0.973 (0.990)\n",
      "  Subject 2: median (max) validation accuracy = 0.972 (0.991)\n",
      "  Subject 3: median (max) validation accuracy = 0.981 (0.992)\n",
      "  Subject 4: median (max) validation accuracy = 0.981 (0.992)\n",
      "  Subject 5: median (max) validation accuracy = 0.978 (0.993)\n",
      "  Subject 6: median (max) validation accuracy = 0.978 (0.993)\n",
      "  Subject 7: median (max) validation accuracy = 0.980 (0.992)\n",
      "  Subject 8: median (max) validation accuracy = 0.981 (0.991)\n",
      "** Saving params with joint score = 0.978 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 22 of 50 took       74.778s\n",
      "  training loss:               0.092400\n",
      "  holdout loss (batch):        0.058767\n",
      "  Subject 1: median (max) validation accuracy = 0.974 (0.990)\n",
      "  Subject 2: median (max) validation accuracy = 0.973 (0.991)\n",
      "  Subject 3: median (max) validation accuracy = 0.982 (0.993)\n",
      "  Subject 4: median (max) validation accuracy = 0.981 (0.993)\n",
      "  Subject 5: median (max) validation accuracy = 0.979 (0.993)\n",
      "  Subject 6: median (max) validation accuracy = 0.979 (0.993)\n",
      "  Subject 7: median (max) validation accuracy = 0.981 (0.992)\n",
      "  Subject 8: median (max) validation accuracy = 0.981 (0.992)\n",
      "** Saving params with joint score = 0.979 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 23 of 50 took       74.839s\n",
      "  training loss:               0.089526\n",
      "  holdout loss (batch):        0.057370\n",
      "  Subject 1: median (max) validation accuracy = 0.975 (0.991)\n",
      "  Subject 2: median (max) validation accuracy = 0.974 (0.992)\n",
      "  Subject 3: median (max) validation accuracy = 0.982 (0.993)\n",
      "  Subject 4: median (max) validation accuracy = 0.981 (0.993)\n",
      "  Subject 5: median (max) validation accuracy = 0.979 (0.993)\n",
      "  Subject 6: median (max) validation accuracy = 0.979 (0.993)\n",
      "  Subject 7: median (max) validation accuracy = 0.981 (0.992)\n",
      "  Subject 8: median (max) validation accuracy = 0.981 (0.992)\n",
      "** Saving params with joint score = 0.979 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 24 of 50 took       75.141s\n",
      "  training loss:               0.087289\n",
      "  holdout loss (batch):        0.059781\n",
      "  Subject 1: median (max) validation accuracy = 0.976 (0.991)\n",
      "  Subject 2: median (max) validation accuracy = 0.976 (0.992)\n",
      "  Subject 3: median (max) validation accuracy = 0.983 (0.993)\n",
      "  Subject 4: median (max) validation accuracy = 0.982 (0.993)\n",
      "  Subject 5: median (max) validation accuracy = 0.980 (0.993)\n",
      "  Subject 6: median (max) validation accuracy = 0.980 (0.993)\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "  Subject 7: median (max) validation accuracy = 0.981 (0.992)\n",
      "  Subject 8: median (max) validation accuracy = 0.982 (0.992)\n",
      "** Saving params with joint score = 0.980 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 25 of 50 took       74.915s\n",
      "  training loss:               0.085319\n",
      "  holdout loss (batch):        0.054452\n",
      "  Subject 1: median (max) validation accuracy = 0.977 (0.992)\n",
      "  Subject 2: median (max) validation accuracy = 0.977 (0.992)\n",
      "  Subject 3: median (max) validation accuracy = 0.983 (0.993)\n",
      "  Subject 4: median (max) validation accuracy = 0.983 (0.993)\n",
      "  Subject 5: median (max) validation accuracy = 0.981 (0.994)\n",
      "  Subject 6: median (max) validation accuracy = 0.981 (0.994)\n",
      "  Subject 7: median (max) validation accuracy = 0.982 (0.993)\n",
      "  Subject 8: median (max) validation accuracy = 0.983 (0.992)\n",
      "** Saving params with joint score = 0.981 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 26 of 50 took       74.718s\n",
      "  training loss:               0.083635\n",
      "  holdout loss (batch):        0.057095\n",
      "  Subject 1: median (max) validation accuracy = 0.977 (0.991)\n",
      "  Subject 2: median (max) validation accuracy = 0.977 (0.993)\n",
      "  Subject 3: median (max) validation accuracy = 0.983 (0.993)\n",
      "  Subject 4: median (max) validation accuracy = 0.983 (0.993)\n",
      "  Subject 5: median (max) validation accuracy = 0.981 (0.994)\n",
      "  Subject 6: median (max) validation accuracy = 0.981 (0.994)\n",
      "  Subject 7: median (max) validation accuracy = 0.982 (0.993)\n",
      "  Subject 8: median (max) validation accuracy = 0.983 (0.993)\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 27 of 50 took       74.736s\n",
      "  training loss:               0.081718\n",
      "  holdout loss (batch):        0.052616\n",
      "  Subject 1: median (max) validation accuracy = 0.979 (0.992)\n",
      "  Subject 2: median (max) validation accuracy = 0.978 (0.993)\n",
      "  Subject 3: median (max) validation accuracy = 0.984 (0.994)\n",
      "  Subject 4: median (max) validation accuracy = 0.984 (0.994)\n",
      "  Subject 5: median (max) validation accuracy = 0.981 (0.994)\n",
      "  Subject 6: median (max) validation accuracy = 0.981 (0.994)\n",
      "  Subject 7: median (max) validation accuracy = 0.983 (0.993)\n",
      "  Subject 8: median (max) validation accuracy = 0.983 (0.993)\n",
      "** Saving params with joint score = 0.981 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 28 of 50 took       74.700s\n",
      "  training loss:               0.080782\n",
      "  holdout loss (batch):        0.050364\n",
      "  Subject 1: median (max) validation accuracy = 0.979 (0.992)\n",
      "  Subject 2: median (max) validation accuracy = 0.979 (0.993)\n",
      "  Subject 3: median (max) validation accuracy = 0.985 (0.994)\n",
      "  Subject 4: median (max) validation accuracy = 0.984 (0.994)\n",
      "  Subject 5: median (max) validation accuracy = 0.982 (0.994)\n",
      "  Subject 6: median (max) validation accuracy = 0.982 (0.994)\n",
      "  Subject 7: median (max) validation accuracy = 0.983 (0.993)\n",
      "  Subject 8: median (max) validation accuracy = 0.984 (0.993)\n",
      "** Saving params with joint score = 0.982 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 29 of 50 took       74.718s\n",
      "  training loss:               0.079543\n",
      "  holdout loss (batch):        0.049640\n",
      "  Subject 1: median (max) validation accuracy = 0.980 (0.992)\n",
      "  Subject 2: median (max) validation accuracy = 0.979 (0.993)\n",
      "  Subject 3: median (max) validation accuracy = 0.985 (0.994)\n",
      "  Subject 4: median (max) validation accuracy = 0.984 (0.994)\n",
      "  Subject 5: median (max) validation accuracy = 0.982 (0.994)\n",
      "  Subject 6: median (max) validation accuracy = 0.982 (0.994)\n",
      "  Subject 7: median (max) validation accuracy = 0.984 (0.994)\n",
      "  Subject 8: median (max) validation accuracy = 0.984 (0.993)\n",
      "** Saving params with joint score = 0.982 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 30 of 50 took       74.693s\n",
      "  training loss:               0.078381\n",
      "  holdout loss (batch):        0.046871\n",
      "  Subject 1: median (max) validation accuracy = 0.981 (0.993)\n",
      "  Subject 2: median (max) validation accuracy = 0.980 (0.993)\n",
      "  Subject 3: median (max) validation accuracy = 0.985 (0.994)\n",
      "  Subject 4: median (max) validation accuracy = 0.985 (0.994)\n",
      "  Subject 5: median (max) validation accuracy = 0.983 (0.994)\n",
      "  Subject 6: median (max) validation accuracy = 0.983 (0.994)\n",
      "  Subject 7: median (max) validation accuracy = 0.984 (0.993)\n",
      "  Subject 8: median (max) validation accuracy = 0.985 (0.993)\n",
      "** Saving params with joint score = 0.983 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 31 of 50 took       74.693s\n",
      "  training loss:               0.077048\n",
      "  holdout loss (batch):        0.047239\n",
      "  Subject 1: median (max) validation accuracy = 0.981 (0.993)\n",
      "  Subject 2: median (max) validation accuracy = 0.980 (0.993)\n",
      "  Subject 3: median (max) validation accuracy = 0.986 (0.994)\n",
      "  Subject 4: median (max) validation accuracy = 0.985 (0.994)\n",
      "  Subject 5: median (max) validation accuracy = 0.983 (0.994)\n",
      "  Subject 6: median (max) validation accuracy = 0.983 (0.994)\n",
      "  Subject 7: median (max) validation accuracy = 0.984 (0.994)\n",
      "  Subject 8: median (max) validation accuracy = 0.985 (0.993)\n",
      "** Saving params with joint score = 0.983 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 32 of 50 took       74.720s\n",
      "  training loss:               0.075935\n",
      "  holdout loss (batch):        0.049222\n",
      "  Subject 1: median (max) validation accuracy = 0.981 (0.992)\n",
      "  Subject 2: median (max) validation accuracy = 0.981 (0.993)\n",
      "  Subject 3: median (max) validation accuracy = 0.986 (0.994)\n",
      "  Subject 4: median (max) validation accuracy = 0.985 (0.994)\n",
      "  Subject 5: median (max) validation accuracy = 0.983 (0.994)\n",
      "  Subject 6: median (max) validation accuracy = 0.983 (0.995)\n",
      "  Subject 7: median (max) validation accuracy = 0.985 (0.994)\n",
      "  Subject 8: median (max) validation accuracy = 0.985 (0.993)\n",
      "** Saving params with joint score = 0.983 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 33 of 50 took       74.780s\n",
      "  training loss:               0.075401\n",
      "  holdout loss (batch):        0.046002\n",
      "  Subject 1: median (max) validation accuracy = 0.982 (0.993)\n",
      "  Subject 2: median (max) validation accuracy = 0.981 (0.994)\n",
      "  Subject 3: median (max) validation accuracy = 0.986 (0.994)\n",
      "  Subject 4: median (max) validation accuracy = 0.985 (0.994)\n",
      "  Subject 5: median (max) validation accuracy = 0.984 (0.994)\n",
      "  Subject 6: median (max) validation accuracy = 0.983 (0.995)\n",
      "  Subject 7: median (max) validation accuracy = 0.985 (0.994)\n",
      "  Subject 8: median (max) validation accuracy = 0.985 (0.994)\n",
      "** Saving params with joint score = 0.984 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 34 of 50 took       74.744s\n",
      "  training loss:               0.074816\n",
      "  holdout loss (batch):        0.044585\n",
      "  Subject 1: median (max) validation accuracy = 0.982 (0.993)\n",
      "  Subject 2: median (max) validation accuracy = 0.981 (0.994)\n",
      "  Subject 3: median (max) validation accuracy = 0.986 (0.994)\n",
      "  Subject 4: median (max) validation accuracy = 0.985 (0.994)\n",
      "  Subject 5: median (max) validation accuracy = 0.983 (0.994)\n",
      "  Subject 6: median (max) validation accuracy = 0.984 (0.995)\n",
      "  Subject 7: median (max) validation accuracy = 0.985 (0.994)\n",
      "  Subject 8: median (max) validation accuracy = 0.985 (0.994)\n",
      "** Saving params with joint score = 0.984 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 35 of 50 took       74.747s\n",
      "  training loss:               0.073764\n",
      "  holdout loss (batch):        0.046127\n",
      "  Subject 1: median (max) validation accuracy = 0.982 (0.993)\n",
      "  Subject 2: median (max) validation accuracy = 0.982 (0.994)\n",
      "  Subject 3: median (max) validation accuracy = 0.986 (0.994)\n",
      "  Subject 4: median (max) validation accuracy = 0.986 (0.994)\n",
      "  Subject 5: median (max) validation accuracy = 0.984 (0.994)\n",
      "  Subject 6: median (max) validation accuracy = 0.984 (0.995)\n",
      "  Subject 7: median (max) validation accuracy = 0.985 (0.994)\n",
      "  Subject 8: median (max) validation accuracy = 0.985 (0.994)\n",
      "** Saving params with joint score = 0.984 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 36 of 50 took       74.799s\n",
      "  training loss:               0.073085\n",
      "  holdout loss (batch):        0.043250\n",
      "  Subject 1: median (max) validation accuracy = 0.983 (0.993)\n",
      "  Subject 2: median (max) validation accuracy = 0.982 (0.994)\n",
      "  Subject 3: median (max) validation accuracy = 0.987 (0.995)\n",
      "  Subject 4: median (max) validation accuracy = 0.986 (0.995)\n",
      "  Subject 5: median (max) validation accuracy = 0.984 (0.994)\n",
      "  Subject 6: median (max) validation accuracy = 0.984 (0.995)\n",
      "  Subject 7: median (max) validation accuracy = 0.986 (0.994)\n",
      "  Subject 8: median (max) validation accuracy = 0.986 (0.994)\n",
      "** Saving params with joint score = 0.984 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 37 of 50 took       74.798s\n",
      "  training loss:               0.072168\n",
      "  holdout loss (batch):        0.043774\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "  Subject 1: median (max) validation accuracy = 0.983 (0.994)\n",
      "  Subject 2: median (max) validation accuracy = 0.982 (0.994)\n",
      "  Subject 3: median (max) validation accuracy = 0.987 (0.995)\n",
      "  Subject 4: median (max) validation accuracy = 0.986 (0.995)\n",
      "  Subject 5: median (max) validation accuracy = 0.984 (0.994)\n",
      "  Subject 6: median (max) validation accuracy = 0.984 (0.995)\n",
      "  Subject 7: median (max) validation accuracy = 0.986 (0.994)\n",
      "  Subject 8: median (max) validation accuracy = 0.986 (0.994)\n",
      "** Saving params with joint score = 0.984 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 38 of 50 took       74.742s\n",
      "  training loss:               0.071249\n",
      "  holdout loss (batch):        0.043986\n",
      "  Subject 1: median (max) validation accuracy = 0.983 (0.994)\n",
      "  Subject 2: median (max) validation accuracy = 0.982 (0.994)\n",
      "  Subject 3: median (max) validation accuracy = 0.987 (0.995)\n",
      "  Subject 4: median (max) validation accuracy = 0.986 (0.995)\n",
      "  Subject 5: median (max) validation accuracy = 0.985 (0.995)\n",
      "  Subject 6: median (max) validation accuracy = 0.984 (0.995)\n",
      "  Subject 7: median (max) validation accuracy = 0.985 (0.994)\n",
      "  Subject 8: median (max) validation accuracy = 0.986 (0.994)\n",
      "** Saving params with joint score = 0.985 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 39 of 50 took       74.794s\n",
      "  training loss:               0.071147\n",
      "  holdout loss (batch):        0.044935\n",
      "  Subject 1: median (max) validation accuracy = 0.983 (0.994)\n",
      "  Subject 2: median (max) validation accuracy = 0.983 (0.994)\n",
      "  Subject 3: median (max) validation accuracy = 0.987 (0.995)\n",
      "  Subject 4: median (max) validation accuracy = 0.986 (0.995)\n",
      "  Subject 5: median (max) validation accuracy = 0.984 (0.995)\n",
      "  Subject 6: median (max) validation accuracy = 0.984 (0.995)\n",
      "  Subject 7: median (max) validation accuracy = 0.986 (0.994)\n",
      "  Subject 8: median (max) validation accuracy = 0.986 (0.994)\n",
      "** Saving params with joint score = 0.985 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 40 of 50 took       74.831s\n",
      "  training loss:               0.070462\n",
      "  holdout loss (batch):        0.044319\n",
      "  Subject 1: median (max) validation accuracy = 0.983 (0.994)\n",
      "  Subject 2: median (max) validation accuracy = 0.983 (0.994)\n",
      "  Subject 3: median (max) validation accuracy = 0.987 (0.995)\n",
      "  Subject 4: median (max) validation accuracy = 0.986 (0.994)\n",
      "  Subject 5: median (max) validation accuracy = 0.985 (0.994)\n",
      "  Subject 6: median (max) validation accuracy = 0.984 (0.995)\n",
      "  Subject 7: median (max) validation accuracy = 0.986 (0.994)\n",
      "  Subject 8: median (max) validation accuracy = 0.986 (0.994)\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 41 of 50 took       74.721s\n",
      "  training loss:               0.070058\n",
      "  holdout loss (batch):        0.042290\n",
      "  Subject 1: median (max) validation accuracy = 0.984 (0.994)\n",
      "  Subject 2: median (max) validation accuracy = 0.983 (0.994)\n",
      "  Subject 3: median (max) validation accuracy = 0.987 (0.995)\n",
      "  Subject 4: median (max) validation accuracy = 0.987 (0.995)\n",
      "  Subject 5: median (max) validation accuracy = 0.985 (0.995)\n",
      "  Subject 6: median (max) validation accuracy = 0.985 (0.995)\n",
      "  Subject 7: median (max) validation accuracy = 0.986 (0.994)\n",
      "  Subject 8: median (max) validation accuracy = 0.986 (0.994)\n",
      "** Saving params with joint score = 0.985 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 42 of 50 took       74.668s\n",
      "  training loss:               0.069174\n",
      "  holdout loss (batch):        0.043231\n",
      "  Subject 1: median (max) validation accuracy = 0.984 (0.994)\n",
      "  Subject 2: median (max) validation accuracy = 0.983 (0.994)\n",
      "  Subject 3: median (max) validation accuracy = 0.987 (0.995)\n",
      "  Subject 4: median (max) validation accuracy = 0.987 (0.995)\n",
      "  Subject 5: median (max) validation accuracy = 0.985 (0.994)\n",
      "  Subject 6: median (max) validation accuracy = 0.985 (0.995)\n",
      "  Subject 7: median (max) validation accuracy = 0.986 (0.994)\n",
      "  Subject 8: median (max) validation accuracy = 0.986 (0.994)\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 43 of 50 took       74.685s\n",
      "  training loss:               0.068634\n",
      "  holdout loss (batch):        0.044712\n",
      "  Subject 1: median (max) validation accuracy = 0.984 (0.994)\n",
      "  Subject 2: median (max) validation accuracy = 0.983 (0.994)\n",
      "  Subject 3: median (max) validation accuracy = 0.987 (0.995)\n",
      "  Subject 4: median (max) validation accuracy = 0.986 (0.994)\n",
      "  Subject 5: median (max) validation accuracy = 0.985 (0.994)\n",
      "  Subject 6: median (max) validation accuracy = 0.985 (0.995)\n",
      "  Subject 7: median (max) validation accuracy = 0.986 (0.994)\n",
      "  Subject 8: median (max) validation accuracy = 0.986 (0.994)\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 44 of 50 took       74.665s\n",
      "  training loss:               0.068944\n",
      "  holdout loss (batch):        0.041709\n",
      "  Subject 1: median (max) validation accuracy = 0.984 (0.994)\n",
      "  Subject 2: median (max) validation accuracy = 0.984 (0.995)\n",
      "  Subject 3: median (max) validation accuracy = 0.987 (0.995)\n",
      "  Subject 4: median (max) validation accuracy = 0.987 (0.995)\n",
      "  Subject 5: median (max) validation accuracy = 0.985 (0.995)\n",
      "  Subject 6: median (max) validation accuracy = 0.985 (0.995)\n",
      "  Subject 7: median (max) validation accuracy = 0.986 (0.995)\n",
      "  Subject 8: median (max) validation accuracy = 0.987 (0.994)\n",
      "** Saving params with joint score = 0.986 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 45 of 50 took       74.734s\n",
      "  training loss:               0.068420\n",
      "  holdout loss (batch):        0.046901\n",
      "  Subject 1: median (max) validation accuracy = 0.984 (0.994)\n",
      "  Subject 2: median (max) validation accuracy = 0.984 (0.994)\n",
      "  Subject 3: median (max) validation accuracy = 0.987 (0.995)\n",
      "  Subject 4: median (max) validation accuracy = 0.987 (0.994)\n",
      "  Subject 5: median (max) validation accuracy = 0.985 (0.995)\n",
      "  Subject 6: median (max) validation accuracy = 0.985 (0.995)\n",
      "  Subject 7: median (max) validation accuracy = 0.986 (0.995)\n",
      "  Subject 8: median (max) validation accuracy = 0.987 (0.994)\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 46 of 50 took       74.696s\n",
      "  training loss:               0.067738\n",
      "  holdout loss (batch):        0.043483\n",
      "  Subject 1: median (max) validation accuracy = 0.985 (0.994)\n",
      "  Subject 2: median (max) validation accuracy = 0.984 (0.995)\n",
      "  Subject 3: median (max) validation accuracy = 0.987 (0.995)\n",
      "  Subject 4: median (max) validation accuracy = 0.987 (0.994)\n",
      "  Subject 5: median (max) validation accuracy = 0.985 (0.994)\n",
      "  Subject 6: median (max) validation accuracy = 0.985 (0.995)\n",
      "  Subject 7: median (max) validation accuracy = 0.986 (0.994)\n",
      "  Subject 8: median (max) validation accuracy = 0.987 (0.994)\n",
      "** Saving params with joint score = 0.986 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 47 of 50 took       74.751s\n",
      "  training loss:               0.067090\n",
      "  holdout loss (batch):        0.041684\n",
      "  Subject 1: median (max) validation accuracy = 0.985 (0.994)\n",
      "  Subject 2: median (max) validation accuracy = 0.984 (0.994)\n",
      "  Subject 3: median (max) validation accuracy = 0.988 (0.995)\n",
      "  Subject 4: median (max) validation accuracy = 0.987 (0.995)\n",
      "  Subject 5: median (max) validation accuracy = 0.985 (0.995)\n",
      "  Subject 6: median (max) validation accuracy = 0.985 (0.995)\n",
      "  Subject 7: median (max) validation accuracy = 0.986 (0.995)\n",
      "  Subject 8: median (max) validation accuracy = 0.987 (0.994)\n",
      "** Saving params with joint score = 0.986 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 48 of 50 took       74.820s\n",
      "  training loss:               0.066829\n",
      "  holdout loss (batch):        0.040054\n",
      "  Subject 1: median (max) validation accuracy = 0.985 (0.994)\n",
      "  Subject 2: median (max) validation accuracy = 0.984 (0.995)\n",
      "  Subject 3: median (max) validation accuracy = 0.988 (0.995)\n",
      "  Subject 4: median (max) validation accuracy = 0.987 (0.995)\n",
      "  Subject 5: median (max) validation accuracy = 0.985 (0.995)\n",
      "  Subject 6: median (max) validation accuracy = 0.985 (0.995)\n",
      "  Subject 7: median (max) validation accuracy = 0.987 (0.995)\n",
      "  Subject 8: median (max) validation accuracy = 0.987 (0.994)\n",
      "** Saving params with joint score = 0.986 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 49 of 50 took       74.776s\n",
      "  training loss:               0.066485\n",
      "  holdout loss (batch):        0.045030\n",
      "  Subject 1: median (max) validation accuracy = 0.985 (0.994)\n",
      "  Subject 2: median (max) validation accuracy = 0.984 (0.995)\n",
      "  Subject 3: median (max) validation accuracy = 0.988 (0.995)\n",
      "  Subject 4: median (max) validation accuracy = 0.987 (0.995)\n",
      "  Subject 5: median (max) validation accuracy = 0.985 (0.995)\n",
      "  Subject 6: median (max) validation accuracy = 0.985 (0.995)\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "  Subject 7: median (max) validation accuracy = 0.987 (0.995)\n",
      "  Subject 8: median (max) validation accuracy = 0.987 (0.994)\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 50 of 50 took       74.759s\n",
      "  training loss:               0.066681\n",
      "  holdout loss (batch):        0.040155\n",
      "  Subject 1: median (max) validation accuracy = 0.985 (0.995)\n",
      "  Subject 2: median (max) validation accuracy = 0.984 (0.994)\n",
      "  Subject 3: median (max) validation accuracy = 0.988 (0.995)\n",
      "  Subject 4: median (max) validation accuracy = 0.987 (0.995)\n",
      "  Subject 5: median (max) validation accuracy = 0.986 (0.995)\n",
      "  Subject 6: median (max) validation accuracy = 0.986 (0.995)\n",
      "  Subject 7: median (max) validation accuracy = 0.987 (0.995)\n",
      "  Subject 8: median (max) validation accuracy = 0.987 (0.994)\n",
      "** Saving params with joint score = 0.986 **\n",
      "\n"
     ]
    }
   ],
   "source": [
    "batch_size = 100\n",
    "num_epochs = 50\n",
    "holdout_frac = .1\n",
    "\n",
    "best_params, final_params, hold_cc_hist, hold_hist, trn_hist, best_epoch, best_joint_cc_score = \\\n",
    "    learn_params_(_training_fn, _holdout_fn, _pred_fn, shared_model, subject_fwrfs, subject_opts, \n",
    "        stim_data, trn_voxel_data, trn_stim_ordering,\n",
    "        num_epochs=num_epochs, batch_size=batch_size, holdout_frac=holdout_frac, masks=None, randomize=False)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Results"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 32,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 1728x216 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.figure(figsize=(24,3))\n",
    "plt.plot(trn_hist, color='r', marker='o', ms=10, lw=2)\n",
    "plt.plot(np.array(hold_hist), color='b', marker='o', ms=10, lw=2)\n",
    "plt.ylabel('loss')\n",
    "plt.xlabel('epoch')\n",
    "plt.grid()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 33,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "No handles with labels found to put in legend.\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "Text(0.5, 0, 'epoch')"
      ]
     },
     "execution_count": 33,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 1728x432 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.figure(figsize=(24,6))\n",
    "plt.subplot(2,1,1)\n",
    "for s,cc in hold_cc_hist.items():\n",
    "    plt.gca().violinplot(np.nan_to_num(np.array(cc)).T, showmeans=True, showmedians=True, showextrema=True)\n",
    "plt.ylabel('Validation history')\n",
    "plt.subplot(2,1,2)\n",
    "for s in hold_cc_hist.keys():\n",
    "    p=plt.plot([np.median(cc) for cc in hold_cc_hist[s]], lw=2, linestyle='--')\n",
    "plt.legend()\n",
    "plt.ylabel('Median history')\n",
    "plt.xlabel('epoch')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "---\n",
    "## Final validation accuracy"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 34,
   "metadata": {},
   "outputs": [],
   "source": [
    "val_voxel = {s: val_voxel_data[s] for s in val_voxel_data.keys()}\n",
    "shared_model.load_state_dict(best_params['enc'])\n",
    "shared_model.eval() \n",
    "for s,sd in subject_fwrfs.items():\n",
    "    sd.load_state_dict(best_params['fwrfs'][s])\n",
    "    sd.eval() \n",
    "\n",
    "batch_size = 50\n",
    "subject_val_cc = validation_(_pred_fn, shared_model, subject_fwrfs, stim_data, val_voxel, val_stim_ordering, batch_size)\n",
    "joined_val_cc = np.concatenate(list(subject_val_cc.values()), axis=0)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 35,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "best joint score = 0.986\n",
      "best joint val cc = 0.987\n",
      "subject 1: val cc = 0.986\n",
      "subject 2: val cc = 0.985\n",
      "subject 3: val cc = 0.989\n",
      "subject 4: val cc = 0.988\n",
      "subject 5: val cc = 0.987\n",
      "subject 6: val cc = 0.987\n",
      "subject 7: val cc = 0.987\n",
      "subject 8: val cc = 0.988\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 864x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "print (\"best joint score = %.3f\"%best_joint_cc_score)\n",
    "print (\"best joint val cc = %.3f\"% np.median(joined_val_cc))\n",
    "for s,v in subject_val_cc.items():\n",
    "    print (\"subject %s: val cc = %.3f\"%(s, np.median(v)))\n",
    "\n",
    "fig=plt.figure(figsize=(12,4))\n",
    "plt.subplots_adjust(left=0., bottom=0.2, right=1., top=1., wspace=0., hspace=0.)\n",
    "_=plt.hist(joined_val_cc, bins=100, density=True, range=(-.5, 1.))\n",
    "_=plt.vlines(x=[0], ymin=1e-4, ymax=25, color='r')\n",
    "_=plt.yscale('log')\n",
    "_=plt.xlim([-.1, 1.])\n",
    "_=plt.xlabel(r'$\\rho$')\n",
    "_=plt.ylabel('Relative frequency')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Save final parameters"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 38,
   "metadata": {},
   "outputs": [],
   "source": [
    "torch.save({\n",
    "            'num_epochs': num_epochs,\n",
    "            'batch_size': batch_size,\n",
    "            'holdout_frac': holdout_frac,   \n",
    "            'best_params': best_params, \n",
    "            'final_params': final_params,\n",
    "            'trn_loss_history': trn_hist,\n",
    "            'hold_loss_history': hold_hist,\n",
    "            'hold_cc_history': hold_cc_hist,\n",
    "            'best_epoch': best_epoch,\n",
    "            'best_joint_cc_score': best_joint_cc_score,\n",
    "            'val_cc': subject_val_cc,\n",
    "            'input_mean': trn_stim_mean,\n",
    "            'voxel_roi': voxel_roi,\n",
    "            'voxel_mask': voxel_mask,\n",
    "            'voxel_index': voxel_idx\n",
    "            }, output_dir+'model_params_%s'%noise_model)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "---\n",
    "# ROI-wise models"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 39,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "['V1.1: 1348', 'V1.2: 1097', 'V1.3: 1251', 'V1.4: 871', 'V1.5: 1115', 'V1.6: 1130', 'V1.7: 1148', 'V1.8: 1081', 'V2.1: 1415', 'V2.2: 1083', 'V2.3: 1160', 'V2.4: 883', 'V2.5: 1088', 'V2.6: 1191', 'V2.7: 958', 'V2.8: 1040', 'V3.1: 1188', 'V3.2: 1081', 'V3.3: 945', 'V3.4: 800', 'V3.5: 934', 'V3.6: 1215', 'V3.7: 721', 'V3.8: 879', 'hV4.1: 676', 'hV4.2: 504', 'hV4.3: 422', 'hV4.4: 473', 'hV4.5: 548', 'hV4.6: 488', 'hV4.7: 374', 'hV4.8: 490']\n"
     ]
    }
   ],
   "source": [
    "block_voxel_masks = {}\n",
    "for roi_name in block_group_names:\n",
    "    block_voxel_masks[roi_name] = {}\n",
    "\n",
    "for s in subjects:\n",
    "    for roi_mask, roi_name in iterate_roi(block_group, voxel_roi[s], roi_map, group_name=block_group_names):    \n",
    "        block_voxel_masks[roi_name][s] = roi_mask\n",
    "\n",
    "print (['%s: %d'%(s,np.sum(v)) for s,v in flatten_dict(block_voxel_masks).items()])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 42,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "***************************************\n",
      "***                  V1       ***\n",
      "{1: 1348, 2: 1097, 3: 1251, 4: 871, 5: 1115, 6: 1130, 7: 1148, 8: 1081}\n",
      "***************************************\n",
      "subject 1 masked 1348 of 4627\n",
      "subject 1 training/holdout 8100 900\n",
      "subject 2 masked 1097 of 3765\n",
      "subject 2 training/holdout 8100 900\n",
      "subject 3 masked 1251 of 3778\n",
      "subject 3 training/holdout 8100 900\n",
      "subject 4 masked 871 of 3027\n",
      "subject 4 training/holdout 8100 900\n",
      "subject 5 masked 1115 of 3685\n",
      "subject 5 training/holdout 8100 900\n",
      "subject 6 masked 1130 of 4024\n",
      "subject 6 training/holdout 8100 900\n",
      "subject 7 masked 1148 of 3201\n",
      "subject 7 training/holdout 8100 900\n",
      "subject 8 masked 1081 of 3490\n",
      "subject 8 training/holdout 8100 900\n",
      "\\ : 100.0 %\n",
      "  Epoch 1 of 50 took       46.562s\n",
      "  training loss:               0.610270\n",
      "  holdout loss (batch):        0.469995\n",
      "  Subject 1: median (max) validation accuracy = 0.577 (0.888)\n",
      "  Subject 2: median (max) validation accuracy = 0.635 (0.912)\n",
      "  Subject 3: median (max) validation accuracy = 0.748 (0.946)\n",
      "  Subject 4: median (max) validation accuracy = 0.778 (0.929)\n",
      "  Subject 5: median (max) validation accuracy = 0.849 (0.950)\n",
      "  Subject 6: median (max) validation accuracy = 0.755 (0.950)\n",
      "  Subject 7: median (max) validation accuracy = 0.717 (0.918)\n",
      "  Subject 8: median (max) validation accuracy = 0.795 (0.933)\n",
      "** Saving params with joint score = 0.742 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 2 of 50 took       46.598s\n",
      "  training loss:               0.462144\n",
      "  holdout loss (batch):        0.387482\n",
      "  Subject 1: median (max) validation accuracy = 0.650 (0.910)\n",
      "  Subject 2: median (max) validation accuracy = 0.693 (0.934)\n",
      "  Subject 3: median (max) validation accuracy = 0.800 (0.957)\n",
      "  Subject 4: median (max) validation accuracy = 0.825 (0.949)\n",
      "  Subject 5: median (max) validation accuracy = 0.894 (0.961)\n",
      "  Subject 6: median (max) validation accuracy = 0.815 (0.962)\n",
      "  Subject 7: median (max) validation accuracy = 0.780 (0.932)\n",
      "  Subject 8: median (max) validation accuracy = 0.836 (0.954)\n",
      "** Saving params with joint score = 0.798 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 3 of 50 took       46.842s\n",
      "  training loss:               0.399984\n",
      "  holdout loss (batch):        0.330333\n",
      "  Subject 1: median (max) validation accuracy = 0.700 (0.933)\n",
      "  Subject 2: median (max) validation accuracy = 0.741 (0.945)\n",
      "  Subject 3: median (max) validation accuracy = 0.834 (0.964)\n",
      "  Subject 4: median (max) validation accuracy = 0.860 (0.957)\n",
      "  Subject 5: median (max) validation accuracy = 0.917 (0.967)\n",
      "  Subject 6: median (max) validation accuracy = 0.850 (0.966)\n",
      "  Subject 7: median (max) validation accuracy = 0.823 (0.948)\n",
      "  Subject 8: median (max) validation accuracy = 0.868 (0.960)\n",
      "** Saving params with joint score = 0.835 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 4 of 50 took       46.714s\n",
      "  training loss:               0.347258\n",
      "  holdout loss (batch):        0.285211\n",
      "  Subject 1: median (max) validation accuracy = 0.744 (0.944)\n",
      "  Subject 2: median (max) validation accuracy = 0.782 (0.954)\n",
      "  Subject 3: median (max) validation accuracy = 0.862 (0.968)\n",
      "  Subject 4: median (max) validation accuracy = 0.886 (0.965)\n",
      "  Subject 5: median (max) validation accuracy = 0.932 (0.973)\n",
      "  Subject 6: median (max) validation accuracy = 0.876 (0.971)\n",
      "  Subject 7: median (max) validation accuracy = 0.857 (0.962)\n",
      "  Subject 8: median (max) validation accuracy = 0.892 (0.966)\n",
      "** Saving params with joint score = 0.863 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 5 of 50 took       46.758s\n",
      "  training loss:               0.307102\n",
      "  holdout loss (batch):        0.256919\n",
      "  Subject 1: median (max) validation accuracy = 0.776 (0.958)\n",
      "  Subject 2: median (max) validation accuracy = 0.816 (0.965)\n",
      "  Subject 3: median (max) validation accuracy = 0.881 (0.970)\n",
      "  Subject 4: median (max) validation accuracy = 0.903 (0.969)\n",
      "  Subject 5: median (max) validation accuracy = 0.939 (0.976)\n",
      "  Subject 6: median (max) validation accuracy = 0.894 (0.977)\n",
      "  Subject 7: median (max) validation accuracy = 0.881 (0.969)\n",
      "  Subject 8: median (max) validation accuracy = 0.907 (0.972)\n",
      "** Saving params with joint score = 0.884 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 6 of 50 took       46.878s\n",
      "  training loss:               0.271830\n",
      "  holdout loss (batch):        0.212417\n",
      "  Subject 1: median (max) validation accuracy = 0.812 (0.967)\n",
      "  Subject 2: median (max) validation accuracy = 0.845 (0.973)\n",
      "  Subject 3: median (max) validation accuracy = 0.903 (0.973)\n",
      "  Subject 4: median (max) validation accuracy = 0.922 (0.977)\n",
      "  Subject 5: median (max) validation accuracy = 0.948 (0.981)\n",
      "  Subject 6: median (max) validation accuracy = 0.913 (0.982)\n",
      "  Subject 7: median (max) validation accuracy = 0.906 (0.974)\n",
      "  Subject 8: median (max) validation accuracy = 0.924 (0.978)\n",
      "** Saving params with joint score = 0.905 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 7 of 50 took       46.783s\n",
      "  training loss:               0.238234\n",
      "  holdout loss (batch):        0.190984\n",
      "  Subject 1: median (max) validation accuracy = 0.838 (0.974)\n",
      "  Subject 2: median (max) validation accuracy = 0.867 (0.979)\n",
      "  Subject 3: median (max) validation accuracy = 0.918 (0.976)\n",
      "  Subject 4: median (max) validation accuracy = 0.935 (0.982)\n",
      "  Subject 5: median (max) validation accuracy = 0.955 (0.984)\n",
      "  Subject 6: median (max) validation accuracy = 0.927 (0.985)\n",
      "  Subject 7: median (max) validation accuracy = 0.923 (0.977)\n",
      "  Subject 8: median (max) validation accuracy = 0.936 (0.982)\n",
      "** Saving params with joint score = 0.920 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 8 of 50 took       46.806s\n",
      "  training loss:               0.210799\n",
      "  holdout loss (batch):        0.161137\n",
      "  Subject 1: median (max) validation accuracy = 0.861 (0.978)\n",
      "  Subject 2: median (max) validation accuracy = 0.888 (0.982)\n",
      "  Subject 3: median (max) validation accuracy = 0.930 (0.981)\n",
      "  Subject 4: median (max) validation accuracy = 0.945 (0.985)\n",
      "  Subject 5: median (max) validation accuracy = 0.961 (0.986)\n",
      "  Subject 6: median (max) validation accuracy = 0.937 (0.987)\n",
      "  Subject 7: median (max) validation accuracy = 0.937 (0.980)\n",
      "  Subject 8: median (max) validation accuracy = 0.946 (0.985)\n",
      "** Saving params with joint score = 0.933 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 9 of 50 took       46.759s\n",
      "  training loss:               0.187728\n",
      "  holdout loss (batch):        0.136407\n",
      "  Subject 1: median (max) validation accuracy = 0.883 (0.980)\n",
      "  Subject 2: median (max) validation accuracy = 0.905 (0.983)\n",
      "  Subject 3: median (max) validation accuracy = 0.942 (0.984)\n",
      "  Subject 4: median (max) validation accuracy = 0.954 (0.987)\n",
      "  Subject 5: median (max) validation accuracy = 0.967 (0.988)\n",
      "  Subject 6: median (max) validation accuracy = 0.948 (0.989)\n",
      "  Subject 7: median (max) validation accuracy = 0.948 (0.983)\n",
      "  Subject 8: median (max) validation accuracy = 0.955 (0.987)\n",
      "** Saving params with joint score = 0.944 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 10 of 50 took       46.842s\n",
      "  training loss:               0.169374\n",
      "  holdout loss (batch):        0.135824\n",
      "  Subject 1: median (max) validation accuracy = 0.898 (0.981)\n",
      "  Subject 2: median (max) validation accuracy = 0.917 (0.985)\n",
      "  Subject 3: median (max) validation accuracy = 0.949 (0.984)\n",
      "  Subject 4: median (max) validation accuracy = 0.961 (0.987)\n",
      "  Subject 5: median (max) validation accuracy = 0.968 (0.988)\n",
      "  Subject 6: median (max) validation accuracy = 0.955 (0.989)\n",
      "  Subject 7: median (max) validation accuracy = 0.953 (0.983)\n",
      "  Subject 8: median (max) validation accuracy = 0.960 (0.988)\n",
      "** Saving params with joint score = 0.950 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 11 of 50 took       46.788s\n",
      "  training loss:               0.152265\n",
      "  holdout loss (batch):        0.111012\n",
      "  Subject 1: median (max) validation accuracy = 0.912 (0.983)\n",
      "  Subject 2: median (max) validation accuracy = 0.929 (0.986)\n",
      "  Subject 3: median (max) validation accuracy = 0.956 (0.985)\n",
      "  Subject 4: median (max) validation accuracy = 0.967 (0.989)\n",
      "  Subject 5: median (max) validation accuracy = 0.970 (0.989)\n",
      "  Subject 6: median (max) validation accuracy = 0.961 (0.990)\n",
      "  Subject 7: median (max) validation accuracy = 0.960 (0.984)\n",
      "  Subject 8: median (max) validation accuracy = 0.965 (0.989)\n",
      "** Saving params with joint score = 0.956 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 12 of 50 took       46.784s\n",
      "  training loss:               0.139166\n",
      "  holdout loss (batch):        0.098378\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "  Subject 1: median (max) validation accuracy = 0.927 (0.985)\n",
      "  Subject 2: median (max) validation accuracy = 0.939 (0.987)\n",
      "  Subject 3: median (max) validation accuracy = 0.963 (0.987)\n",
      "  Subject 4: median (max) validation accuracy = 0.971 (0.990)\n",
      "  Subject 5: median (max) validation accuracy = 0.974 (0.991)\n",
      "  Subject 6: median (max) validation accuracy = 0.966 (0.991)\n",
      "  Subject 7: median (max) validation accuracy = 0.966 (0.986)\n",
      "  Subject 8: median (max) validation accuracy = 0.971 (0.990)\n",
      "** Saving params with joint score = 0.962 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 13 of 50 took       46.891s\n",
      "  training loss:               0.124828\n",
      "  holdout loss (batch):        0.092113\n",
      "  Subject 1: median (max) validation accuracy = 0.937 (0.986)\n",
      "  Subject 2: median (max) validation accuracy = 0.947 (0.989)\n",
      "  Subject 3: median (max) validation accuracy = 0.968 (0.988)\n",
      "  Subject 4: median (max) validation accuracy = 0.974 (0.991)\n",
      "  Subject 5: median (max) validation accuracy = 0.976 (0.991)\n",
      "  Subject 6: median (max) validation accuracy = 0.970 (0.992)\n",
      "  Subject 7: median (max) validation accuracy = 0.970 (0.987)\n",
      "  Subject 8: median (max) validation accuracy = 0.974 (0.991)\n",
      "** Saving params with joint score = 0.967 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 14 of 50 took       46.890s\n",
      "  training loss:               0.114143\n",
      "  holdout loss (batch):        0.075728\n",
      "  Subject 1: median (max) validation accuracy = 0.947 (0.987)\n",
      "  Subject 2: median (max) validation accuracy = 0.954 (0.989)\n",
      "  Subject 3: median (max) validation accuracy = 0.973 (0.989)\n",
      "  Subject 4: median (max) validation accuracy = 0.978 (0.991)\n",
      "  Subject 5: median (max) validation accuracy = 0.978 (0.992)\n",
      "  Subject 6: median (max) validation accuracy = 0.974 (0.992)\n",
      "  Subject 7: median (max) validation accuracy = 0.974 (0.989)\n",
      "  Subject 8: median (max) validation accuracy = 0.977 (0.991)\n",
      "** Saving params with joint score = 0.971 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 15 of 50 took       46.908s\n",
      "  training loss:               0.103206\n",
      "  holdout loss (batch):        0.067690\n",
      "  Subject 1: median (max) validation accuracy = 0.955 (0.987)\n",
      "  Subject 2: median (max) validation accuracy = 0.961 (0.990)\n",
      "  Subject 3: median (max) validation accuracy = 0.976 (0.990)\n",
      "  Subject 4: median (max) validation accuracy = 0.980 (0.992)\n",
      "  Subject 5: median (max) validation accuracy = 0.979 (0.992)\n",
      "  Subject 6: median (max) validation accuracy = 0.977 (0.993)\n",
      "  Subject 7: median (max) validation accuracy = 0.977 (0.990)\n",
      "  Subject 8: median (max) validation accuracy = 0.979 (0.992)\n",
      "** Saving params with joint score = 0.974 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 16 of 50 took       47.164s\n",
      "  training loss:               0.096026\n",
      "  holdout loss (batch):        0.068846\n",
      "  Subject 1: median (max) validation accuracy = 0.961 (0.988)\n",
      "  Subject 2: median (max) validation accuracy = 0.965 (0.990)\n",
      "  Subject 3: median (max) validation accuracy = 0.978 (0.991)\n",
      "  Subject 4: median (max) validation accuracy = 0.982 (0.992)\n",
      "  Subject 5: median (max) validation accuracy = 0.980 (0.992)\n",
      "  Subject 6: median (max) validation accuracy = 0.979 (0.992)\n",
      "  Subject 7: median (max) validation accuracy = 0.979 (0.990)\n",
      "  Subject 8: median (max) validation accuracy = 0.980 (0.991)\n",
      "** Saving params with joint score = 0.977 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 17 of 50 took       46.977s\n",
      "  training loss:               0.088087\n",
      "  holdout loss (batch):        0.057314\n",
      "  Subject 1: median (max) validation accuracy = 0.966 (0.989)\n",
      "  Subject 2: median (max) validation accuracy = 0.971 (0.991)\n",
      "  Subject 3: median (max) validation accuracy = 0.981 (0.991)\n",
      "  Subject 4: median (max) validation accuracy = 0.984 (0.993)\n",
      "  Subject 5: median (max) validation accuracy = 0.982 (0.993)\n",
      "  Subject 6: median (max) validation accuracy = 0.981 (0.994)\n",
      "  Subject 7: median (max) validation accuracy = 0.982 (0.991)\n",
      "  Subject 8: median (max) validation accuracy = 0.983 (0.993)\n",
      "** Saving params with joint score = 0.979 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 18 of 50 took       46.872s\n",
      "  training loss:               0.081523\n",
      "  holdout loss (batch):        0.050878\n",
      "  Subject 1: median (max) validation accuracy = 0.972 (0.990)\n",
      "  Subject 2: median (max) validation accuracy = 0.974 (0.992)\n",
      "  Subject 3: median (max) validation accuracy = 0.983 (0.992)\n",
      "  Subject 4: median (max) validation accuracy = 0.985 (0.993)\n",
      "  Subject 5: median (max) validation accuracy = 0.983 (0.994)\n",
      "  Subject 6: median (max) validation accuracy = 0.983 (0.994)\n",
      "  Subject 7: median (max) validation accuracy = 0.983 (0.992)\n",
      "  Subject 8: median (max) validation accuracy = 0.984 (0.993)\n",
      "** Saving params with joint score = 0.982 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 19 of 50 took       47.155s\n",
      "  training loss:               0.077816\n",
      "  holdout loss (batch):        0.051639\n",
      "  Subject 1: median (max) validation accuracy = 0.975 (0.990)\n",
      "  Subject 2: median (max) validation accuracy = 0.977 (0.992)\n",
      "  Subject 3: median (max) validation accuracy = 0.985 (0.992)\n",
      "  Subject 4: median (max) validation accuracy = 0.986 (0.993)\n",
      "  Subject 5: median (max) validation accuracy = 0.984 (0.994)\n",
      "  Subject 6: median (max) validation accuracy = 0.984 (0.994)\n",
      "  Subject 7: median (max) validation accuracy = 0.984 (0.992)\n",
      "  Subject 8: median (max) validation accuracy = 0.986 (0.993)\n",
      "** Saving params with joint score = 0.983 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 20 of 50 took       47.356s\n",
      "  training loss:               0.072844\n",
      "  holdout loss (batch):        0.045345\n",
      "  Subject 1: median (max) validation accuracy = 0.978 (0.991)\n",
      "  Subject 2: median (max) validation accuracy = 0.979 (0.993)\n",
      "  Subject 3: median (max) validation accuracy = 0.985 (0.993)\n",
      "  Subject 4: median (max) validation accuracy = 0.987 (0.994)\n",
      "  Subject 5: median (max) validation accuracy = 0.985 (0.994)\n",
      "  Subject 6: median (max) validation accuracy = 0.985 (0.994)\n",
      "  Subject 7: median (max) validation accuracy = 0.985 (0.993)\n",
      "  Subject 8: median (max) validation accuracy = 0.986 (0.993)\n",
      "** Saving params with joint score = 0.984 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 21 of 50 took       47.363s\n",
      "  training loss:               0.070750\n",
      "  holdout loss (batch):        0.042137\n",
      "  Subject 1: median (max) validation accuracy = 0.980 (0.991)\n",
      "  Subject 2: median (max) validation accuracy = 0.981 (0.993)\n",
      "  Subject 3: median (max) validation accuracy = 0.987 (0.993)\n",
      "  Subject 4: median (max) validation accuracy = 0.988 (0.994)\n",
      "  Subject 5: median (max) validation accuracy = 0.985 (0.994)\n",
      "  Subject 6: median (max) validation accuracy = 0.986 (0.994)\n",
      "  Subject 7: median (max) validation accuracy = 0.987 (0.993)\n",
      "  Subject 8: median (max) validation accuracy = 0.987 (0.994)\n",
      "** Saving params with joint score = 0.985 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 22 of 50 took       46.742s\n",
      "  training loss:               0.066918\n",
      "  holdout loss (batch):        0.041097\n",
      "  Subject 1: median (max) validation accuracy = 0.981 (0.992)\n",
      "  Subject 2: median (max) validation accuracy = 0.983 (0.993)\n",
      "  Subject 3: median (max) validation accuracy = 0.988 (0.993)\n",
      "  Subject 4: median (max) validation accuracy = 0.989 (0.994)\n",
      "  Subject 5: median (max) validation accuracy = 0.986 (0.994)\n",
      "  Subject 6: median (max) validation accuracy = 0.987 (0.995)\n",
      "  Subject 7: median (max) validation accuracy = 0.987 (0.993)\n",
      "  Subject 8: median (max) validation accuracy = 0.988 (0.994)\n",
      "** Saving params with joint score = 0.986 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 23 of 50 took       46.762s\n",
      "  training loss:               0.064585\n",
      "  holdout loss (batch):        0.036605\n",
      "  Subject 1: median (max) validation accuracy = 0.983 (0.992)\n",
      "  Subject 2: median (max) validation accuracy = 0.984 (0.994)\n",
      "  Subject 3: median (max) validation accuracy = 0.989 (0.994)\n",
      "  Subject 4: median (max) validation accuracy = 0.989 (0.995)\n",
      "  Subject 5: median (max) validation accuracy = 0.986 (0.995)\n",
      "  Subject 6: median (max) validation accuracy = 0.988 (0.995)\n",
      "  Subject 7: median (max) validation accuracy = 0.988 (0.994)\n",
      "  Subject 8: median (max) validation accuracy = 0.989 (0.994)\n",
      "** Saving params with joint score = 0.987 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 24 of 50 took       46.752s\n",
      "  training loss:               0.062237\n",
      "  holdout loss (batch):        0.040061\n",
      "  Subject 1: median (max) validation accuracy = 0.984 (0.992)\n",
      "  Subject 2: median (max) validation accuracy = 0.985 (0.994)\n",
      "  Subject 3: median (max) validation accuracy = 0.989 (0.994)\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "  Subject 4: median (max) validation accuracy = 0.990 (0.995)\n",
      "  Subject 5: median (max) validation accuracy = 0.987 (0.995)\n",
      "  Subject 6: median (max) validation accuracy = 0.988 (0.995)\n",
      "  Subject 7: median (max) validation accuracy = 0.988 (0.993)\n",
      "  Subject 8: median (max) validation accuracy = 0.989 (0.994)\n",
      "** Saving params with joint score = 0.987 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 25 of 50 took       46.756s\n",
      "  training loss:               0.060931\n",
      "  holdout loss (batch):        0.036839\n",
      "  Subject 1: median (max) validation accuracy = 0.985 (0.993)\n",
      "  Subject 2: median (max) validation accuracy = 0.986 (0.994)\n",
      "  Subject 3: median (max) validation accuracy = 0.989 (0.994)\n",
      "  Subject 4: median (max) validation accuracy = 0.990 (0.995)\n",
      "  Subject 5: median (max) validation accuracy = 0.987 (0.995)\n",
      "  Subject 6: median (max) validation accuracy = 0.989 (0.995)\n",
      "  Subject 7: median (max) validation accuracy = 0.989 (0.994)\n",
      "  Subject 8: median (max) validation accuracy = 0.989 (0.995)\n",
      "** Saving params with joint score = 0.988 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 26 of 50 took       46.798s\n",
      "  training loss:               0.059407\n",
      "  holdout loss (batch):        0.037174\n",
      "  Subject 1: median (max) validation accuracy = 0.985 (0.993)\n",
      "  Subject 2: median (max) validation accuracy = 0.986 (0.994)\n",
      "  Subject 3: median (max) validation accuracy = 0.990 (0.994)\n",
      "  Subject 4: median (max) validation accuracy = 0.990 (0.995)\n",
      "  Subject 5: median (max) validation accuracy = 0.988 (0.995)\n",
      "  Subject 6: median (max) validation accuracy = 0.990 (0.996)\n",
      "  Subject 7: median (max) validation accuracy = 0.989 (0.994)\n",
      "  Subject 8: median (max) validation accuracy = 0.990 (0.995)\n",
      "** Saving params with joint score = 0.988 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 27 of 50 took       46.770s\n",
      "  training loss:               0.058125\n",
      "  holdout loss (batch):        0.038403\n",
      "  Subject 1: median (max) validation accuracy = 0.986 (0.993)\n",
      "  Subject 2: median (max) validation accuracy = 0.987 (0.995)\n",
      "  Subject 3: median (max) validation accuracy = 0.990 (0.995)\n",
      "  Subject 4: median (max) validation accuracy = 0.991 (0.995)\n",
      "  Subject 5: median (max) validation accuracy = 0.988 (0.995)\n",
      "  Subject 6: median (max) validation accuracy = 0.990 (0.995)\n",
      "  Subject 7: median (max) validation accuracy = 0.990 (0.994)\n",
      "  Subject 8: median (max) validation accuracy = 0.990 (0.995)\n",
      "** Saving params with joint score = 0.989 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 28 of 50 took       46.815s\n",
      "  training loss:               0.057302\n",
      "  holdout loss (batch):        0.036011\n",
      "  Subject 1: median (max) validation accuracy = 0.987 (0.994)\n",
      "  Subject 2: median (max) validation accuracy = 0.988 (0.995)\n",
      "  Subject 3: median (max) validation accuracy = 0.991 (0.995)\n",
      "  Subject 4: median (max) validation accuracy = 0.991 (0.995)\n",
      "  Subject 5: median (max) validation accuracy = 0.989 (0.995)\n",
      "  Subject 6: median (max) validation accuracy = 0.990 (0.996)\n",
      "  Subject 7: median (max) validation accuracy = 0.990 (0.995)\n",
      "  Subject 8: median (max) validation accuracy = 0.991 (0.995)\n",
      "** Saving params with joint score = 0.990 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 29 of 50 took       46.810s\n",
      "  training loss:               0.055849\n",
      "  holdout loss (batch):        0.033305\n",
      "  Subject 1: median (max) validation accuracy = 0.987 (0.994)\n",
      "  Subject 2: median (max) validation accuracy = 0.987 (0.994)\n",
      "  Subject 3: median (max) validation accuracy = 0.990 (0.995)\n",
      "  Subject 4: median (max) validation accuracy = 0.991 (0.995)\n",
      "  Subject 5: median (max) validation accuracy = 0.989 (0.995)\n",
      "  Subject 6: median (max) validation accuracy = 0.990 (0.996)\n",
      "  Subject 7: median (max) validation accuracy = 0.990 (0.994)\n",
      "  Subject 8: median (max) validation accuracy = 0.990 (0.995)\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 30 of 50 took       46.755s\n",
      "  training loss:               0.055050\n",
      "  holdout loss (batch):        0.029190\n",
      "  Subject 1: median (max) validation accuracy = 0.988 (0.994)\n",
      "  Subject 2: median (max) validation accuracy = 0.988 (0.995)\n",
      "  Subject 3: median (max) validation accuracy = 0.991 (0.995)\n",
      "  Subject 4: median (max) validation accuracy = 0.991 (0.995)\n",
      "  Subject 5: median (max) validation accuracy = 0.989 (0.995)\n",
      "  Subject 6: median (max) validation accuracy = 0.991 (0.996)\n",
      "  Subject 7: median (max) validation accuracy = 0.990 (0.995)\n",
      "  Subject 8: median (max) validation accuracy = 0.991 (0.995)\n",
      "** Saving params with joint score = 0.990 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 31 of 50 took       46.756s\n",
      "  training loss:               0.053691\n",
      "  holdout loss (batch):        0.030302\n",
      "  Subject 1: median (max) validation accuracy = 0.988 (0.994)\n",
      "  Subject 2: median (max) validation accuracy = 0.989 (0.995)\n",
      "  Subject 3: median (max) validation accuracy = 0.991 (0.995)\n",
      "  Subject 4: median (max) validation accuracy = 0.992 (0.996)\n",
      "  Subject 5: median (max) validation accuracy = 0.990 (0.996)\n",
      "  Subject 6: median (max) validation accuracy = 0.991 (0.996)\n",
      "  Subject 7: median (max) validation accuracy = 0.991 (0.995)\n",
      "  Subject 8: median (max) validation accuracy = 0.991 (0.995)\n",
      "** Saving params with joint score = 0.990 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 32 of 50 took       46.757s\n",
      "  training loss:               0.053004\n",
      "  holdout loss (batch):        0.029225\n",
      "  Subject 1: median (max) validation accuracy = 0.989 (0.995)\n",
      "  Subject 2: median (max) validation accuracy = 0.989 (0.995)\n",
      "  Subject 3: median (max) validation accuracy = 0.992 (0.995)\n",
      "  Subject 4: median (max) validation accuracy = 0.992 (0.996)\n",
      "  Subject 5: median (max) validation accuracy = 0.990 (0.996)\n",
      "  Subject 6: median (max) validation accuracy = 0.991 (0.996)\n",
      "  Subject 7: median (max) validation accuracy = 0.991 (0.995)\n",
      "  Subject 8: median (max) validation accuracy = 0.991 (0.995)\n",
      "** Saving params with joint score = 0.990 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 33 of 50 took       46.760s\n",
      "  training loss:               0.052203\n",
      "  holdout loss (batch):        0.028467\n",
      "  Subject 1: median (max) validation accuracy = 0.989 (0.995)\n",
      "  Subject 2: median (max) validation accuracy = 0.989 (0.995)\n",
      "  Subject 3: median (max) validation accuracy = 0.992 (0.995)\n",
      "  Subject 4: median (max) validation accuracy = 0.992 (0.996)\n",
      "  Subject 5: median (max) validation accuracy = 0.990 (0.996)\n",
      "  Subject 6: median (max) validation accuracy = 0.991 (0.996)\n",
      "  Subject 7: median (max) validation accuracy = 0.991 (0.995)\n",
      "  Subject 8: median (max) validation accuracy = 0.991 (0.995)\n",
      "** Saving params with joint score = 0.991 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 34 of 50 took       46.824s\n",
      "  training loss:               0.052372\n",
      "  holdout loss (batch):        0.031703\n",
      "  Subject 1: median (max) validation accuracy = 0.989 (0.995)\n",
      "  Subject 2: median (max) validation accuracy = 0.990 (0.995)\n",
      "  Subject 3: median (max) validation accuracy = 0.992 (0.996)\n",
      "  Subject 4: median (max) validation accuracy = 0.992 (0.996)\n",
      "  Subject 5: median (max) validation accuracy = 0.990 (0.996)\n",
      "  Subject 6: median (max) validation accuracy = 0.992 (0.996)\n",
      "  Subject 7: median (max) validation accuracy = 0.991 (0.995)\n",
      "  Subject 8: median (max) validation accuracy = 0.992 (0.996)\n",
      "** Saving params with joint score = 0.991 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 35 of 50 took       47.032s\n",
      "  training loss:               0.051110\n",
      "  holdout loss (batch):        0.027003\n",
      "  Subject 1: median (max) validation accuracy = 0.990 (0.995)\n",
      "  Subject 2: median (max) validation accuracy = 0.990 (0.996)\n",
      "  Subject 3: median (max) validation accuracy = 0.992 (0.995)\n",
      "  Subject 4: median (max) validation accuracy = 0.992 (0.996)\n",
      "  Subject 5: median (max) validation accuracy = 0.990 (0.996)\n",
      "  Subject 6: median (max) validation accuracy = 0.992 (0.996)\n",
      "  Subject 7: median (max) validation accuracy = 0.991 (0.995)\n",
      "  Subject 8: median (max) validation accuracy = 0.992 (0.995)\n",
      "** Saving params with joint score = 0.991 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 36 of 50 took       46.912s\n",
      "  training loss:               0.050964\n",
      "  holdout loss (batch):        0.034859\n",
      "  Subject 1: median (max) validation accuracy = 0.989 (0.995)\n",
      "  Subject 2: median (max) validation accuracy = 0.990 (0.995)\n",
      "  Subject 3: median (max) validation accuracy = 0.992 (0.995)\n",
      "  Subject 4: median (max) validation accuracy = 0.992 (0.996)\n",
      "  Subject 5: median (max) validation accuracy = 0.990 (0.996)\n",
      "  Subject 6: median (max) validation accuracy = 0.991 (0.996)\n",
      "  Subject 7: median (max) validation accuracy = 0.991 (0.995)\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "  Subject 8: median (max) validation accuracy = 0.991 (0.995)\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 37 of 50 took       47.029s\n",
      "  training loss:               0.049823\n",
      "  holdout loss (batch):        0.032246\n",
      "  Subject 1: median (max) validation accuracy = 0.990 (0.995)\n",
      "  Subject 2: median (max) validation accuracy = 0.990 (0.995)\n",
      "  Subject 3: median (max) validation accuracy = 0.992 (0.995)\n",
      "  Subject 4: median (max) validation accuracy = 0.992 (0.996)\n",
      "  Subject 5: median (max) validation accuracy = 0.990 (0.996)\n",
      "  Subject 6: median (max) validation accuracy = 0.992 (0.996)\n",
      "  Subject 7: median (max) validation accuracy = 0.991 (0.996)\n",
      "  Subject 8: median (max) validation accuracy = 0.992 (0.996)\n",
      "** Saving params with joint score = 0.991 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 38 of 50 took       46.969s\n",
      "  training loss:               0.049570\n",
      "  holdout loss (batch):        0.029662\n",
      "  Subject 1: median (max) validation accuracy = 0.990 (0.995)\n",
      "  Subject 2: median (max) validation accuracy = 0.991 (0.996)\n",
      "  Subject 3: median (max) validation accuracy = 0.993 (0.996)\n",
      "  Subject 4: median (max) validation accuracy = 0.993 (0.996)\n",
      "  Subject 5: median (max) validation accuracy = 0.991 (0.996)\n",
      "  Subject 6: median (max) validation accuracy = 0.992 (0.996)\n",
      "  Subject 7: median (max) validation accuracy = 0.992 (0.996)\n",
      "  Subject 8: median (max) validation accuracy = 0.992 (0.996)\n",
      "** Saving params with joint score = 0.992 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 39 of 50 took       46.966s\n",
      "  training loss:               0.048649\n",
      "  holdout loss (batch):        0.025661\n",
      "  Subject 1: median (max) validation accuracy = 0.990 (0.995)\n",
      "  Subject 2: median (max) validation accuracy = 0.991 (0.995)\n",
      "  Subject 3: median (max) validation accuracy = 0.992 (0.996)\n",
      "  Subject 4: median (max) validation accuracy = 0.993 (0.996)\n",
      "  Subject 5: median (max) validation accuracy = 0.991 (0.996)\n",
      "  Subject 6: median (max) validation accuracy = 0.992 (0.996)\n",
      "  Subject 7: median (max) validation accuracy = 0.992 (0.996)\n",
      "  Subject 8: median (max) validation accuracy = 0.992 (0.996)\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 40 of 50 took       46.999s\n",
      "  training loss:               0.048562\n",
      "  holdout loss (batch):        0.029442\n",
      "  Subject 1: median (max) validation accuracy = 0.990 (0.995)\n",
      "  Subject 2: median (max) validation accuracy = 0.991 (0.996)\n",
      "  Subject 3: median (max) validation accuracy = 0.992 (0.995)\n",
      "  Subject 4: median (max) validation accuracy = 0.993 (0.996)\n",
      "  Subject 5: median (max) validation accuracy = 0.991 (0.996)\n",
      "  Subject 6: median (max) validation accuracy = 0.992 (0.996)\n",
      "  Subject 7: median (max) validation accuracy = 0.992 (0.995)\n",
      "  Subject 8: median (max) validation accuracy = 0.992 (0.996)\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 41 of 50 took       46.994s\n",
      "  training loss:               0.048142\n",
      "  holdout loss (batch):        0.027300\n",
      "  Subject 1: median (max) validation accuracy = 0.991 (0.996)\n",
      "  Subject 2: median (max) validation accuracy = 0.991 (0.996)\n",
      "  Subject 3: median (max) validation accuracy = 0.993 (0.996)\n",
      "  Subject 4: median (max) validation accuracy = 0.993 (0.996)\n",
      "  Subject 5: median (max) validation accuracy = 0.991 (0.996)\n",
      "  Subject 6: median (max) validation accuracy = 0.992 (0.996)\n",
      "  Subject 7: median (max) validation accuracy = 0.992 (0.996)\n",
      "  Subject 8: median (max) validation accuracy = 0.992 (0.996)\n",
      "** Saving params with joint score = 0.992 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 42 of 50 took       46.927s\n",
      "  training loss:               0.047559\n",
      "  holdout loss (batch):        0.028713\n",
      "  Subject 1: median (max) validation accuracy = 0.990 (0.996)\n",
      "  Subject 2: median (max) validation accuracy = 0.991 (0.996)\n",
      "  Subject 3: median (max) validation accuracy = 0.993 (0.996)\n",
      "  Subject 4: median (max) validation accuracy = 0.993 (0.996)\n",
      "  Subject 5: median (max) validation accuracy = 0.991 (0.996)\n",
      "  Subject 6: median (max) validation accuracy = 0.992 (0.996)\n",
      "  Subject 7: median (max) validation accuracy = 0.992 (0.995)\n",
      "  Subject 8: median (max) validation accuracy = 0.992 (0.995)\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 43 of 50 took       46.896s\n",
      "  training loss:               0.047347\n",
      "  holdout loss (batch):        0.024499\n",
      "  Subject 1: median (max) validation accuracy = 0.990 (0.996)\n",
      "  Subject 2: median (max) validation accuracy = 0.991 (0.996)\n",
      "  Subject 3: median (max) validation accuracy = 0.993 (0.996)\n",
      "  Subject 4: median (max) validation accuracy = 0.993 (0.996)\n",
      "  Subject 5: median (max) validation accuracy = 0.991 (0.996)\n",
      "  Subject 6: median (max) validation accuracy = 0.992 (0.996)\n",
      "  Subject 7: median (max) validation accuracy = 0.992 (0.996)\n",
      "  Subject 8: median (max) validation accuracy = 0.992 (0.996)\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 44 of 50 took       47.446s\n",
      "  training loss:               0.047628\n",
      "  holdout loss (batch):        0.027270\n",
      "  Subject 1: median (max) validation accuracy = 0.990 (0.996)\n",
      "  Subject 2: median (max) validation accuracy = 0.991 (0.995)\n",
      "  Subject 3: median (max) validation accuracy = 0.992 (0.996)\n",
      "  Subject 4: median (max) validation accuracy = 0.993 (0.996)\n",
      "  Subject 5: median (max) validation accuracy = 0.991 (0.996)\n",
      "  Subject 6: median (max) validation accuracy = 0.992 (0.996)\n",
      "  Subject 7: median (max) validation accuracy = 0.992 (0.996)\n",
      "  Subject 8: median (max) validation accuracy = 0.992 (0.996)\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 45 of 50 took       47.319s\n",
      "  training loss:               0.047019\n",
      "  holdout loss (batch):        0.024062\n",
      "  Subject 1: median (max) validation accuracy = 0.991 (0.996)\n",
      "  Subject 2: median (max) validation accuracy = 0.991 (0.996)\n",
      "  Subject 3: median (max) validation accuracy = 0.993 (0.996)\n",
      "  Subject 4: median (max) validation accuracy = 0.993 (0.996)\n",
      "  Subject 5: median (max) validation accuracy = 0.992 (0.996)\n",
      "  Subject 6: median (max) validation accuracy = 0.992 (0.996)\n",
      "  Subject 7: median (max) validation accuracy = 0.992 (0.996)\n",
      "  Subject 8: median (max) validation accuracy = 0.992 (0.996)\n",
      "** Saving params with joint score = 0.992 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 46 of 50 took       47.325s\n",
      "  training loss:               0.046303\n",
      "  holdout loss (batch):        0.026502\n",
      "  Subject 1: median (max) validation accuracy = 0.991 (0.996)\n",
      "  Subject 2: median (max) validation accuracy = 0.991 (0.996)\n",
      "  Subject 3: median (max) validation accuracy = 0.993 (0.996)\n",
      "  Subject 4: median (max) validation accuracy = 0.993 (0.996)\n",
      "  Subject 5: median (max) validation accuracy = 0.991 (0.996)\n",
      "  Subject 6: median (max) validation accuracy = 0.992 (0.996)\n",
      "  Subject 7: median (max) validation accuracy = 0.992 (0.996)\n",
      "  Subject 8: median (max) validation accuracy = 0.992 (0.996)\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 47 of 50 took       47.357s\n",
      "  training loss:               0.046167\n",
      "  holdout loss (batch):        0.025458\n",
      "  Subject 1: median (max) validation accuracy = 0.991 (0.996)\n",
      "  Subject 2: median (max) validation accuracy = 0.991 (0.996)\n",
      "  Subject 3: median (max) validation accuracy = 0.993 (0.996)\n",
      "  Subject 4: median (max) validation accuracy = 0.993 (0.996)\n",
      "  Subject 5: median (max) validation accuracy = 0.992 (0.996)\n",
      "  Subject 6: median (max) validation accuracy = 0.992 (0.996)\n",
      "  Subject 7: median (max) validation accuracy = 0.992 (0.996)\n",
      "  Subject 8: median (max) validation accuracy = 0.992 (0.996)\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 48 of 50 took       47.376s\n",
      "  training loss:               0.045508\n",
      "  holdout loss (batch):        0.025504\n",
      "  Subject 1: median (max) validation accuracy = 0.991 (0.996)\n",
      "  Subject 2: median (max) validation accuracy = 0.991 (0.996)\n",
      "  Subject 3: median (max) validation accuracy = 0.993 (0.996)\n",
      "  Subject 4: median (max) validation accuracy = 0.993 (0.996)\n",
      "  Subject 5: median (max) validation accuracy = 0.992 (0.996)\n",
      "  Subject 6: median (max) validation accuracy = 0.992 (0.996)\n",
      "  Subject 7: median (max) validation accuracy = 0.992 (0.996)\n",
      "  Subject 8: median (max) validation accuracy = 0.992 (0.996)\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 49 of 50 took       47.231s\n",
      "  training loss:               0.045388\n",
      "  holdout loss (batch):        0.024866\n",
      "  Subject 1: median (max) validation accuracy = 0.991 (0.996)\n",
      "  Subject 2: median (max) validation accuracy = 0.992 (0.996)\n",
      "  Subject 3: median (max) validation accuracy = 0.993 (0.996)\n",
      "  Subject 4: median (max) validation accuracy = 0.993 (0.996)\n",
      "  Subject 5: median (max) validation accuracy = 0.992 (0.996)\n",
      "  Subject 6: median (max) validation accuracy = 0.993 (0.996)\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "  Subject 7: median (max) validation accuracy = 0.992 (0.996)\n",
      "  Subject 8: median (max) validation accuracy = 0.993 (0.996)\n",
      "** Saving params with joint score = 0.992 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 50 of 50 took       47.036s\n",
      "  training loss:               0.045505\n",
      "  holdout loss (batch):        0.027010\n",
      "  Subject 1: median (max) validation accuracy = 0.991 (0.996)\n",
      "  Subject 2: median (max) validation accuracy = 0.992 (0.996)\n",
      "  Subject 3: median (max) validation accuracy = 0.993 (0.996)\n",
      "  Subject 4: median (max) validation accuracy = 0.994 (0.997)\n",
      "  Subject 5: median (max) validation accuracy = 0.992 (0.996)\n",
      "  Subject 6: median (max) validation accuracy = 0.993 (0.997)\n",
      "  Subject 7: median (max) validation accuracy = 0.992 (0.996)\n",
      "  Subject 8: median (max) validation accuracy = 0.993 (0.996)\n",
      "** Saving params with joint score = 0.992 **\n",
      "\n",
      "***************************************\n",
      "best joint val cc = 0.993\n",
      "subject 1: val cc = 0.992\n",
      "subject 2: val cc = 0.992\n",
      "subject 3: val cc = 0.994\n",
      "subject 4: val cc = 0.994\n",
      "subject 5: val cc = 0.993\n",
      "subject 6: val cc = 0.993\n",
      "subject 7: val cc = 0.993\n",
      "subject 8: val cc = 0.993\n",
      "***************************************\n",
      "***************************************\n",
      "***                  V2       ***\n",
      "{1: 1415, 2: 1083, 3: 1160, 4: 883, 5: 1088, 6: 1191, 7: 958, 8: 1040}\n",
      "***************************************\n",
      "subject 1 masked 1415 of 4627\n",
      "subject 1 training/holdout 8100 900\n",
      "subject 2 masked 1083 of 3765\n",
      "subject 2 training/holdout 8100 900\n",
      "subject 3 masked 1160 of 3778\n",
      "subject 3 training/holdout 8100 900\n",
      "subject 4 masked 883 of 3027\n",
      "subject 4 training/holdout 8100 900\n",
      "subject 5 masked 1088 of 3685\n",
      "subject 5 training/holdout 8100 900\n",
      "subject 6 masked 1191 of 4024\n",
      "subject 6 training/holdout 8100 900\n",
      "subject 7 masked 958 of 3201\n",
      "subject 7 training/holdout 8100 900\n",
      "subject 8 masked 1040 of 3490\n",
      "subject 8 training/holdout 8100 900\n",
      "\\ : 100.0 %\n",
      "  Epoch 1 of 50 took       46.582s\n",
      "  training loss:               0.680129\n",
      "  holdout loss (batch):        0.538496\n",
      "  Subject 1: median (max) validation accuracy = 0.538 (0.914)\n",
      "  Subject 2: median (max) validation accuracy = 0.555 (0.920)\n",
      "  Subject 3: median (max) validation accuracy = 0.671 (0.915)\n",
      "  Subject 4: median (max) validation accuracy = 0.718 (0.905)\n",
      "  Subject 5: median (max) validation accuracy = 0.815 (0.945)\n",
      "  Subject 6: median (max) validation accuracy = 0.703 (0.924)\n",
      "  Subject 7: median (max) validation accuracy = 0.654 (0.885)\n",
      "  Subject 8: median (max) validation accuracy = 0.749 (0.911)\n",
      "** Saving params with joint score = 0.681 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 2 of 50 took       46.524s\n",
      "  training loss:               0.523577\n",
      "  holdout loss (batch):        0.432662\n",
      "  Subject 1: median (max) validation accuracy = 0.636 (0.947)\n",
      "  Subject 2: median (max) validation accuracy = 0.638 (0.950)\n",
      "  Subject 3: median (max) validation accuracy = 0.764 (0.942)\n",
      "  Subject 4: median (max) validation accuracy = 0.790 (0.930)\n",
      "  Subject 5: median (max) validation accuracy = 0.876 (0.959)\n",
      "  Subject 6: median (max) validation accuracy = 0.780 (0.939)\n",
      "  Subject 7: median (max) validation accuracy = 0.753 (0.933)\n",
      "  Subject 8: median (max) validation accuracy = 0.812 (0.942)\n",
      "** Saving params with joint score = 0.766 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 3 of 50 took       46.502s\n",
      "  training loss:               0.438252\n",
      "  holdout loss (batch):        0.369675\n",
      "  Subject 1: median (max) validation accuracy = 0.709 (0.957)\n",
      "  Subject 2: median (max) validation accuracy = 0.711 (0.952)\n",
      "  Subject 3: median (max) validation accuracy = 0.816 (0.948)\n",
      "  Subject 4: median (max) validation accuracy = 0.839 (0.940)\n",
      "  Subject 5: median (max) validation accuracy = 0.904 (0.964)\n",
      "  Subject 6: median (max) validation accuracy = 0.831 (0.948)\n",
      "  Subject 7: median (max) validation accuracy = 0.812 (0.944)\n",
      "  Subject 8: median (max) validation accuracy = 0.853 (0.956)\n",
      "** Saving params with joint score = 0.818 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 4 of 50 took       46.517s\n",
      "  training loss:               0.373678\n",
      "  holdout loss (batch):        0.301995\n",
      "  Subject 1: median (max) validation accuracy = 0.755 (0.962)\n",
      "  Subject 2: median (max) validation accuracy = 0.763 (0.959)\n",
      "  Subject 3: median (max) validation accuracy = 0.852 (0.954)\n",
      "  Subject 4: median (max) validation accuracy = 0.873 (0.950)\n",
      "  Subject 5: median (max) validation accuracy = 0.920 (0.970)\n",
      "  Subject 6: median (max) validation accuracy = 0.860 (0.960)\n",
      "  Subject 7: median (max) validation accuracy = 0.854 (0.955)\n",
      "  Subject 8: median (max) validation accuracy = 0.877 (0.961)\n",
      "** Saving params with joint score = 0.851 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 5 of 50 took       46.495s\n",
      "  training loss:               0.330698\n",
      "  holdout loss (batch):        0.268955\n",
      "  Subject 1: median (max) validation accuracy = 0.788 (0.965)\n",
      "  Subject 2: median (max) validation accuracy = 0.793 (0.962)\n",
      "  Subject 3: median (max) validation accuracy = 0.871 (0.958)\n",
      "  Subject 4: median (max) validation accuracy = 0.892 (0.955)\n",
      "  Subject 5: median (max) validation accuracy = 0.930 (0.973)\n",
      "  Subject 6: median (max) validation accuracy = 0.880 (0.963)\n",
      "  Subject 7: median (max) validation accuracy = 0.877 (0.961)\n",
      "  Subject 8: median (max) validation accuracy = 0.893 (0.966)\n",
      "** Saving params with joint score = 0.872 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 6 of 50 took       46.411s\n",
      "  training loss:               0.295089\n",
      "  holdout loss (batch):        0.236163\n",
      "  Subject 1: median (max) validation accuracy = 0.820 (0.967)\n",
      "  Subject 2: median (max) validation accuracy = 0.824 (0.969)\n",
      "  Subject 3: median (max) validation accuracy = 0.890 (0.962)\n",
      "  Subject 4: median (max) validation accuracy = 0.909 (0.963)\n",
      "  Subject 5: median (max) validation accuracy = 0.939 (0.977)\n",
      "  Subject 6: median (max) validation accuracy = 0.899 (0.969)\n",
      "  Subject 7: median (max) validation accuracy = 0.900 (0.965)\n",
      "  Subject 8: median (max) validation accuracy = 0.910 (0.970)\n",
      "** Saving params with joint score = 0.892 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 7 of 50 took       46.428s\n",
      "  training loss:               0.264486\n",
      "  holdout loss (batch):        0.201747\n",
      "  Subject 1: median (max) validation accuracy = 0.848 (0.970)\n",
      "  Subject 2: median (max) validation accuracy = 0.851 (0.972)\n",
      "  Subject 3: median (max) validation accuracy = 0.907 (0.966)\n",
      "  Subject 4: median (max) validation accuracy = 0.922 (0.968)\n",
      "  Subject 5: median (max) validation accuracy = 0.947 (0.979)\n",
      "  Subject 6: median (max) validation accuracy = 0.914 (0.973)\n",
      "  Subject 7: median (max) validation accuracy = 0.916 (0.969)\n",
      "  Subject 8: median (max) validation accuracy = 0.923 (0.973)\n",
      "** Saving params with joint score = 0.909 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 8 of 50 took       46.401s\n",
      "  training loss:               0.238332\n",
      "  holdout loss (batch):        0.185831\n",
      "  Subject 1: median (max) validation accuracy = 0.868 (0.972)\n",
      "  Subject 2: median (max) validation accuracy = 0.868 (0.975)\n",
      "  Subject 3: median (max) validation accuracy = 0.919 (0.969)\n",
      "  Subject 4: median (max) validation accuracy = 0.932 (0.972)\n",
      "  Subject 5: median (max) validation accuracy = 0.952 (0.981)\n",
      "  Subject 6: median (max) validation accuracy = 0.924 (0.977)\n",
      "  Subject 7: median (max) validation accuracy = 0.927 (0.972)\n",
      "  Subject 8: median (max) validation accuracy = 0.932 (0.977)\n",
      "** Saving params with joint score = 0.919 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 9 of 50 took       46.519s\n",
      "  training loss:               0.215452\n",
      "  holdout loss (batch):        0.161129\n",
      "  Subject 1: median (max) validation accuracy = 0.888 (0.978)\n",
      "  Subject 2: median (max) validation accuracy = 0.887 (0.979)\n",
      "  Subject 3: median (max) validation accuracy = 0.931 (0.973)\n",
      "  Subject 4: median (max) validation accuracy = 0.942 (0.975)\n",
      "  Subject 5: median (max) validation accuracy = 0.957 (0.984)\n",
      "  Subject 6: median (max) validation accuracy = 0.934 (0.981)\n",
      "  Subject 7: median (max) validation accuracy = 0.938 (0.976)\n",
      "  Subject 8: median (max) validation accuracy = 0.942 (0.978)\n",
      "** Saving params with joint score = 0.931 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 10 of 50 took       46.528s\n",
      "  training loss:               0.194240\n",
      "  holdout loss (batch):        0.147806\n",
      "  Subject 1: median (max) validation accuracy = 0.904 (0.979)\n",
      "  Subject 2: median (max) validation accuracy = 0.902 (0.981)\n",
      "  Subject 3: median (max) validation accuracy = 0.939 (0.975)\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "  Subject 4: median (max) validation accuracy = 0.950 (0.978)\n",
      "  Subject 5: median (max) validation accuracy = 0.960 (0.985)\n",
      "  Subject 6: median (max) validation accuracy = 0.941 (0.983)\n",
      "  Subject 7: median (max) validation accuracy = 0.945 (0.978)\n",
      "  Subject 8: median (max) validation accuracy = 0.948 (0.981)\n",
      "** Saving params with joint score = 0.939 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 11 of 50 took       46.509s\n",
      "  training loss:               0.175749\n",
      "  holdout loss (batch):        0.132151\n",
      "  Subject 1: median (max) validation accuracy = 0.920 (0.981)\n",
      "  Subject 2: median (max) validation accuracy = 0.918 (0.984)\n",
      "  Subject 3: median (max) validation accuracy = 0.949 (0.979)\n",
      "  Subject 4: median (max) validation accuracy = 0.956 (0.980)\n",
      "  Subject 5: median (max) validation accuracy = 0.965 (0.986)\n",
      "  Subject 6: median (max) validation accuracy = 0.949 (0.985)\n",
      "  Subject 7: median (max) validation accuracy = 0.953 (0.982)\n",
      "  Subject 8: median (max) validation accuracy = 0.956 (0.984)\n",
      "** Saving params with joint score = 0.947 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 12 of 50 took       46.526s\n",
      "  training loss:               0.157946\n",
      "  holdout loss (batch):        0.105228\n",
      "  Subject 1: median (max) validation accuracy = 0.934 (0.984)\n",
      "  Subject 2: median (max) validation accuracy = 0.931 (0.986)\n",
      "  Subject 3: median (max) validation accuracy = 0.958 (0.982)\n",
      "  Subject 4: median (max) validation accuracy = 0.963 (0.983)\n",
      "  Subject 5: median (max) validation accuracy = 0.968 (0.989)\n",
      "  Subject 6: median (max) validation accuracy = 0.957 (0.987)\n",
      "  Subject 7: median (max) validation accuracy = 0.961 (0.985)\n",
      "  Subject 8: median (max) validation accuracy = 0.963 (0.984)\n",
      "** Saving params with joint score = 0.956 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 13 of 50 took       46.646s\n",
      "  training loss:               0.142741\n",
      "  holdout loss (batch):        0.100670\n",
      "  Subject 1: median (max) validation accuracy = 0.944 (0.986)\n",
      "  Subject 2: median (max) validation accuracy = 0.941 (0.986)\n",
      "  Subject 3: median (max) validation accuracy = 0.964 (0.983)\n",
      "  Subject 4: median (max) validation accuracy = 0.968 (0.984)\n",
      "  Subject 5: median (max) validation accuracy = 0.970 (0.989)\n",
      "  Subject 6: median (max) validation accuracy = 0.962 (0.988)\n",
      "  Subject 7: median (max) validation accuracy = 0.965 (0.986)\n",
      "  Subject 8: median (max) validation accuracy = 0.967 (0.986)\n",
      "** Saving params with joint score = 0.961 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 14 of 50 took       46.608s\n",
      "  training loss:               0.128783\n",
      "  holdout loss (batch):        0.083847\n",
      "  Subject 1: median (max) validation accuracy = 0.952 (0.987)\n",
      "  Subject 2: median (max) validation accuracy = 0.950 (0.988)\n",
      "  Subject 3: median (max) validation accuracy = 0.969 (0.986)\n",
      "  Subject 4: median (max) validation accuracy = 0.972 (0.986)\n",
      "  Subject 5: median (max) validation accuracy = 0.973 (0.990)\n",
      "  Subject 6: median (max) validation accuracy = 0.967 (0.989)\n",
      "  Subject 7: median (max) validation accuracy = 0.970 (0.988)\n",
      "  Subject 8: median (max) validation accuracy = 0.972 (0.987)\n",
      "** Saving params with joint score = 0.966 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 15 of 50 took       46.629s\n",
      "  training loss:               0.118193\n",
      "  holdout loss (batch):        0.073797\n",
      "  Subject 1: median (max) validation accuracy = 0.959 (0.988)\n",
      "  Subject 2: median (max) validation accuracy = 0.957 (0.988)\n",
      "  Subject 3: median (max) validation accuracy = 0.974 (0.987)\n",
      "  Subject 4: median (max) validation accuracy = 0.975 (0.987)\n",
      "  Subject 5: median (max) validation accuracy = 0.975 (0.991)\n",
      "  Subject 6: median (max) validation accuracy = 0.971 (0.989)\n",
      "  Subject 7: median (max) validation accuracy = 0.974 (0.989)\n",
      "  Subject 8: median (max) validation accuracy = 0.975 (0.989)\n",
      "** Saving params with joint score = 0.970 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 16 of 50 took       46.608s\n",
      "  training loss:               0.109264\n",
      "  holdout loss (batch):        0.072003\n",
      "  Subject 1: median (max) validation accuracy = 0.963 (0.989)\n",
      "  Subject 2: median (max) validation accuracy = 0.961 (0.989)\n",
      "  Subject 3: median (max) validation accuracy = 0.975 (0.987)\n",
      "  Subject 4: median (max) validation accuracy = 0.977 (0.989)\n",
      "  Subject 5: median (max) validation accuracy = 0.976 (0.991)\n",
      "  Subject 6: median (max) validation accuracy = 0.973 (0.990)\n",
      "  Subject 7: median (max) validation accuracy = 0.976 (0.989)\n",
      "  Subject 8: median (max) validation accuracy = 0.977 (0.989)\n",
      "** Saving params with joint score = 0.973 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 17 of 50 took       46.612s\n",
      "  training loss:               0.101458\n",
      "  holdout loss (batch):        0.069215\n",
      "  Subject 1: median (max) validation accuracy = 0.968 (0.989)\n",
      "  Subject 2: median (max) validation accuracy = 0.966 (0.989)\n",
      "  Subject 3: median (max) validation accuracy = 0.978 (0.989)\n",
      "  Subject 4: median (max) validation accuracy = 0.980 (0.989)\n",
      "  Subject 5: median (max) validation accuracy = 0.977 (0.991)\n",
      "  Subject 6: median (max) validation accuracy = 0.976 (0.991)\n",
      "  Subject 7: median (max) validation accuracy = 0.978 (0.990)\n",
      "  Subject 8: median (max) validation accuracy = 0.979 (0.991)\n",
      "** Saving params with joint score = 0.975 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 18 of 50 took       46.621s\n",
      "  training loss:               0.096071\n",
      "  holdout loss (batch):        0.059249\n",
      "  Subject 1: median (max) validation accuracy = 0.971 (0.991)\n",
      "  Subject 2: median (max) validation accuracy = 0.969 (0.989)\n",
      "  Subject 3: median (max) validation accuracy = 0.981 (0.990)\n",
      "  Subject 4: median (max) validation accuracy = 0.982 (0.991)\n",
      "  Subject 5: median (max) validation accuracy = 0.979 (0.992)\n",
      "  Subject 6: median (max) validation accuracy = 0.979 (0.991)\n",
      "  Subject 7: median (max) validation accuracy = 0.981 (0.991)\n",
      "  Subject 8: median (max) validation accuracy = 0.982 (0.991)\n",
      "** Saving params with joint score = 0.978 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 19 of 50 took       46.608s\n",
      "  training loss:               0.091312\n",
      "  holdout loss (batch):        0.058511\n",
      "  Subject 1: median (max) validation accuracy = 0.973 (0.991)\n",
      "  Subject 2: median (max) validation accuracy = 0.971 (0.990)\n",
      "  Subject 3: median (max) validation accuracy = 0.981 (0.990)\n",
      "  Subject 4: median (max) validation accuracy = 0.982 (0.991)\n",
      "  Subject 5: median (max) validation accuracy = 0.980 (0.992)\n",
      "  Subject 6: median (max) validation accuracy = 0.980 (0.992)\n",
      "  Subject 7: median (max) validation accuracy = 0.981 (0.991)\n",
      "  Subject 8: median (max) validation accuracy = 0.982 (0.992)\n",
      "** Saving params with joint score = 0.979 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 20 of 50 took       46.559s\n",
      "  training loss:               0.087251\n",
      "  holdout loss (batch):        0.052604\n",
      "  Subject 1: median (max) validation accuracy = 0.976 (0.992)\n",
      "  Subject 2: median (max) validation accuracy = 0.974 (0.990)\n",
      "  Subject 3: median (max) validation accuracy = 0.983 (0.991)\n",
      "  Subject 4: median (max) validation accuracy = 0.984 (0.992)\n",
      "  Subject 5: median (max) validation accuracy = 0.982 (0.992)\n",
      "  Subject 6: median (max) validation accuracy = 0.981 (0.992)\n",
      "  Subject 7: median (max) validation accuracy = 0.983 (0.992)\n",
      "  Subject 8: median (max) validation accuracy = 0.984 (0.992)\n",
      "** Saving params with joint score = 0.981 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 21 of 50 took       46.470s\n",
      "  training loss:               0.084077\n",
      "  holdout loss (batch):        0.052067\n",
      "  Subject 1: median (max) validation accuracy = 0.977 (0.992)\n",
      "  Subject 2: median (max) validation accuracy = 0.975 (0.991)\n",
      "  Subject 3: median (max) validation accuracy = 0.984 (0.992)\n",
      "  Subject 4: median (max) validation accuracy = 0.984 (0.992)\n",
      "  Subject 5: median (max) validation accuracy = 0.982 (0.993)\n",
      "  Subject 6: median (max) validation accuracy = 0.982 (0.992)\n",
      "  Subject 7: median (max) validation accuracy = 0.984 (0.992)\n",
      "  Subject 8: median (max) validation accuracy = 0.984 (0.992)\n",
      "** Saving params with joint score = 0.982 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 22 of 50 took       46.500s\n",
      "  training loss:               0.082133\n",
      "  holdout loss (batch):        0.049015\n",
      "  Subject 1: median (max) validation accuracy = 0.979 (0.992)\n",
      "  Subject 2: median (max) validation accuracy = 0.977 (0.991)\n",
      "  Subject 3: median (max) validation accuracy = 0.985 (0.992)\n",
      "  Subject 4: median (max) validation accuracy = 0.985 (0.993)\n",
      "  Subject 5: median (max) validation accuracy = 0.983 (0.993)\n",
      "  Subject 6: median (max) validation accuracy = 0.983 (0.993)\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "  Subject 7: median (max) validation accuracy = 0.984 (0.992)\n",
      "  Subject 8: median (max) validation accuracy = 0.985 (0.993)\n",
      "** Saving params with joint score = 0.983 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 23 of 50 took       46.487s\n",
      "  training loss:               0.079093\n",
      "  holdout loss (batch):        0.043630\n",
      "  Subject 1: median (max) validation accuracy = 0.980 (0.993)\n",
      "  Subject 2: median (max) validation accuracy = 0.978 (0.991)\n",
      "  Subject 3: median (max) validation accuracy = 0.986 (0.992)\n",
      "  Subject 4: median (max) validation accuracy = 0.986 (0.993)\n",
      "  Subject 5: median (max) validation accuracy = 0.984 (0.993)\n",
      "  Subject 6: median (max) validation accuracy = 0.984 (0.993)\n",
      "  Subject 7: median (max) validation accuracy = 0.985 (0.993)\n",
      "  Subject 8: median (max) validation accuracy = 0.986 (0.993)\n",
      "** Saving params with joint score = 0.984 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 24 of 50 took       46.492s\n",
      "  training loss:               0.077205\n",
      "  holdout loss (batch):        0.047638\n",
      "  Subject 1: median (max) validation accuracy = 0.980 (0.993)\n",
      "  Subject 2: median (max) validation accuracy = 0.978 (0.991)\n",
      "  Subject 3: median (max) validation accuracy = 0.985 (0.992)\n",
      "  Subject 4: median (max) validation accuracy = 0.986 (0.992)\n",
      "  Subject 5: median (max) validation accuracy = 0.984 (0.993)\n",
      "  Subject 6: median (max) validation accuracy = 0.984 (0.993)\n",
      "  Subject 7: median (max) validation accuracy = 0.985 (0.993)\n",
      "  Subject 8: median (max) validation accuracy = 0.985 (0.993)\n",
      "** Saving params with joint score = 0.984 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 25 of 50 took       46.495s\n",
      "  training loss:               0.074268\n",
      "  holdout loss (batch):        0.043160\n",
      "  Subject 1: median (max) validation accuracy = 0.982 (0.993)\n",
      "  Subject 2: median (max) validation accuracy = 0.979 (0.991)\n",
      "  Subject 3: median (max) validation accuracy = 0.986 (0.993)\n",
      "  Subject 4: median (max) validation accuracy = 0.986 (0.993)\n",
      "  Subject 5: median (max) validation accuracy = 0.984 (0.993)\n",
      "  Subject 6: median (max) validation accuracy = 0.985 (0.993)\n",
      "  Subject 7: median (max) validation accuracy = 0.986 (0.993)\n",
      "  Subject 8: median (max) validation accuracy = 0.987 (0.993)\n",
      "** Saving params with joint score = 0.984 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 26 of 50 took       46.497s\n",
      "  training loss:               0.073678\n",
      "  holdout loss (batch):        0.042703\n",
      "  Subject 1: median (max) validation accuracy = 0.982 (0.993)\n",
      "  Subject 2: median (max) validation accuracy = 0.981 (0.992)\n",
      "  Subject 3: median (max) validation accuracy = 0.987 (0.993)\n",
      "  Subject 4: median (max) validation accuracy = 0.987 (0.993)\n",
      "  Subject 5: median (max) validation accuracy = 0.985 (0.993)\n",
      "  Subject 6: median (max) validation accuracy = 0.985 (0.993)\n",
      "  Subject 7: median (max) validation accuracy = 0.986 (0.993)\n",
      "  Subject 8: median (max) validation accuracy = 0.987 (0.993)\n",
      "** Saving params with joint score = 0.985 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 27 of 50 took       46.460s\n",
      "  training loss:               0.071844\n",
      "  holdout loss (batch):        0.041188\n",
      "  Subject 1: median (max) validation accuracy = 0.983 (0.994)\n",
      "  Subject 2: median (max) validation accuracy = 0.981 (0.992)\n",
      "  Subject 3: median (max) validation accuracy = 0.987 (0.993)\n",
      "  Subject 4: median (max) validation accuracy = 0.987 (0.994)\n",
      "  Subject 5: median (max) validation accuracy = 0.986 (0.994)\n",
      "  Subject 6: median (max) validation accuracy = 0.986 (0.994)\n",
      "  Subject 7: median (max) validation accuracy = 0.987 (0.994)\n",
      "  Subject 8: median (max) validation accuracy = 0.987 (0.994)\n",
      "** Saving params with joint score = 0.985 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 28 of 50 took       46.484s\n",
      "  training loss:               0.070366\n",
      "  holdout loss (batch):        0.042423\n",
      "  Subject 1: median (max) validation accuracy = 0.983 (0.993)\n",
      "  Subject 2: median (max) validation accuracy = 0.982 (0.992)\n",
      "  Subject 3: median (max) validation accuracy = 0.987 (0.993)\n",
      "  Subject 4: median (max) validation accuracy = 0.987 (0.994)\n",
      "  Subject 5: median (max) validation accuracy = 0.985 (0.993)\n",
      "  Subject 6: median (max) validation accuracy = 0.986 (0.993)\n",
      "  Subject 7: median (max) validation accuracy = 0.987 (0.993)\n",
      "  Subject 8: median (max) validation accuracy = 0.987 (0.993)\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 29 of 50 took       46.405s\n",
      "  training loss:               0.069169\n",
      "  holdout loss (batch):        0.041295\n",
      "  Subject 1: median (max) validation accuracy = 0.984 (0.993)\n",
      "  Subject 2: median (max) validation accuracy = 0.982 (0.992)\n",
      "  Subject 3: median (max) validation accuracy = 0.987 (0.993)\n",
      "  Subject 4: median (max) validation accuracy = 0.988 (0.994)\n",
      "  Subject 5: median (max) validation accuracy = 0.986 (0.993)\n",
      "  Subject 6: median (max) validation accuracy = 0.986 (0.994)\n",
      "  Subject 7: median (max) validation accuracy = 0.987 (0.994)\n",
      "  Subject 8: median (max) validation accuracy = 0.987 (0.994)\n",
      "** Saving params with joint score = 0.986 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 30 of 50 took       46.394s\n",
      "  training loss:               0.068518\n",
      "  holdout loss (batch):        0.036743\n",
      "  Subject 1: median (max) validation accuracy = 0.985 (0.994)\n",
      "  Subject 2: median (max) validation accuracy = 0.983 (0.993)\n",
      "  Subject 3: median (max) validation accuracy = 0.988 (0.994)\n",
      "  Subject 4: median (max) validation accuracy = 0.988 (0.994)\n",
      "  Subject 5: median (max) validation accuracy = 0.986 (0.994)\n",
      "  Subject 6: median (max) validation accuracy = 0.987 (0.994)\n",
      "  Subject 7: median (max) validation accuracy = 0.988 (0.994)\n",
      "  Subject 8: median (max) validation accuracy = 0.988 (0.994)\n",
      "** Saving params with joint score = 0.987 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 31 of 50 took       46.410s\n",
      "  training loss:               0.067183\n",
      "  holdout loss (batch):        0.036310\n",
      "  Subject 1: median (max) validation accuracy = 0.985 (0.994)\n",
      "  Subject 2: median (max) validation accuracy = 0.983 (0.993)\n",
      "  Subject 3: median (max) validation accuracy = 0.988 (0.994)\n",
      "  Subject 4: median (max) validation accuracy = 0.989 (0.994)\n",
      "  Subject 5: median (max) validation accuracy = 0.987 (0.994)\n",
      "  Subject 6: median (max) validation accuracy = 0.987 (0.994)\n",
      "  Subject 7: median (max) validation accuracy = 0.988 (0.994)\n",
      "  Subject 8: median (max) validation accuracy = 0.988 (0.994)\n",
      "** Saving params with joint score = 0.987 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 32 of 50 took       46.445s\n",
      "  training loss:               0.065735\n",
      "  holdout loss (batch):        0.035027\n",
      "  Subject 1: median (max) validation accuracy = 0.985 (0.995)\n",
      "  Subject 2: median (max) validation accuracy = 0.984 (0.993)\n",
      "  Subject 3: median (max) validation accuracy = 0.989 (0.994)\n",
      "  Subject 4: median (max) validation accuracy = 0.988 (0.994)\n",
      "  Subject 5: median (max) validation accuracy = 0.987 (0.994)\n",
      "  Subject 6: median (max) validation accuracy = 0.987 (0.994)\n",
      "  Subject 7: median (max) validation accuracy = 0.988 (0.994)\n",
      "  Subject 8: median (max) validation accuracy = 0.988 (0.994)\n",
      "** Saving params with joint score = 0.987 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 33 of 50 took       46.382s\n",
      "  training loss:               0.064816\n",
      "  holdout loss (batch):        0.033517\n",
      "  Subject 1: median (max) validation accuracy = 0.986 (0.995)\n",
      "  Subject 2: median (max) validation accuracy = 0.985 (0.993)\n",
      "  Subject 3: median (max) validation accuracy = 0.989 (0.994)\n",
      "  Subject 4: median (max) validation accuracy = 0.989 (0.994)\n",
      "  Subject 5: median (max) validation accuracy = 0.987 (0.994)\n",
      "  Subject 6: median (max) validation accuracy = 0.988 (0.994)\n",
      "  Subject 7: median (max) validation accuracy = 0.989 (0.994)\n",
      "  Subject 8: median (max) validation accuracy = 0.989 (0.995)\n",
      "** Saving params with joint score = 0.988 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 34 of 50 took       46.382s\n",
      "  training loss:               0.064499\n",
      "  holdout loss (batch):        0.032753\n",
      "  Subject 1: median (max) validation accuracy = 0.986 (0.995)\n",
      "  Subject 2: median (max) validation accuracy = 0.984 (0.993)\n",
      "  Subject 3: median (max) validation accuracy = 0.989 (0.994)\n",
      "  Subject 4: median (max) validation accuracy = 0.989 (0.994)\n",
      "  Subject 5: median (max) validation accuracy = 0.988 (0.994)\n",
      "  Subject 6: median (max) validation accuracy = 0.988 (0.995)\n",
      "  Subject 7: median (max) validation accuracy = 0.988 (0.994)\n",
      "  Subject 8: median (max) validation accuracy = 0.989 (0.995)\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 35 of 50 took       46.400s\n",
      "  training loss:               0.063641\n",
      "  holdout loss (batch):        0.036702\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "  Subject 1: median (max) validation accuracy = 0.986 (0.994)\n",
      "  Subject 2: median (max) validation accuracy = 0.985 (0.993)\n",
      "  Subject 3: median (max) validation accuracy = 0.989 (0.994)\n",
      "  Subject 4: median (max) validation accuracy = 0.989 (0.994)\n",
      "  Subject 5: median (max) validation accuracy = 0.987 (0.994)\n",
      "  Subject 6: median (max) validation accuracy = 0.988 (0.994)\n",
      "  Subject 7: median (max) validation accuracy = 0.988 (0.994)\n",
      "  Subject 8: median (max) validation accuracy = 0.989 (0.994)\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 36 of 50 took       46.408s\n",
      "  training loss:               0.063342\n",
      "  holdout loss (batch):        0.032436\n",
      "  Subject 1: median (max) validation accuracy = 0.987 (0.995)\n",
      "  Subject 2: median (max) validation accuracy = 0.985 (0.994)\n",
      "  Subject 3: median (max) validation accuracy = 0.990 (0.994)\n",
      "  Subject 4: median (max) validation accuracy = 0.989 (0.995)\n",
      "  Subject 5: median (max) validation accuracy = 0.988 (0.994)\n",
      "  Subject 6: median (max) validation accuracy = 0.988 (0.994)\n",
      "  Subject 7: median (max) validation accuracy = 0.989 (0.995)\n",
      "  Subject 8: median (max) validation accuracy = 0.989 (0.995)\n",
      "** Saving params with joint score = 0.988 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 37 of 50 took       46.358s\n",
      "  training loss:               0.062535\n",
      "  holdout loss (batch):        0.033378\n",
      "  Subject 1: median (max) validation accuracy = 0.987 (0.995)\n",
      "  Subject 2: median (max) validation accuracy = 0.986 (0.994)\n",
      "  Subject 3: median (max) validation accuracy = 0.990 (0.994)\n",
      "  Subject 4: median (max) validation accuracy = 0.989 (0.994)\n",
      "  Subject 5: median (max) validation accuracy = 0.988 (0.994)\n",
      "  Subject 6: median (max) validation accuracy = 0.988 (0.995)\n",
      "  Subject 7: median (max) validation accuracy = 0.989 (0.995)\n",
      "  Subject 8: median (max) validation accuracy = 0.989 (0.995)\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 38 of 50 took       46.363s\n",
      "  training loss:               0.061607\n",
      "  holdout loss (batch):        0.032778\n",
      "  Subject 1: median (max) validation accuracy = 0.987 (0.995)\n",
      "  Subject 2: median (max) validation accuracy = 0.986 (0.994)\n",
      "  Subject 3: median (max) validation accuracy = 0.990 (0.994)\n",
      "  Subject 4: median (max) validation accuracy = 0.990 (0.995)\n",
      "  Subject 5: median (max) validation accuracy = 0.988 (0.994)\n",
      "  Subject 6: median (max) validation accuracy = 0.988 (0.995)\n",
      "  Subject 7: median (max) validation accuracy = 0.989 (0.995)\n",
      "  Subject 8: median (max) validation accuracy = 0.989 (0.995)\n",
      "** Saving params with joint score = 0.988 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 39 of 50 took       46.457s\n",
      "  training loss:               0.061362\n",
      "  holdout loss (batch):        0.041259\n",
      "  Subject 1: median (max) validation accuracy = 0.987 (0.995)\n",
      "  Subject 2: median (max) validation accuracy = 0.986 (0.994)\n",
      "  Subject 3: median (max) validation accuracy = 0.990 (0.995)\n",
      "  Subject 4: median (max) validation accuracy = 0.990 (0.995)\n",
      "  Subject 5: median (max) validation accuracy = 0.988 (0.994)\n",
      "  Subject 6: median (max) validation accuracy = 0.988 (0.994)\n",
      "  Subject 7: median (max) validation accuracy = 0.989 (0.995)\n",
      "  Subject 8: median (max) validation accuracy = 0.989 (0.995)\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 40 of 50 took       46.530s\n",
      "  training loss:               0.060568\n",
      "  holdout loss (batch):        0.030388\n",
      "  Subject 1: median (max) validation accuracy = 0.988 (0.995)\n",
      "  Subject 2: median (max) validation accuracy = 0.986 (0.994)\n",
      "  Subject 3: median (max) validation accuracy = 0.990 (0.995)\n",
      "  Subject 4: median (max) validation accuracy = 0.990 (0.995)\n",
      "  Subject 5: median (max) validation accuracy = 0.989 (0.995)\n",
      "  Subject 6: median (max) validation accuracy = 0.989 (0.995)\n",
      "  Subject 7: median (max) validation accuracy = 0.989 (0.995)\n",
      "  Subject 8: median (max) validation accuracy = 0.990 (0.995)\n",
      "** Saving params with joint score = 0.989 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 41 of 50 took       46.445s\n",
      "  training loss:               0.060632\n",
      "  holdout loss (batch):        0.030791\n",
      "  Subject 1: median (max) validation accuracy = 0.988 (0.995)\n",
      "  Subject 2: median (max) validation accuracy = 0.986 (0.994)\n",
      "  Subject 3: median (max) validation accuracy = 0.990 (0.995)\n",
      "  Subject 4: median (max) validation accuracy = 0.990 (0.995)\n",
      "  Subject 5: median (max) validation accuracy = 0.989 (0.995)\n",
      "  Subject 6: median (max) validation accuracy = 0.989 (0.995)\n",
      "  Subject 7: median (max) validation accuracy = 0.989 (0.995)\n",
      "  Subject 8: median (max) validation accuracy = 0.990 (0.995)\n",
      "** Saving params with joint score = 0.989 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 42 of 50 took       46.414s\n",
      "  training loss:               0.059801\n",
      "  holdout loss (batch):        0.030095\n",
      "  Subject 1: median (max) validation accuracy = 0.988 (0.996)\n",
      "  Subject 2: median (max) validation accuracy = 0.987 (0.994)\n",
      "  Subject 3: median (max) validation accuracy = 0.990 (0.995)\n",
      "  Subject 4: median (max) validation accuracy = 0.990 (0.995)\n",
      "  Subject 5: median (max) validation accuracy = 0.989 (0.995)\n",
      "  Subject 6: median (max) validation accuracy = 0.989 (0.995)\n",
      "  Subject 7: median (max) validation accuracy = 0.990 (0.995)\n",
      "  Subject 8: median (max) validation accuracy = 0.990 (0.995)\n",
      "** Saving params with joint score = 0.989 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 43 of 50 took       46.383s\n",
      "  training loss:               0.059119\n",
      "  holdout loss (batch):        0.030221\n",
      "  Subject 1: median (max) validation accuracy = 0.988 (0.995)\n",
      "  Subject 2: median (max) validation accuracy = 0.987 (0.994)\n",
      "  Subject 3: median (max) validation accuracy = 0.990 (0.995)\n",
      "  Subject 4: median (max) validation accuracy = 0.990 (0.995)\n",
      "  Subject 5: median (max) validation accuracy = 0.989 (0.995)\n",
      "  Subject 6: median (max) validation accuracy = 0.989 (0.995)\n",
      "  Subject 7: median (max) validation accuracy = 0.990 (0.995)\n",
      "  Subject 8: median (max) validation accuracy = 0.990 (0.995)\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 44 of 50 took       46.375s\n",
      "  training loss:               0.058798\n",
      "  holdout loss (batch):        0.029776\n",
      "  Subject 1: median (max) validation accuracy = 0.988 (0.995)\n",
      "  Subject 2: median (max) validation accuracy = 0.987 (0.994)\n",
      "  Subject 3: median (max) validation accuracy = 0.991 (0.995)\n",
      "  Subject 4: median (max) validation accuracy = 0.991 (0.995)\n",
      "  Subject 5: median (max) validation accuracy = 0.989 (0.995)\n",
      "  Subject 6: median (max) validation accuracy = 0.989 (0.995)\n",
      "  Subject 7: median (max) validation accuracy = 0.990 (0.995)\n",
      "  Subject 8: median (max) validation accuracy = 0.990 (0.995)\n",
      "** Saving params with joint score = 0.989 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 45 of 50 took       46.410s\n",
      "  training loss:               0.058551\n",
      "  holdout loss (batch):        0.034530\n",
      "  Subject 1: median (max) validation accuracy = 0.988 (0.995)\n",
      "  Subject 2: median (max) validation accuracy = 0.986 (0.994)\n",
      "  Subject 3: median (max) validation accuracy = 0.990 (0.994)\n",
      "  Subject 4: median (max) validation accuracy = 0.990 (0.995)\n",
      "  Subject 5: median (max) validation accuracy = 0.989 (0.994)\n",
      "  Subject 6: median (max) validation accuracy = 0.989 (0.995)\n",
      "  Subject 7: median (max) validation accuracy = 0.989 (0.995)\n",
      "  Subject 8: median (max) validation accuracy = 0.990 (0.995)\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 46 of 50 took       46.549s\n",
      "  training loss:               0.057732\n",
      "  holdout loss (batch):        0.031472\n",
      "  Subject 1: median (max) validation accuracy = 0.988 (0.995)\n",
      "  Subject 2: median (max) validation accuracy = 0.987 (0.994)\n",
      "  Subject 3: median (max) validation accuracy = 0.990 (0.995)\n",
      "  Subject 4: median (max) validation accuracy = 0.990 (0.995)\n",
      "  Subject 5: median (max) validation accuracy = 0.989 (0.994)\n",
      "  Subject 6: median (max) validation accuracy = 0.989 (0.995)\n",
      "  Subject 7: median (max) validation accuracy = 0.990 (0.995)\n",
      "  Subject 8: median (max) validation accuracy = 0.990 (0.995)\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 47 of 50 took       46.551s\n",
      "  training loss:               0.057959\n",
      "  holdout loss (batch):        0.029148\n",
      "  Subject 1: median (max) validation accuracy = 0.988 (0.995)\n",
      "  Subject 2: median (max) validation accuracy = 0.987 (0.994)\n",
      "  Subject 3: median (max) validation accuracy = 0.990 (0.995)\n",
      "  Subject 4: median (max) validation accuracy = 0.991 (0.995)\n",
      "  Subject 5: median (max) validation accuracy = 0.989 (0.995)\n",
      "  Subject 6: median (max) validation accuracy = 0.989 (0.995)\n",
      "  Subject 7: median (max) validation accuracy = 0.990 (0.995)\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "  Subject 8: median (max) validation accuracy = 0.990 (0.995)\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 48 of 50 took       46.512s\n",
      "  training loss:               0.057204\n",
      "  holdout loss (batch):        0.032345\n",
      "  Subject 1: median (max) validation accuracy = 0.988 (0.995)\n",
      "  Subject 2: median (max) validation accuracy = 0.987 (0.994)\n",
      "  Subject 3: median (max) validation accuracy = 0.990 (0.995)\n",
      "  Subject 4: median (max) validation accuracy = 0.990 (0.995)\n",
      "  Subject 5: median (max) validation accuracy = 0.989 (0.995)\n",
      "  Subject 6: median (max) validation accuracy = 0.989 (0.995)\n",
      "  Subject 7: median (max) validation accuracy = 0.990 (0.995)\n",
      "  Subject 8: median (max) validation accuracy = 0.990 (0.995)\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 49 of 50 took       46.461s\n",
      "  training loss:               0.057009\n",
      "  holdout loss (batch):        0.029962\n",
      "  Subject 1: median (max) validation accuracy = 0.988 (0.995)\n",
      "  Subject 2: median (max) validation accuracy = 0.987 (0.994)\n",
      "  Subject 3: median (max) validation accuracy = 0.990 (0.995)\n",
      "  Subject 4: median (max) validation accuracy = 0.991 (0.995)\n",
      "  Subject 5: median (max) validation accuracy = 0.989 (0.995)\n",
      "  Subject 6: median (max) validation accuracy = 0.989 (0.995)\n",
      "  Subject 7: median (max) validation accuracy = 0.990 (0.995)\n",
      "  Subject 8: median (max) validation accuracy = 0.990 (0.995)\n",
      "** Saving params with joint score = 0.989 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 50 of 50 took       46.410s\n",
      "  training loss:               0.056864\n",
      "  holdout loss (batch):        0.029560\n",
      "  Subject 1: median (max) validation accuracy = 0.988 (0.995)\n",
      "  Subject 2: median (max) validation accuracy = 0.987 (0.994)\n",
      "  Subject 3: median (max) validation accuracy = 0.990 (0.995)\n",
      "  Subject 4: median (max) validation accuracy = 0.991 (0.995)\n",
      "  Subject 5: median (max) validation accuracy = 0.990 (0.995)\n",
      "  Subject 6: median (max) validation accuracy = 0.989 (0.995)\n",
      "  Subject 7: median (max) validation accuracy = 0.990 (0.995)\n",
      "  Subject 8: median (max) validation accuracy = 0.990 (0.995)\n",
      "** Saving params with joint score = 0.989 **\n",
      "\n",
      "***************************************\n",
      "best joint val cc = 0.990\n",
      "subject 1: val cc = 0.989\n",
      "subject 2: val cc = 0.988\n",
      "subject 3: val cc = 0.991\n",
      "subject 4: val cc = 0.991\n",
      "subject 5: val cc = 0.991\n",
      "subject 6: val cc = 0.990\n",
      "subject 7: val cc = 0.991\n",
      "subject 8: val cc = 0.991\n",
      "***************************************\n",
      "***************************************\n",
      "***                  V3       ***\n",
      "{1: 1188, 2: 1081, 3: 945, 4: 800, 5: 934, 6: 1215, 7: 721, 8: 879}\n",
      "***************************************\n",
      "subject 1 masked 1188 of 4627\n",
      "subject 1 training/holdout 8100 900\n",
      "subject 2 masked 1081 of 3765\n",
      "subject 2 training/holdout 8100 900\n",
      "subject 3 masked 945 of 3778\n",
      "subject 3 training/holdout 8100 900\n",
      "subject 4 masked 800 of 3027\n",
      "subject 4 training/holdout 8100 900\n",
      "subject 5 masked 934 of 3685\n",
      "subject 5 training/holdout 8100 900\n",
      "subject 6 masked 1215 of 4024\n",
      "subject 6 training/holdout 8100 900\n",
      "subject 7 masked 721 of 3201\n",
      "subject 7 training/holdout 8100 900\n",
      "subject 8 masked 879 of 3490\n",
      "subject 8 training/holdout 8100 900\n",
      "\\ : 100.0 %\n",
      "  Epoch 1 of 50 took       45.051s\n",
      "  training loss:               0.657562\n",
      "  holdout loss (batch):        0.509172\n",
      "  Subject 1: median (max) validation accuracy = 0.604 (0.920)\n",
      "  Subject 2: median (max) validation accuracy = 0.606 (0.912)\n",
      "  Subject 3: median (max) validation accuracy = 0.715 (0.902)\n",
      "  Subject 4: median (max) validation accuracy = 0.729 (0.906)\n",
      "  Subject 5: median (max) validation accuracy = 0.801 (0.931)\n",
      "  Subject 6: median (max) validation accuracy = 0.719 (0.915)\n",
      "  Subject 7: median (max) validation accuracy = 0.674 (0.920)\n",
      "  Subject 8: median (max) validation accuracy = 0.775 (0.952)\n",
      "** Saving params with joint score = 0.710 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 2 of 50 took       45.023s\n",
      "  training loss:               0.486277\n",
      "  holdout loss (batch):        0.382592\n",
      "  Subject 1: median (max) validation accuracy = 0.711 (0.944)\n",
      "  Subject 2: median (max) validation accuracy = 0.690 (0.936)\n",
      "  Subject 3: median (max) validation accuracy = 0.806 (0.939)\n",
      "  Subject 4: median (max) validation accuracy = 0.803 (0.933)\n",
      "  Subject 5: median (max) validation accuracy = 0.875 (0.951)\n",
      "  Subject 6: median (max) validation accuracy = 0.798 (0.937)\n",
      "  Subject 7: median (max) validation accuracy = 0.773 (0.935)\n",
      "  Subject 8: median (max) validation accuracy = 0.845 (0.959)\n",
      "** Saving params with joint score = 0.797 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 3 of 50 took       45.073s\n",
      "  training loss:               0.399000\n",
      "  holdout loss (batch):        0.317396\n",
      "  Subject 1: median (max) validation accuracy = 0.773 (0.953)\n",
      "  Subject 2: median (max) validation accuracy = 0.752 (0.950)\n",
      "  Subject 3: median (max) validation accuracy = 0.848 (0.946)\n",
      "  Subject 4: median (max) validation accuracy = 0.845 (0.947)\n",
      "  Subject 5: median (max) validation accuracy = 0.901 (0.960)\n",
      "  Subject 6: median (max) validation accuracy = 0.841 (0.947)\n",
      "  Subject 7: median (max) validation accuracy = 0.826 (0.946)\n",
      "  Subject 8: median (max) validation accuracy = 0.879 (0.965)\n",
      "** Saving params with joint score = 0.840 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 4 of 50 took       45.020s\n",
      "  training loss:               0.346610\n",
      "  holdout loss (batch):        0.280503\n",
      "  Subject 1: median (max) validation accuracy = 0.810 (0.961)\n",
      "  Subject 2: median (max) validation accuracy = 0.794 (0.959)\n",
      "  Subject 3: median (max) validation accuracy = 0.870 (0.951)\n",
      "  Subject 4: median (max) validation accuracy = 0.870 (0.954)\n",
      "  Subject 5: median (max) validation accuracy = 0.914 (0.965)\n",
      "  Subject 6: median (max) validation accuracy = 0.864 (0.955)\n",
      "  Subject 7: median (max) validation accuracy = 0.853 (0.950)\n",
      "  Subject 8: median (max) validation accuracy = 0.896 (0.971)\n",
      "** Saving params with joint score = 0.865 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 5 of 50 took       45.073s\n",
      "  training loss:               0.305567\n",
      "  holdout loss (batch):        0.234794\n",
      "  Subject 1: median (max) validation accuracy = 0.847 (0.966)\n",
      "  Subject 2: median (max) validation accuracy = 0.831 (0.961)\n",
      "  Subject 3: median (max) validation accuracy = 0.895 (0.959)\n",
      "  Subject 4: median (max) validation accuracy = 0.894 (0.961)\n",
      "  Subject 5: median (max) validation accuracy = 0.929 (0.973)\n",
      "  Subject 6: median (max) validation accuracy = 0.889 (0.963)\n",
      "  Subject 7: median (max) validation accuracy = 0.882 (0.956)\n",
      "  Subject 8: median (max) validation accuracy = 0.915 (0.974)\n",
      "** Saving params with joint score = 0.889 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 6 of 50 took       45.075s\n",
      "  training loss:               0.272626\n",
      "  holdout loss (batch):        0.202803\n",
      "  Subject 1: median (max) validation accuracy = 0.878 (0.971)\n",
      "  Subject 2: median (max) validation accuracy = 0.860 (0.964)\n",
      "  Subject 3: median (max) validation accuracy = 0.912 (0.965)\n",
      "  Subject 4: median (max) validation accuracy = 0.910 (0.968)\n",
      "  Subject 5: median (max) validation accuracy = 0.938 (0.977)\n",
      "  Subject 6: median (max) validation accuracy = 0.905 (0.971)\n",
      "  Subject 7: median (max) validation accuracy = 0.901 (0.960)\n",
      "  Subject 8: median (max) validation accuracy = 0.927 (0.978)\n",
      "** Saving params with joint score = 0.906 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 7 of 50 took       45.128s\n",
      "  training loss:               0.243621\n",
      "  holdout loss (batch):        0.182809\n",
      "  Subject 1: median (max) validation accuracy = 0.895 (0.974)\n",
      "  Subject 2: median (max) validation accuracy = 0.881 (0.968)\n",
      "  Subject 3: median (max) validation accuracy = 0.924 (0.970)\n",
      "  Subject 4: median (max) validation accuracy = 0.921 (0.971)\n",
      "  Subject 5: median (max) validation accuracy = 0.945 (0.979)\n",
      "  Subject 6: median (max) validation accuracy = 0.917 (0.975)\n",
      "  Subject 7: median (max) validation accuracy = 0.915 (0.963)\n",
      "  Subject 8: median (max) validation accuracy = 0.937 (0.980)\n",
      "** Saving params with joint score = 0.919 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 8 of 50 took       45.149s\n",
      "  training loss:               0.218614\n",
      "  holdout loss (batch):        0.161571\n",
      "  Subject 1: median (max) validation accuracy = 0.911 (0.977)\n",
      "  Subject 2: median (max) validation accuracy = 0.899 (0.971)\n",
      "  Subject 3: median (max) validation accuracy = 0.936 (0.976)\n",
      "  Subject 4: median (max) validation accuracy = 0.932 (0.975)\n",
      "  Subject 5: median (max) validation accuracy = 0.950 (0.982)\n",
      "  Subject 6: median (max) validation accuracy = 0.929 (0.980)\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "  Subject 7: median (max) validation accuracy = 0.926 (0.968)\n",
      "  Subject 8: median (max) validation accuracy = 0.944 (0.982)\n",
      "** Saving params with joint score = 0.929 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 9 of 50 took       45.403s\n",
      "  training loss:               0.198261\n",
      "  holdout loss (batch):        0.140478\n",
      "  Subject 1: median (max) validation accuracy = 0.924 (0.979)\n",
      "  Subject 2: median (max) validation accuracy = 0.913 (0.972)\n",
      "  Subject 3: median (max) validation accuracy = 0.945 (0.980)\n",
      "  Subject 4: median (max) validation accuracy = 0.941 (0.977)\n",
      "  Subject 5: median (max) validation accuracy = 0.955 (0.984)\n",
      "  Subject 6: median (max) validation accuracy = 0.937 (0.983)\n",
      "  Subject 7: median (max) validation accuracy = 0.938 (0.971)\n",
      "  Subject 8: median (max) validation accuracy = 0.951 (0.983)\n",
      "** Saving params with joint score = 0.939 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 10 of 50 took       45.177s\n",
      "  training loss:               0.179448\n",
      "  holdout loss (batch):        0.128017\n",
      "  Subject 1: median (max) validation accuracy = 0.934 (0.981)\n",
      "  Subject 2: median (max) validation accuracy = 0.925 (0.973)\n",
      "  Subject 3: median (max) validation accuracy = 0.951 (0.983)\n",
      "  Subject 4: median (max) validation accuracy = 0.948 (0.979)\n",
      "  Subject 5: median (max) validation accuracy = 0.958 (0.985)\n",
      "  Subject 6: median (max) validation accuracy = 0.944 (0.985)\n",
      "  Subject 7: median (max) validation accuracy = 0.946 (0.974)\n",
      "  Subject 8: median (max) validation accuracy = 0.956 (0.984)\n",
      "** Saving params with joint score = 0.945 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 11 of 50 took       45.113s\n",
      "  training loss:               0.162367\n",
      "  holdout loss (batch):        0.112014\n",
      "  Subject 1: median (max) validation accuracy = 0.942 (0.982)\n",
      "  Subject 2: median (max) validation accuracy = 0.935 (0.974)\n",
      "  Subject 3: median (max) validation accuracy = 0.957 (0.984)\n",
      "  Subject 4: median (max) validation accuracy = 0.955 (0.981)\n",
      "  Subject 5: median (max) validation accuracy = 0.962 (0.986)\n",
      "  Subject 6: median (max) validation accuracy = 0.949 (0.985)\n",
      "  Subject 7: median (max) validation accuracy = 0.952 (0.976)\n",
      "  Subject 8: median (max) validation accuracy = 0.961 (0.986)\n",
      "** Saving params with joint score = 0.951 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 12 of 50 took       45.080s\n",
      "  training loss:               0.146965\n",
      "  holdout loss (batch):        0.097668\n",
      "  Subject 1: median (max) validation accuracy = 0.951 (0.984)\n",
      "  Subject 2: median (max) validation accuracy = 0.946 (0.976)\n",
      "  Subject 3: median (max) validation accuracy = 0.964 (0.987)\n",
      "  Subject 4: median (max) validation accuracy = 0.961 (0.983)\n",
      "  Subject 5: median (max) validation accuracy = 0.966 (0.988)\n",
      "  Subject 6: median (max) validation accuracy = 0.956 (0.988)\n",
      "  Subject 7: median (max) validation accuracy = 0.959 (0.980)\n",
      "  Subject 8: median (max) validation accuracy = 0.966 (0.987)\n",
      "** Saving params with joint score = 0.958 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 13 of 50 took       45.085s\n",
      "  training loss:               0.133444\n",
      "  holdout loss (batch):        0.089302\n",
      "  Subject 1: median (max) validation accuracy = 0.958 (0.985)\n",
      "  Subject 2: median (max) validation accuracy = 0.953 (0.979)\n",
      "  Subject 3: median (max) validation accuracy = 0.968 (0.988)\n",
      "  Subject 4: median (max) validation accuracy = 0.966 (0.985)\n",
      "  Subject 5: median (max) validation accuracy = 0.968 (0.989)\n",
      "  Subject 6: median (max) validation accuracy = 0.960 (0.989)\n",
      "  Subject 7: median (max) validation accuracy = 0.963 (0.983)\n",
      "  Subject 8: median (max) validation accuracy = 0.970 (0.988)\n",
      "** Saving params with joint score = 0.962 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 14 of 50 took       45.094s\n",
      "  training loss:               0.122309\n",
      "  holdout loss (batch):        0.076794\n",
      "  Subject 1: median (max) validation accuracy = 0.963 (0.986)\n",
      "  Subject 2: median (max) validation accuracy = 0.959 (0.983)\n",
      "  Subject 3: median (max) validation accuracy = 0.972 (0.988)\n",
      "  Subject 4: median (max) validation accuracy = 0.970 (0.986)\n",
      "  Subject 5: median (max) validation accuracy = 0.970 (0.990)\n",
      "  Subject 6: median (max) validation accuracy = 0.965 (0.990)\n",
      "  Subject 7: median (max) validation accuracy = 0.968 (0.984)\n",
      "  Subject 8: median (max) validation accuracy = 0.973 (0.989)\n",
      "** Saving params with joint score = 0.967 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 15 of 50 took       45.045s\n",
      "  training loss:               0.113823\n",
      "  holdout loss (batch):        0.071154\n",
      "  Subject 1: median (max) validation accuracy = 0.967 (0.987)\n",
      "  Subject 2: median (max) validation accuracy = 0.963 (0.984)\n",
      "  Subject 3: median (max) validation accuracy = 0.974 (0.989)\n",
      "  Subject 4: median (max) validation accuracy = 0.973 (0.987)\n",
      "  Subject 5: median (max) validation accuracy = 0.973 (0.991)\n",
      "  Subject 6: median (max) validation accuracy = 0.968 (0.991)\n",
      "  Subject 7: median (max) validation accuracy = 0.970 (0.986)\n",
      "  Subject 8: median (max) validation accuracy = 0.976 (0.990)\n",
      "** Saving params with joint score = 0.970 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 16 of 50 took       44.995s\n",
      "  training loss:               0.106193\n",
      "  holdout loss (batch):        0.065927\n",
      "  Subject 1: median (max) validation accuracy = 0.971 (0.988)\n",
      "  Subject 2: median (max) validation accuracy = 0.967 (0.985)\n",
      "  Subject 3: median (max) validation accuracy = 0.977 (0.990)\n",
      "  Subject 4: median (max) validation accuracy = 0.976 (0.988)\n",
      "  Subject 5: median (max) validation accuracy = 0.974 (0.991)\n",
      "  Subject 6: median (max) validation accuracy = 0.972 (0.992)\n",
      "  Subject 7: median (max) validation accuracy = 0.974 (0.987)\n",
      "  Subject 8: median (max) validation accuracy = 0.978 (0.991)\n",
      "** Saving params with joint score = 0.973 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 17 of 50 took       44.937s\n",
      "  training loss:               0.100528\n",
      "  holdout loss (batch):        0.065987\n",
      "  Subject 1: median (max) validation accuracy = 0.973 (0.988)\n",
      "  Subject 2: median (max) validation accuracy = 0.969 (0.987)\n",
      "  Subject 3: median (max) validation accuracy = 0.978 (0.990)\n",
      "  Subject 4: median (max) validation accuracy = 0.977 (0.988)\n",
      "  Subject 5: median (max) validation accuracy = 0.975 (0.991)\n",
      "  Subject 6: median (max) validation accuracy = 0.973 (0.992)\n",
      "  Subject 7: median (max) validation accuracy = 0.975 (0.987)\n",
      "  Subject 8: median (max) validation accuracy = 0.979 (0.991)\n",
      "** Saving params with joint score = 0.975 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 18 of 50 took       44.965s\n",
      "  training loss:               0.096604\n",
      "  holdout loss (batch):        0.056970\n",
      "  Subject 1: median (max) validation accuracy = 0.976 (0.990)\n",
      "  Subject 2: median (max) validation accuracy = 0.972 (0.988)\n",
      "  Subject 3: median (max) validation accuracy = 0.981 (0.991)\n",
      "  Subject 4: median (max) validation accuracy = 0.979 (0.989)\n",
      "  Subject 5: median (max) validation accuracy = 0.977 (0.992)\n",
      "  Subject 6: median (max) validation accuracy = 0.976 (0.993)\n",
      "  Subject 7: median (max) validation accuracy = 0.977 (0.989)\n",
      "  Subject 8: median (max) validation accuracy = 0.981 (0.992)\n",
      "** Saving params with joint score = 0.977 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 19 of 50 took       45.053s\n",
      "  training loss:               0.092106\n",
      "  holdout loss (batch):        0.057686\n",
      "  Subject 1: median (max) validation accuracy = 0.977 (0.990)\n",
      "  Subject 2: median (max) validation accuracy = 0.973 (0.988)\n",
      "  Subject 3: median (max) validation accuracy = 0.981 (0.990)\n",
      "  Subject 4: median (max) validation accuracy = 0.980 (0.989)\n",
      "  Subject 5: median (max) validation accuracy = 0.977 (0.992)\n",
      "  Subject 6: median (max) validation accuracy = 0.977 (0.993)\n",
      "  Subject 7: median (max) validation accuracy = 0.978 (0.989)\n",
      "  Subject 8: median (max) validation accuracy = 0.982 (0.992)\n",
      "** Saving params with joint score = 0.978 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 20 of 50 took       45.059s\n",
      "  training loss:               0.089480\n",
      "  holdout loss (batch):        0.052156\n",
      "  Subject 1: median (max) validation accuracy = 0.979 (0.990)\n",
      "  Subject 2: median (max) validation accuracy = 0.975 (0.989)\n",
      "  Subject 3: median (max) validation accuracy = 0.982 (0.991)\n",
      "  Subject 4: median (max) validation accuracy = 0.981 (0.990)\n",
      "  Subject 5: median (max) validation accuracy = 0.978 (0.992)\n",
      "  Subject 6: median (max) validation accuracy = 0.978 (0.993)\n",
      "  Subject 7: median (max) validation accuracy = 0.979 (0.990)\n",
      "  Subject 8: median (max) validation accuracy = 0.983 (0.992)\n",
      "** Saving params with joint score = 0.979 **\n",
      "\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\\ : 100.0 %\n",
      "  Epoch 21 of 50 took       45.008s\n",
      "  training loss:               0.086348\n",
      "  holdout loss (batch):        0.053356\n",
      "  Subject 1: median (max) validation accuracy = 0.979 (0.991)\n",
      "  Subject 2: median (max) validation accuracy = 0.976 (0.989)\n",
      "  Subject 3: median (max) validation accuracy = 0.983 (0.991)\n",
      "  Subject 4: median (max) validation accuracy = 0.982 (0.990)\n",
      "  Subject 5: median (max) validation accuracy = 0.979 (0.993)\n",
      "  Subject 6: median (max) validation accuracy = 0.979 (0.994)\n",
      "  Subject 7: median (max) validation accuracy = 0.980 (0.990)\n",
      "  Subject 8: median (max) validation accuracy = 0.984 (0.992)\n",
      "** Saving params with joint score = 0.980 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 22 of 50 took       44.958s\n",
      "  training loss:               0.084129\n",
      "  holdout loss (batch):        0.052926\n",
      "  Subject 1: median (max) validation accuracy = 0.981 (0.991)\n",
      "  Subject 2: median (max) validation accuracy = 0.978 (0.990)\n",
      "  Subject 3: median (max) validation accuracy = 0.984 (0.992)\n",
      "  Subject 4: median (max) validation accuracy = 0.982 (0.991)\n",
      "  Subject 5: median (max) validation accuracy = 0.979 (0.993)\n",
      "  Subject 6: median (max) validation accuracy = 0.980 (0.994)\n",
      "  Subject 7: median (max) validation accuracy = 0.981 (0.990)\n",
      "  Subject 8: median (max) validation accuracy = 0.984 (0.992)\n",
      "** Saving params with joint score = 0.981 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 23 of 50 took       44.991s\n",
      "  training loss:               0.082104\n",
      "  holdout loss (batch):        0.049215\n",
      "  Subject 1: median (max) validation accuracy = 0.981 (0.991)\n",
      "  Subject 2: median (max) validation accuracy = 0.978 (0.990)\n",
      "  Subject 3: median (max) validation accuracy = 0.984 (0.992)\n",
      "  Subject 4: median (max) validation accuracy = 0.982 (0.991)\n",
      "  Subject 5: median (max) validation accuracy = 0.980 (0.993)\n",
      "  Subject 6: median (max) validation accuracy = 0.981 (0.994)\n",
      "  Subject 7: median (max) validation accuracy = 0.981 (0.991)\n",
      "  Subject 8: median (max) validation accuracy = 0.984 (0.993)\n",
      "** Saving params with joint score = 0.981 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 24 of 50 took       45.017s\n",
      "  training loss:               0.080281\n",
      "  holdout loss (batch):        0.047888\n",
      "  Subject 1: median (max) validation accuracy = 0.982 (0.992)\n",
      "  Subject 2: median (max) validation accuracy = 0.979 (0.991)\n",
      "  Subject 3: median (max) validation accuracy = 0.985 (0.992)\n",
      "  Subject 4: median (max) validation accuracy = 0.984 (0.991)\n",
      "  Subject 5: median (max) validation accuracy = 0.981 (0.993)\n",
      "  Subject 6: median (max) validation accuracy = 0.981 (0.994)\n",
      "  Subject 7: median (max) validation accuracy = 0.982 (0.991)\n",
      "  Subject 8: median (max) validation accuracy = 0.985 (0.993)\n",
      "** Saving params with joint score = 0.982 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 25 of 50 took       45.004s\n",
      "  training loss:               0.078450\n",
      "  holdout loss (batch):        0.045868\n",
      "  Subject 1: median (max) validation accuracy = 0.982 (0.992)\n",
      "  Subject 2: median (max) validation accuracy = 0.980 (0.991)\n",
      "  Subject 3: median (max) validation accuracy = 0.985 (0.992)\n",
      "  Subject 4: median (max) validation accuracy = 0.984 (0.991)\n",
      "  Subject 5: median (max) validation accuracy = 0.981 (0.993)\n",
      "  Subject 6: median (max) validation accuracy = 0.982 (0.994)\n",
      "  Subject 7: median (max) validation accuracy = 0.983 (0.991)\n",
      "  Subject 8: median (max) validation accuracy = 0.985 (0.993)\n",
      "** Saving params with joint score = 0.983 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 26 of 50 took       45.137s\n",
      "  training loss:               0.077433\n",
      "  holdout loss (batch):        0.046452\n",
      "  Subject 1: median (max) validation accuracy = 0.983 (0.992)\n",
      "  Subject 2: median (max) validation accuracy = 0.981 (0.991)\n",
      "  Subject 3: median (max) validation accuracy = 0.986 (0.992)\n",
      "  Subject 4: median (max) validation accuracy = 0.984 (0.991)\n",
      "  Subject 5: median (max) validation accuracy = 0.981 (0.993)\n",
      "  Subject 6: median (max) validation accuracy = 0.983 (0.994)\n",
      "  Subject 7: median (max) validation accuracy = 0.983 (0.992)\n",
      "  Subject 8: median (max) validation accuracy = 0.986 (0.993)\n",
      "** Saving params with joint score = 0.983 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 27 of 50 took       45.134s\n",
      "  training loss:               0.075765\n",
      "  holdout loss (batch):        0.043249\n",
      "  Subject 1: median (max) validation accuracy = 0.984 (0.993)\n",
      "  Subject 2: median (max) validation accuracy = 0.982 (0.991)\n",
      "  Subject 3: median (max) validation accuracy = 0.986 (0.992)\n",
      "  Subject 4: median (max) validation accuracy = 0.985 (0.992)\n",
      "  Subject 5: median (max) validation accuracy = 0.982 (0.994)\n",
      "  Subject 6: median (max) validation accuracy = 0.983 (0.995)\n",
      "  Subject 7: median (max) validation accuracy = 0.983 (0.992)\n",
      "  Subject 8: median (max) validation accuracy = 0.986 (0.994)\n",
      "** Saving params with joint score = 0.984 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 28 of 50 took       45.087s\n",
      "  training loss:               0.074629\n",
      "  holdout loss (batch):        0.042945\n",
      "  Subject 1: median (max) validation accuracy = 0.984 (0.993)\n",
      "  Subject 2: median (max) validation accuracy = 0.982 (0.992)\n",
      "  Subject 3: median (max) validation accuracy = 0.987 (0.993)\n",
      "  Subject 4: median (max) validation accuracy = 0.985 (0.993)\n",
      "  Subject 5: median (max) validation accuracy = 0.983 (0.994)\n",
      "  Subject 6: median (max) validation accuracy = 0.984 (0.995)\n",
      "  Subject 7: median (max) validation accuracy = 0.984 (0.992)\n",
      "  Subject 8: median (max) validation accuracy = 0.987 (0.994)\n",
      "** Saving params with joint score = 0.984 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 29 of 50 took       45.280s\n",
      "  training loss:               0.073644\n",
      "  holdout loss (batch):        0.043640\n",
      "  Subject 1: median (max) validation accuracy = 0.984 (0.993)\n",
      "  Subject 2: median (max) validation accuracy = 0.982 (0.992)\n",
      "  Subject 3: median (max) validation accuracy = 0.986 (0.993)\n",
      "  Subject 4: median (max) validation accuracy = 0.985 (0.992)\n",
      "  Subject 5: median (max) validation accuracy = 0.982 (0.994)\n",
      "  Subject 6: median (max) validation accuracy = 0.983 (0.994)\n",
      "  Subject 7: median (max) validation accuracy = 0.984 (0.992)\n",
      "  Subject 8: median (max) validation accuracy = 0.987 (0.994)\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 30 of 50 took       45.376s\n",
      "  training loss:               0.072839\n",
      "  holdout loss (batch):        0.044084\n",
      "  Subject 1: median (max) validation accuracy = 0.985 (0.993)\n",
      "  Subject 2: median (max) validation accuracy = 0.983 (0.992)\n",
      "  Subject 3: median (max) validation accuracy = 0.987 (0.993)\n",
      "  Subject 4: median (max) validation accuracy = 0.985 (0.993)\n",
      "  Subject 5: median (max) validation accuracy = 0.983 (0.994)\n",
      "  Subject 6: median (max) validation accuracy = 0.984 (0.995)\n",
      "  Subject 7: median (max) validation accuracy = 0.984 (0.993)\n",
      "  Subject 8: median (max) validation accuracy = 0.987 (0.994)\n",
      "** Saving params with joint score = 0.985 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 31 of 50 took       45.044s\n",
      "  training loss:               0.071326\n",
      "  holdout loss (batch):        0.040031\n",
      "  Subject 1: median (max) validation accuracy = 0.985 (0.993)\n",
      "  Subject 2: median (max) validation accuracy = 0.983 (0.992)\n",
      "  Subject 3: median (max) validation accuracy = 0.987 (0.993)\n",
      "  Subject 4: median (max) validation accuracy = 0.986 (0.992)\n",
      "  Subject 5: median (max) validation accuracy = 0.983 (0.993)\n",
      "  Subject 6: median (max) validation accuracy = 0.985 (0.995)\n",
      "  Subject 7: median (max) validation accuracy = 0.985 (0.993)\n",
      "  Subject 8: median (max) validation accuracy = 0.987 (0.994)\n",
      "** Saving params with joint score = 0.985 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 32 of 50 took       45.028s\n",
      "  training loss:               0.070823\n",
      "  holdout loss (batch):        0.037903\n",
      "  Subject 1: median (max) validation accuracy = 0.986 (0.993)\n",
      "  Subject 2: median (max) validation accuracy = 0.984 (0.991)\n",
      "  Subject 3: median (max) validation accuracy = 0.988 (0.994)\n",
      "  Subject 4: median (max) validation accuracy = 0.986 (0.993)\n",
      "  Subject 5: median (max) validation accuracy = 0.984 (0.994)\n",
      "  Subject 6: median (max) validation accuracy = 0.985 (0.995)\n",
      "  Subject 7: median (max) validation accuracy = 0.985 (0.993)\n",
      "  Subject 8: median (max) validation accuracy = 0.987 (0.994)\n",
      "** Saving params with joint score = 0.986 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 33 of 50 took       45.023s\n",
      "  training loss:               0.069649\n",
      "  holdout loss (batch):        0.039803\n",
      "  Subject 1: median (max) validation accuracy = 0.986 (0.993)\n",
      "  Subject 2: median (max) validation accuracy = 0.984 (0.992)\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "  Subject 3: median (max) validation accuracy = 0.988 (0.993)\n",
      "  Subject 4: median (max) validation accuracy = 0.986 (0.993)\n",
      "  Subject 5: median (max) validation accuracy = 0.984 (0.994)\n",
      "  Subject 6: median (max) validation accuracy = 0.985 (0.995)\n",
      "  Subject 7: median (max) validation accuracy = 0.985 (0.992)\n",
      "  Subject 8: median (max) validation accuracy = 0.988 (0.994)\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 34 of 50 took       45.131s\n",
      "  training loss:               0.069179\n",
      "  holdout loss (batch):        0.041092\n",
      "  Subject 1: median (max) validation accuracy = 0.986 (0.994)\n",
      "  Subject 2: median (max) validation accuracy = 0.984 (0.992)\n",
      "  Subject 3: median (max) validation accuracy = 0.988 (0.994)\n",
      "  Subject 4: median (max) validation accuracy = 0.987 (0.993)\n",
      "  Subject 5: median (max) validation accuracy = 0.984 (0.994)\n",
      "  Subject 6: median (max) validation accuracy = 0.985 (0.995)\n",
      "  Subject 7: median (max) validation accuracy = 0.985 (0.993)\n",
      "  Subject 8: median (max) validation accuracy = 0.987 (0.994)\n",
      "** Saving params with joint score = 0.986 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 35 of 50 took       45.117s\n",
      "  training loss:               0.068280\n",
      "  holdout loss (batch):        0.041423\n",
      "  Subject 1: median (max) validation accuracy = 0.986 (0.994)\n",
      "  Subject 2: median (max) validation accuracy = 0.984 (0.992)\n",
      "  Subject 3: median (max) validation accuracy = 0.987 (0.993)\n",
      "  Subject 4: median (max) validation accuracy = 0.986 (0.993)\n",
      "  Subject 5: median (max) validation accuracy = 0.984 (0.994)\n",
      "  Subject 6: median (max) validation accuracy = 0.985 (0.995)\n",
      "  Subject 7: median (max) validation accuracy = 0.985 (0.993)\n",
      "  Subject 8: median (max) validation accuracy = 0.988 (0.994)\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 36 of 50 took       45.172s\n",
      "  training loss:               0.067813\n",
      "  holdout loss (batch):        0.035301\n",
      "  Subject 1: median (max) validation accuracy = 0.987 (0.994)\n",
      "  Subject 2: median (max) validation accuracy = 0.985 (0.992)\n",
      "  Subject 3: median (max) validation accuracy = 0.989 (0.994)\n",
      "  Subject 4: median (max) validation accuracy = 0.987 (0.994)\n",
      "  Subject 5: median (max) validation accuracy = 0.985 (0.995)\n",
      "  Subject 6: median (max) validation accuracy = 0.986 (0.996)\n",
      "  Subject 7: median (max) validation accuracy = 0.986 (0.993)\n",
      "  Subject 8: median (max) validation accuracy = 0.988 (0.995)\n",
      "** Saving params with joint score = 0.987 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 37 of 50 took       45.076s\n",
      "  training loss:               0.067039\n",
      "  holdout loss (batch):        0.036492\n",
      "  Subject 1: median (max) validation accuracy = 0.986 (0.993)\n",
      "  Subject 2: median (max) validation accuracy = 0.985 (0.992)\n",
      "  Subject 3: median (max) validation accuracy = 0.988 (0.993)\n",
      "  Subject 4: median (max) validation accuracy = 0.987 (0.993)\n",
      "  Subject 5: median (max) validation accuracy = 0.984 (0.994)\n",
      "  Subject 6: median (max) validation accuracy = 0.985 (0.995)\n",
      "  Subject 7: median (max) validation accuracy = 0.986 (0.993)\n",
      "  Subject 8: median (max) validation accuracy = 0.988 (0.994)\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 38 of 50 took       45.091s\n",
      "  training loss:               0.066496\n",
      "  holdout loss (batch):        0.037992\n",
      "  Subject 1: median (max) validation accuracy = 0.986 (0.994)\n",
      "  Subject 2: median (max) validation accuracy = 0.984 (0.992)\n",
      "  Subject 3: median (max) validation accuracy = 0.988 (0.993)\n",
      "  Subject 4: median (max) validation accuracy = 0.986 (0.992)\n",
      "  Subject 5: median (max) validation accuracy = 0.984 (0.994)\n",
      "  Subject 6: median (max) validation accuracy = 0.986 (0.995)\n",
      "  Subject 7: median (max) validation accuracy = 0.985 (0.993)\n",
      "  Subject 8: median (max) validation accuracy = 0.987 (0.993)\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 39 of 50 took       45.039s\n",
      "  training loss:               0.065992\n",
      "  holdout loss (batch):        0.036536\n",
      "  Subject 1: median (max) validation accuracy = 0.987 (0.994)\n",
      "  Subject 2: median (max) validation accuracy = 0.986 (0.993)\n",
      "  Subject 3: median (max) validation accuracy = 0.989 (0.994)\n",
      "  Subject 4: median (max) validation accuracy = 0.987 (0.994)\n",
      "  Subject 5: median (max) validation accuracy = 0.985 (0.995)\n",
      "  Subject 6: median (max) validation accuracy = 0.986 (0.995)\n",
      "  Subject 7: median (max) validation accuracy = 0.987 (0.994)\n",
      "  Subject 8: median (max) validation accuracy = 0.989 (0.994)\n",
      "** Saving params with joint score = 0.987 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 40 of 50 took       44.980s\n",
      "  training loss:               0.065602\n",
      "  holdout loss (batch):        0.036958\n",
      "  Subject 1: median (max) validation accuracy = 0.987 (0.994)\n",
      "  Subject 2: median (max) validation accuracy = 0.986 (0.993)\n",
      "  Subject 3: median (max) validation accuracy = 0.989 (0.994)\n",
      "  Subject 4: median (max) validation accuracy = 0.988 (0.993)\n",
      "  Subject 5: median (max) validation accuracy = 0.985 (0.994)\n",
      "  Subject 6: median (max) validation accuracy = 0.986 (0.995)\n",
      "  Subject 7: median (max) validation accuracy = 0.986 (0.993)\n",
      "  Subject 8: median (max) validation accuracy = 0.988 (0.995)\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 41 of 50 took       45.047s\n",
      "  training loss:               0.064938\n",
      "  holdout loss (batch):        0.034631\n",
      "  Subject 1: median (max) validation accuracy = 0.987 (0.994)\n",
      "  Subject 2: median (max) validation accuracy = 0.986 (0.993)\n",
      "  Subject 3: median (max) validation accuracy = 0.989 (0.994)\n",
      "  Subject 4: median (max) validation accuracy = 0.987 (0.993)\n",
      "  Subject 5: median (max) validation accuracy = 0.985 (0.994)\n",
      "  Subject 6: median (max) validation accuracy = 0.986 (0.995)\n",
      "  Subject 7: median (max) validation accuracy = 0.986 (0.993)\n",
      "  Subject 8: median (max) validation accuracy = 0.989 (0.994)\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 42 of 50 took       45.011s\n",
      "  training loss:               0.064573\n",
      "  holdout loss (batch):        0.034879\n",
      "  Subject 1: median (max) validation accuracy = 0.987 (0.994)\n",
      "  Subject 2: median (max) validation accuracy = 0.986 (0.992)\n",
      "  Subject 3: median (max) validation accuracy = 0.989 (0.994)\n",
      "  Subject 4: median (max) validation accuracy = 0.988 (0.994)\n",
      "  Subject 5: median (max) validation accuracy = 0.986 (0.995)\n",
      "  Subject 6: median (max) validation accuracy = 0.986 (0.995)\n",
      "  Subject 7: median (max) validation accuracy = 0.987 (0.993)\n",
      "  Subject 8: median (max) validation accuracy = 0.989 (0.995)\n",
      "** Saving params with joint score = 0.987 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 43 of 50 took       44.916s\n",
      "  training loss:               0.064393\n",
      "  holdout loss (batch):        0.040272\n",
      "  Subject 1: median (max) validation accuracy = 0.987 (0.994)\n",
      "  Subject 2: median (max) validation accuracy = 0.986 (0.993)\n",
      "  Subject 3: median (max) validation accuracy = 0.989 (0.994)\n",
      "  Subject 4: median (max) validation accuracy = 0.987 (0.994)\n",
      "  Subject 5: median (max) validation accuracy = 0.986 (0.995)\n",
      "  Subject 6: median (max) validation accuracy = 0.986 (0.995)\n",
      "  Subject 7: median (max) validation accuracy = 0.987 (0.994)\n",
      "  Subject 8: median (max) validation accuracy = 0.989 (0.995)\n",
      "** Saving params with joint score = 0.987 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 44 of 50 took       44.952s\n",
      "  training loss:               0.063534\n",
      "  holdout loss (batch):        0.036860\n",
      "  Subject 1: median (max) validation accuracy = 0.988 (0.994)\n",
      "  Subject 2: median (max) validation accuracy = 0.986 (0.993)\n",
      "  Subject 3: median (max) validation accuracy = 0.989 (0.994)\n",
      "  Subject 4: median (max) validation accuracy = 0.988 (0.994)\n",
      "  Subject 5: median (max) validation accuracy = 0.986 (0.995)\n",
      "  Subject 6: median (max) validation accuracy = 0.987 (0.996)\n",
      "  Subject 7: median (max) validation accuracy = 0.987 (0.994)\n",
      "  Subject 8: median (max) validation accuracy = 0.989 (0.995)\n",
      "** Saving params with joint score = 0.987 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 45 of 50 took       44.949s\n",
      "  training loss:               0.063194\n",
      "  holdout loss (batch):        0.033233\n",
      "  Subject 1: median (max) validation accuracy = 0.988 (0.994)\n",
      "  Subject 2: median (max) validation accuracy = 0.986 (0.993)\n",
      "  Subject 3: median (max) validation accuracy = 0.989 (0.994)\n",
      "  Subject 4: median (max) validation accuracy = 0.988 (0.994)\n",
      "  Subject 5: median (max) validation accuracy = 0.986 (0.995)\n",
      "  Subject 6: median (max) validation accuracy = 0.987 (0.996)\n",
      "  Subject 7: median (max) validation accuracy = 0.987 (0.994)\n",
      "  Subject 8: median (max) validation accuracy = 0.989 (0.995)\n",
      "** Saving params with joint score = 0.988 **\n",
      "\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\\ : 100.0 %\n",
      "  Epoch 46 of 50 took       45.052s\n",
      "  training loss:               0.062855\n",
      "  holdout loss (batch):        0.038643\n",
      "  Subject 1: median (max) validation accuracy = 0.987 (0.994)\n",
      "  Subject 2: median (max) validation accuracy = 0.986 (0.993)\n",
      "  Subject 3: median (max) validation accuracy = 0.989 (0.994)\n",
      "  Subject 4: median (max) validation accuracy = 0.988 (0.994)\n",
      "  Subject 5: median (max) validation accuracy = 0.985 (0.995)\n",
      "  Subject 6: median (max) validation accuracy = 0.986 (0.996)\n",
      "  Subject 7: median (max) validation accuracy = 0.987 (0.994)\n",
      "  Subject 8: median (max) validation accuracy = 0.989 (0.995)\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 47 of 50 took       45.112s\n",
      "  training loss:               0.062386\n",
      "  holdout loss (batch):        0.035142\n",
      "  Subject 1: median (max) validation accuracy = 0.988 (0.994)\n",
      "  Subject 2: median (max) validation accuracy = 0.986 (0.993)\n",
      "  Subject 3: median (max) validation accuracy = 0.989 (0.994)\n",
      "  Subject 4: median (max) validation accuracy = 0.988 (0.994)\n",
      "  Subject 5: median (max) validation accuracy = 0.986 (0.994)\n",
      "  Subject 6: median (max) validation accuracy = 0.987 (0.995)\n",
      "  Subject 7: median (max) validation accuracy = 0.987 (0.993)\n",
      "  Subject 8: median (max) validation accuracy = 0.989 (0.994)\n",
      "** Saving params with joint score = 0.988 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 48 of 50 took       45.166s\n",
      "  training loss:               0.062099\n",
      "  holdout loss (batch):        0.033824\n",
      "  Subject 1: median (max) validation accuracy = 0.988 (0.994)\n",
      "  Subject 2: median (max) validation accuracy = 0.987 (0.993)\n",
      "  Subject 3: median (max) validation accuracy = 0.989 (0.994)\n",
      "  Subject 4: median (max) validation accuracy = 0.988 (0.994)\n",
      "  Subject 5: median (max) validation accuracy = 0.986 (0.995)\n",
      "  Subject 6: median (max) validation accuracy = 0.987 (0.996)\n",
      "  Subject 7: median (max) validation accuracy = 0.987 (0.994)\n",
      "  Subject 8: median (max) validation accuracy = 0.989 (0.995)\n",
      "** Saving params with joint score = 0.988 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 49 of 50 took       45.228s\n",
      "  training loss:               0.061919\n",
      "  holdout loss (batch):        0.032750\n",
      "  Subject 1: median (max) validation accuracy = 0.988 (0.994)\n",
      "  Subject 2: median (max) validation accuracy = 0.987 (0.993)\n",
      "  Subject 3: median (max) validation accuracy = 0.990 (0.994)\n",
      "  Subject 4: median (max) validation accuracy = 0.989 (0.994)\n",
      "  Subject 5: median (max) validation accuracy = 0.986 (0.995)\n",
      "  Subject 6: median (max) validation accuracy = 0.987 (0.996)\n",
      "  Subject 7: median (max) validation accuracy = 0.987 (0.994)\n",
      "  Subject 8: median (max) validation accuracy = 0.989 (0.995)\n",
      "** Saving params with joint score = 0.988 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 50 of 50 took       45.355s\n",
      "  training loss:               0.061782\n",
      "  holdout loss (batch):        0.035460\n",
      "  Subject 1: median (max) validation accuracy = 0.988 (0.994)\n",
      "  Subject 2: median (max) validation accuracy = 0.987 (0.993)\n",
      "  Subject 3: median (max) validation accuracy = 0.989 (0.994)\n",
      "  Subject 4: median (max) validation accuracy = 0.988 (0.994)\n",
      "  Subject 5: median (max) validation accuracy = 0.986 (0.995)\n",
      "  Subject 6: median (max) validation accuracy = 0.987 (0.995)\n",
      "  Subject 7: median (max) validation accuracy = 0.987 (0.993)\n",
      "  Subject 8: median (max) validation accuracy = 0.989 (0.994)\n",
      "\n",
      "***************************************\n",
      "best joint val cc = 0.989\n",
      "subject 1: val cc = 0.989\n",
      "subject 2: val cc = 0.988\n",
      "subject 3: val cc = 0.990\n",
      "subject 4: val cc = 0.989\n",
      "subject 5: val cc = 0.988\n",
      "subject 6: val cc = 0.988\n",
      "subject 7: val cc = 0.988\n",
      "subject 8: val cc = 0.990\n",
      "***************************************\n",
      "***************************************\n",
      "***                 hV4       ***\n",
      "{1: 676, 2: 504, 3: 422, 4: 473, 5: 548, 6: 488, 7: 374, 8: 490}\n",
      "***************************************\n",
      "subject 1 masked 676 of 4627\n",
      "subject 1 training/holdout 8100 900\n",
      "subject 2 masked 504 of 3765\n",
      "subject 2 training/holdout 8100 900\n",
      "subject 3 masked 422 of 3778\n",
      "subject 3 training/holdout 8100 900\n",
      "subject 4 masked 473 of 3027\n",
      "subject 4 training/holdout 8100 900\n",
      "subject 5 masked 548 of 3685\n",
      "subject 5 training/holdout 8100 900\n",
      "subject 6 masked 488 of 4024\n",
      "subject 6 training/holdout 8100 900\n",
      "subject 7 masked 374 of 3201\n",
      "subject 7 training/holdout 8100 900\n",
      "subject 8 masked 490 of 3490\n",
      "subject 8 training/holdout 8100 900\n",
      "\\ : 100.0 %\n",
      "  Epoch 1 of 50 took       40.058s\n",
      "  training loss:               0.631003\n",
      "  holdout loss (batch):        0.487825\n",
      "  Subject 1: median (max) validation accuracy = 0.648 (0.904)\n",
      "  Subject 2: median (max) validation accuracy = 0.669 (0.909)\n",
      "  Subject 3: median (max) validation accuracy = 0.734 (0.918)\n",
      "  Subject 4: median (max) validation accuracy = 0.710 (0.940)\n",
      "  Subject 5: median (max) validation accuracy = 0.786 (0.937)\n",
      "  Subject 6: median (max) validation accuracy = 0.743 (0.944)\n",
      "  Subject 7: median (max) validation accuracy = 0.664 (0.897)\n",
      "  Subject 8: median (max) validation accuracy = 0.812 (0.950)\n",
      "** Saving params with joint score = 0.722 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 2 of 50 took       39.907s\n",
      "  training loss:               0.494879\n",
      "  holdout loss (batch):        0.401109\n",
      "  Subject 1: median (max) validation accuracy = 0.729 (0.930)\n",
      "  Subject 2: median (max) validation accuracy = 0.731 (0.930)\n",
      "  Subject 3: median (max) validation accuracy = 0.794 (0.945)\n",
      "  Subject 4: median (max) validation accuracy = 0.788 (0.952)\n",
      "  Subject 5: median (max) validation accuracy = 0.845 (0.951)\n",
      "  Subject 6: median (max) validation accuracy = 0.798 (0.950)\n",
      "  Subject 7: median (max) validation accuracy = 0.737 (0.921)\n",
      "  Subject 8: median (max) validation accuracy = 0.854 (0.961)\n",
      "** Saving params with joint score = 0.789 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 3 of 50 took       39.950s\n",
      "  training loss:               0.413273\n",
      "  holdout loss (batch):        0.310255\n",
      "  Subject 1: median (max) validation accuracy = 0.801 (0.948)\n",
      "  Subject 2: median (max) validation accuracy = 0.802 (0.942)\n",
      "  Subject 3: median (max) validation accuracy = 0.856 (0.960)\n",
      "  Subject 4: median (max) validation accuracy = 0.855 (0.960)\n",
      "  Subject 5: median (max) validation accuracy = 0.888 (0.961)\n",
      "  Subject 6: median (max) validation accuracy = 0.854 (0.961)\n",
      "  Subject 7: median (max) validation accuracy = 0.819 (0.930)\n",
      "  Subject 8: median (max) validation accuracy = 0.898 (0.970)\n",
      "** Saving params with joint score = 0.849 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 4 of 50 took       40.291s\n",
      "  training loss:               0.341709\n",
      "  holdout loss (batch):        0.249483\n",
      "  Subject 1: median (max) validation accuracy = 0.854 (0.968)\n",
      "  Subject 2: median (max) validation accuracy = 0.866 (0.952)\n",
      "  Subject 3: median (max) validation accuracy = 0.889 (0.965)\n",
      "  Subject 4: median (max) validation accuracy = 0.891 (0.968)\n",
      "  Subject 5: median (max) validation accuracy = 0.907 (0.968)\n",
      "  Subject 6: median (max) validation accuracy = 0.884 (0.968)\n",
      "  Subject 7: median (max) validation accuracy = 0.862 (0.937)\n",
      "  Subject 8: median (max) validation accuracy = 0.922 (0.976)\n",
      "** Saving params with joint score = 0.885 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 5 of 50 took       40.329s\n",
      "  training loss:               0.292546\n",
      "  holdout loss (batch):        0.209659\n",
      "  Subject 1: median (max) validation accuracy = 0.881 (0.974)\n",
      "  Subject 2: median (max) validation accuracy = 0.893 (0.958)\n",
      "  Subject 3: median (max) validation accuracy = 0.909 (0.969)\n",
      "  Subject 4: median (max) validation accuracy = 0.909 (0.974)\n",
      "  Subject 5: median (max) validation accuracy = 0.920 (0.973)\n",
      "  Subject 6: median (max) validation accuracy = 0.903 (0.973)\n",
      "  Subject 7: median (max) validation accuracy = 0.883 (0.943)\n",
      "  Subject 8: median (max) validation accuracy = 0.935 (0.979)\n",
      "** Saving params with joint score = 0.904 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 6 of 50 took       40.308s\n",
      "  training loss:               0.260735\n",
      "  holdout loss (batch):        0.178512\n",
      "  Subject 1: median (max) validation accuracy = 0.898 (0.978)\n",
      "  Subject 2: median (max) validation accuracy = 0.910 (0.963)\n",
      "  Subject 3: median (max) validation accuracy = 0.921 (0.972)\n",
      "  Subject 4: median (max) validation accuracy = 0.922 (0.975)\n",
      "  Subject 5: median (max) validation accuracy = 0.930 (0.978)\n",
      "  Subject 6: median (max) validation accuracy = 0.915 (0.978)\n",
      "  Subject 7: median (max) validation accuracy = 0.902 (0.954)\n",
      "  Subject 8: median (max) validation accuracy = 0.945 (0.982)\n",
      "** Saving params with joint score = 0.919 **\n",
      "\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\\ : 100.0 %\n",
      "  Epoch 7 of 50 took       40.292s\n",
      "  training loss:               0.236535\n",
      "  holdout loss (batch):        0.160094\n",
      "  Subject 1: median (max) validation accuracy = 0.912 (0.980)\n",
      "  Subject 2: median (max) validation accuracy = 0.925 (0.967)\n",
      "  Subject 3: median (max) validation accuracy = 0.931 (0.974)\n",
      "  Subject 4: median (max) validation accuracy = 0.931 (0.978)\n",
      "  Subject 5: median (max) validation accuracy = 0.939 (0.980)\n",
      "  Subject 6: median (max) validation accuracy = 0.925 (0.980)\n",
      "  Subject 7: median (max) validation accuracy = 0.913 (0.960)\n",
      "  Subject 8: median (max) validation accuracy = 0.951 (0.983)\n",
      "** Saving params with joint score = 0.929 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 8 of 50 took       40.245s\n",
      "  training loss:               0.216887\n",
      "  holdout loss (batch):        0.144944\n",
      "  Subject 1: median (max) validation accuracy = 0.923 (0.981)\n",
      "  Subject 2: median (max) validation accuracy = 0.933 (0.968)\n",
      "  Subject 3: median (max) validation accuracy = 0.939 (0.976)\n",
      "  Subject 4: median (max) validation accuracy = 0.935 (0.980)\n",
      "  Subject 5: median (max) validation accuracy = 0.945 (0.982)\n",
      "  Subject 6: median (max) validation accuracy = 0.933 (0.982)\n",
      "  Subject 7: median (max) validation accuracy = 0.921 (0.962)\n",
      "  Subject 8: median (max) validation accuracy = 0.956 (0.985)\n",
      "** Saving params with joint score = 0.936 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 9 of 50 took       40.278s\n",
      "  training loss:               0.199925\n",
      "  holdout loss (batch):        0.131988\n",
      "  Subject 1: median (max) validation accuracy = 0.933 (0.982)\n",
      "  Subject 2: median (max) validation accuracy = 0.942 (0.972)\n",
      "  Subject 3: median (max) validation accuracy = 0.946 (0.978)\n",
      "  Subject 4: median (max) validation accuracy = 0.943 (0.981)\n",
      "  Subject 5: median (max) validation accuracy = 0.951 (0.983)\n",
      "  Subject 6: median (max) validation accuracy = 0.940 (0.984)\n",
      "  Subject 7: median (max) validation accuracy = 0.932 (0.967)\n",
      "  Subject 8: median (max) validation accuracy = 0.960 (0.986)\n",
      "** Saving params with joint score = 0.943 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 10 of 50 took       40.336s\n",
      "  training loss:               0.183906\n",
      "  holdout loss (batch):        0.123367\n",
      "  Subject 1: median (max) validation accuracy = 0.942 (0.984)\n",
      "  Subject 2: median (max) validation accuracy = 0.949 (0.974)\n",
      "  Subject 3: median (max) validation accuracy = 0.953 (0.980)\n",
      "  Subject 4: median (max) validation accuracy = 0.949 (0.982)\n",
      "  Subject 5: median (max) validation accuracy = 0.956 (0.985)\n",
      "  Subject 6: median (max) validation accuracy = 0.947 (0.986)\n",
      "  Subject 7: median (max) validation accuracy = 0.941 (0.971)\n",
      "  Subject 8: median (max) validation accuracy = 0.965 (0.987)\n",
      "** Saving params with joint score = 0.950 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 11 of 50 took       40.247s\n",
      "  training loss:               0.171025\n",
      "  holdout loss (batch):        0.111886\n",
      "  Subject 1: median (max) validation accuracy = 0.950 (0.985)\n",
      "  Subject 2: median (max) validation accuracy = 0.955 (0.977)\n",
      "  Subject 3: median (max) validation accuracy = 0.959 (0.982)\n",
      "  Subject 4: median (max) validation accuracy = 0.955 (0.984)\n",
      "  Subject 5: median (max) validation accuracy = 0.961 (0.986)\n",
      "  Subject 6: median (max) validation accuracy = 0.953 (0.987)\n",
      "  Subject 7: median (max) validation accuracy = 0.947 (0.974)\n",
      "  Subject 8: median (max) validation accuracy = 0.969 (0.988)\n",
      "** Saving params with joint score = 0.956 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 12 of 50 took       40.262s\n",
      "  training loss:               0.160308\n",
      "  holdout loss (batch):        0.100146\n",
      "  Subject 1: median (max) validation accuracy = 0.953 (0.986)\n",
      "  Subject 2: median (max) validation accuracy = 0.959 (0.978)\n",
      "  Subject 3: median (max) validation accuracy = 0.961 (0.982)\n",
      "  Subject 4: median (max) validation accuracy = 0.957 (0.985)\n",
      "  Subject 5: median (max) validation accuracy = 0.963 (0.986)\n",
      "  Subject 6: median (max) validation accuracy = 0.955 (0.988)\n",
      "  Subject 7: median (max) validation accuracy = 0.952 (0.977)\n",
      "  Subject 8: median (max) validation accuracy = 0.971 (0.989)\n",
      "** Saving params with joint score = 0.959 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 13 of 50 took       40.241s\n",
      "  training loss:               0.151288\n",
      "  holdout loss (batch):        0.094439\n",
      "  Subject 1: median (max) validation accuracy = 0.957 (0.985)\n",
      "  Subject 2: median (max) validation accuracy = 0.962 (0.979)\n",
      "  Subject 3: median (max) validation accuracy = 0.964 (0.984)\n",
      "  Subject 4: median (max) validation accuracy = 0.961 (0.985)\n",
      "  Subject 5: median (max) validation accuracy = 0.965 (0.987)\n",
      "  Subject 6: median (max) validation accuracy = 0.959 (0.989)\n",
      "  Subject 7: median (max) validation accuracy = 0.955 (0.978)\n",
      "  Subject 8: median (max) validation accuracy = 0.973 (0.988)\n",
      "** Saving params with joint score = 0.962 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 14 of 50 took       40.256s\n",
      "  training loss:               0.143154\n",
      "  holdout loss (batch):        0.089236\n",
      "  Subject 1: median (max) validation accuracy = 0.962 (0.987)\n",
      "  Subject 2: median (max) validation accuracy = 0.966 (0.981)\n",
      "  Subject 3: median (max) validation accuracy = 0.968 (0.985)\n",
      "  Subject 4: median (max) validation accuracy = 0.964 (0.987)\n",
      "  Subject 5: median (max) validation accuracy = 0.968 (0.987)\n",
      "  Subject 6: median (max) validation accuracy = 0.962 (0.989)\n",
      "  Subject 7: median (max) validation accuracy = 0.959 (0.980)\n",
      "  Subject 8: median (max) validation accuracy = 0.976 (0.989)\n",
      "** Saving params with joint score = 0.965 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 15 of 50 took       40.199s\n",
      "  training loss:               0.135942\n",
      "  holdout loss (batch):        0.082106\n",
      "  Subject 1: median (max) validation accuracy = 0.964 (0.988)\n",
      "  Subject 2: median (max) validation accuracy = 0.968 (0.982)\n",
      "  Subject 3: median (max) validation accuracy = 0.969 (0.985)\n",
      "  Subject 4: median (max) validation accuracy = 0.966 (0.988)\n",
      "  Subject 5: median (max) validation accuracy = 0.968 (0.988)\n",
      "  Subject 6: median (max) validation accuracy = 0.966 (0.990)\n",
      "  Subject 7: median (max) validation accuracy = 0.961 (0.982)\n",
      "  Subject 8: median (max) validation accuracy = 0.977 (0.990)\n",
      "** Saving params with joint score = 0.967 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 16 of 50 took       39.938s\n",
      "  training loss:               0.129796\n",
      "  holdout loss (batch):        0.080373\n",
      "  Subject 1: median (max) validation accuracy = 0.966 (0.987)\n",
      "  Subject 2: median (max) validation accuracy = 0.970 (0.983)\n",
      "  Subject 3: median (max) validation accuracy = 0.972 (0.986)\n",
      "  Subject 4: median (max) validation accuracy = 0.967 (0.988)\n",
      "  Subject 5: median (max) validation accuracy = 0.970 (0.988)\n",
      "  Subject 6: median (max) validation accuracy = 0.968 (0.990)\n",
      "  Subject 7: median (max) validation accuracy = 0.963 (0.982)\n",
      "  Subject 8: median (max) validation accuracy = 0.978 (0.990)\n",
      "** Saving params with joint score = 0.969 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 17 of 50 took       39.855s\n",
      "  training loss:               0.124996\n",
      "  holdout loss (batch):        0.074262\n",
      "  Subject 1: median (max) validation accuracy = 0.969 (0.988)\n",
      "  Subject 2: median (max) validation accuracy = 0.973 (0.985)\n",
      "  Subject 3: median (max) validation accuracy = 0.974 (0.987)\n",
      "  Subject 4: median (max) validation accuracy = 0.970 (0.990)\n",
      "  Subject 5: median (max) validation accuracy = 0.972 (0.989)\n",
      "  Subject 6: median (max) validation accuracy = 0.971 (0.991)\n",
      "  Subject 7: median (max) validation accuracy = 0.966 (0.983)\n",
      "  Subject 8: median (max) validation accuracy = 0.979 (0.991)\n",
      "** Saving params with joint score = 0.972 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 18 of 50 took       40.499s\n",
      "  training loss:               0.121213\n",
      "  holdout loss (batch):        0.068330\n",
      "  Subject 1: median (max) validation accuracy = 0.971 (0.989)\n",
      "  Subject 2: median (max) validation accuracy = 0.975 (0.986)\n",
      "  Subject 3: median (max) validation accuracy = 0.975 (0.987)\n",
      "  Subject 4: median (max) validation accuracy = 0.971 (0.990)\n",
      "  Subject 5: median (max) validation accuracy = 0.973 (0.989)\n",
      "  Subject 6: median (max) validation accuracy = 0.972 (0.992)\n",
      "  Subject 7: median (max) validation accuracy = 0.968 (0.985)\n",
      "  Subject 8: median (max) validation accuracy = 0.981 (0.991)\n",
      "** Saving params with joint score = 0.973 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 19 of 50 took       40.473s\n",
      "  training loss:               0.117347\n",
      "  holdout loss (batch):        0.071406\n",
      "  Subject 1: median (max) validation accuracy = 0.972 (0.988)\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "  Subject 2: median (max) validation accuracy = 0.975 (0.985)\n",
      "  Subject 3: median (max) validation accuracy = 0.975 (0.987)\n",
      "  Subject 4: median (max) validation accuracy = 0.972 (0.990)\n",
      "  Subject 5: median (max) validation accuracy = 0.972 (0.988)\n",
      "  Subject 6: median (max) validation accuracy = 0.972 (0.991)\n",
      "  Subject 7: median (max) validation accuracy = 0.968 (0.984)\n",
      "  Subject 8: median (max) validation accuracy = 0.980 (0.990)\n",
      "** Saving params with joint score = 0.973 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 20 of 50 took       40.499s\n",
      "  training loss:               0.113885\n",
      "  holdout loss (batch):        0.075339\n",
      "  Subject 1: median (max) validation accuracy = 0.973 (0.989)\n",
      "  Subject 2: median (max) validation accuracy = 0.976 (0.987)\n",
      "  Subject 3: median (max) validation accuracy = 0.976 (0.988)\n",
      "  Subject 4: median (max) validation accuracy = 0.973 (0.990)\n",
      "  Subject 5: median (max) validation accuracy = 0.974 (0.989)\n",
      "  Subject 6: median (max) validation accuracy = 0.974 (0.992)\n",
      "  Subject 7: median (max) validation accuracy = 0.970 (0.985)\n",
      "  Subject 8: median (max) validation accuracy = 0.981 (0.991)\n",
      "** Saving params with joint score = 0.975 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 21 of 50 took       40.586s\n",
      "  training loss:               0.111613\n",
      "  holdout loss (batch):        0.064319\n",
      "  Subject 1: median (max) validation accuracy = 0.974 (0.989)\n",
      "  Subject 2: median (max) validation accuracy = 0.977 (0.987)\n",
      "  Subject 3: median (max) validation accuracy = 0.977 (0.988)\n",
      "  Subject 4: median (max) validation accuracy = 0.974 (0.991)\n",
      "  Subject 5: median (max) validation accuracy = 0.975 (0.989)\n",
      "  Subject 6: median (max) validation accuracy = 0.975 (0.992)\n",
      "  Subject 7: median (max) validation accuracy = 0.971 (0.986)\n",
      "  Subject 8: median (max) validation accuracy = 0.982 (0.991)\n",
      "** Saving params with joint score = 0.976 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 22 of 50 took       40.589s\n",
      "  training loss:               0.109095\n",
      "  holdout loss (batch):        0.058877\n",
      "  Subject 1: median (max) validation accuracy = 0.975 (0.990)\n",
      "  Subject 2: median (max) validation accuracy = 0.978 (0.988)\n",
      "  Subject 3: median (max) validation accuracy = 0.978 (0.989)\n",
      "  Subject 4: median (max) validation accuracy = 0.975 (0.992)\n",
      "  Subject 5: median (max) validation accuracy = 0.976 (0.990)\n",
      "  Subject 6: median (max) validation accuracy = 0.976 (0.992)\n",
      "  Subject 7: median (max) validation accuracy = 0.972 (0.987)\n",
      "  Subject 8: median (max) validation accuracy = 0.983 (0.991)\n",
      "** Saving params with joint score = 0.977 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 23 of 50 took       40.593s\n",
      "  training loss:               0.106351\n",
      "  holdout loss (batch):        0.063891\n",
      "  Subject 1: median (max) validation accuracy = 0.976 (0.990)\n",
      "  Subject 2: median (max) validation accuracy = 0.978 (0.988)\n",
      "  Subject 3: median (max) validation accuracy = 0.979 (0.989)\n",
      "  Subject 4: median (max) validation accuracy = 0.976 (0.992)\n",
      "  Subject 5: median (max) validation accuracy = 0.976 (0.990)\n",
      "  Subject 6: median (max) validation accuracy = 0.976 (0.992)\n",
      "  Subject 7: median (max) validation accuracy = 0.972 (0.987)\n",
      "  Subject 8: median (max) validation accuracy = 0.983 (0.992)\n",
      "** Saving params with joint score = 0.977 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 24 of 50 took       40.609s\n",
      "  training loss:               0.104260\n",
      "  holdout loss (batch):        0.057070\n",
      "  Subject 1: median (max) validation accuracy = 0.977 (0.991)\n",
      "  Subject 2: median (max) validation accuracy = 0.980 (0.989)\n",
      "  Subject 3: median (max) validation accuracy = 0.980 (0.989)\n",
      "  Subject 4: median (max) validation accuracy = 0.976 (0.992)\n",
      "  Subject 5: median (max) validation accuracy = 0.977 (0.990)\n",
      "  Subject 6: median (max) validation accuracy = 0.977 (0.993)\n",
      "  Subject 7: median (max) validation accuracy = 0.973 (0.987)\n",
      "  Subject 8: median (max) validation accuracy = 0.984 (0.992)\n",
      "** Saving params with joint score = 0.978 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 25 of 50 took       40.356s\n",
      "  training loss:               0.102181\n",
      "  holdout loss (batch):        0.057128\n",
      "  Subject 1: median (max) validation accuracy = 0.977 (0.990)\n",
      "  Subject 2: median (max) validation accuracy = 0.980 (0.988)\n",
      "  Subject 3: median (max) validation accuracy = 0.979 (0.989)\n",
      "  Subject 4: median (max) validation accuracy = 0.977 (0.992)\n",
      "  Subject 5: median (max) validation accuracy = 0.977 (0.990)\n",
      "  Subject 6: median (max) validation accuracy = 0.978 (0.992)\n",
      "  Subject 7: median (max) validation accuracy = 0.973 (0.988)\n",
      "  Subject 8: median (max) validation accuracy = 0.984 (0.992)\n",
      "** Saving params with joint score = 0.978 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 26 of 50 took       40.147s\n",
      "  training loss:               0.101033\n",
      "  holdout loss (batch):        0.057708\n",
      "  Subject 1: median (max) validation accuracy = 0.978 (0.991)\n",
      "  Subject 2: median (max) validation accuracy = 0.981 (0.989)\n",
      "  Subject 3: median (max) validation accuracy = 0.981 (0.990)\n",
      "  Subject 4: median (max) validation accuracy = 0.978 (0.992)\n",
      "  Subject 5: median (max) validation accuracy = 0.978 (0.991)\n",
      "  Subject 6: median (max) validation accuracy = 0.979 (0.993)\n",
      "  Subject 7: median (max) validation accuracy = 0.974 (0.988)\n",
      "  Subject 8: median (max) validation accuracy = 0.984 (0.992)\n",
      "** Saving params with joint score = 0.979 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 27 of 50 took       40.210s\n",
      "  training loss:               0.099863\n",
      "  holdout loss (batch):        0.063600\n",
      "  Subject 1: median (max) validation accuracy = 0.979 (0.991)\n",
      "  Subject 2: median (max) validation accuracy = 0.980 (0.989)\n",
      "  Subject 3: median (max) validation accuracy = 0.981 (0.990)\n",
      "  Subject 4: median (max) validation accuracy = 0.978 (0.992)\n",
      "  Subject 5: median (max) validation accuracy = 0.978 (0.991)\n",
      "  Subject 6: median (max) validation accuracy = 0.979 (0.993)\n",
      "  Subject 7: median (max) validation accuracy = 0.975 (0.988)\n",
      "  Subject 8: median (max) validation accuracy = 0.985 (0.992)\n",
      "** Saving params with joint score = 0.979 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 28 of 50 took       40.254s\n",
      "  training loss:               0.098433\n",
      "  holdout loss (batch):        0.059563\n",
      "  Subject 1: median (max) validation accuracy = 0.979 (0.991)\n",
      "  Subject 2: median (max) validation accuracy = 0.981 (0.989)\n",
      "  Subject 3: median (max) validation accuracy = 0.981 (0.990)\n",
      "  Subject 4: median (max) validation accuracy = 0.978 (0.993)\n",
      "  Subject 5: median (max) validation accuracy = 0.979 (0.991)\n",
      "  Subject 6: median (max) validation accuracy = 0.979 (0.993)\n",
      "  Subject 7: median (max) validation accuracy = 0.975 (0.988)\n",
      "  Subject 8: median (max) validation accuracy = 0.985 (0.992)\n",
      "** Saving params with joint score = 0.980 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 29 of 50 took       40.311s\n",
      "  training loss:               0.097432\n",
      "  holdout loss (batch):        0.062247\n",
      "  Subject 1: median (max) validation accuracy = 0.980 (0.991)\n",
      "  Subject 2: median (max) validation accuracy = 0.982 (0.990)\n",
      "  Subject 3: median (max) validation accuracy = 0.982 (0.990)\n",
      "  Subject 4: median (max) validation accuracy = 0.978 (0.992)\n",
      "  Subject 5: median (max) validation accuracy = 0.979 (0.991)\n",
      "  Subject 6: median (max) validation accuracy = 0.979 (0.993)\n",
      "  Subject 7: median (max) validation accuracy = 0.976 (0.988)\n",
      "  Subject 8: median (max) validation accuracy = 0.985 (0.992)\n",
      "** Saving params with joint score = 0.980 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 30 of 50 took       40.187s\n",
      "  training loss:               0.096055\n",
      "  holdout loss (batch):        0.062245\n",
      "  Subject 1: median (max) validation accuracy = 0.980 (0.991)\n",
      "  Subject 2: median (max) validation accuracy = 0.982 (0.990)\n",
      "  Subject 3: median (max) validation accuracy = 0.982 (0.990)\n",
      "  Subject 4: median (max) validation accuracy = 0.979 (0.993)\n",
      "  Subject 5: median (max) validation accuracy = 0.979 (0.991)\n",
      "  Subject 6: median (max) validation accuracy = 0.979 (0.993)\n",
      "  Subject 7: median (max) validation accuracy = 0.976 (0.989)\n",
      "  Subject 8: median (max) validation accuracy = 0.985 (0.992)\n",
      "** Saving params with joint score = 0.980 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 31 of 50 took       40.055s\n",
      "  training loss:               0.094705\n",
      "  holdout loss (batch):        0.053629\n",
      "  Subject 1: median (max) validation accuracy = 0.980 (0.991)\n",
      "  Subject 2: median (max) validation accuracy = 0.982 (0.990)\n",
      "  Subject 3: median (max) validation accuracy = 0.982 (0.990)\n",
      "  Subject 4: median (max) validation accuracy = 0.979 (0.993)\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "  Subject 5: median (max) validation accuracy = 0.980 (0.991)\n",
      "  Subject 6: median (max) validation accuracy = 0.980 (0.993)\n",
      "  Subject 7: median (max) validation accuracy = 0.976 (0.989)\n",
      "  Subject 8: median (max) validation accuracy = 0.985 (0.992)\n",
      "** Saving params with joint score = 0.980 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 32 of 50 took       39.944s\n",
      "  training loss:               0.093808\n",
      "  holdout loss (batch):        0.058168\n",
      "  Subject 1: median (max) validation accuracy = 0.980 (0.991)\n",
      "  Subject 2: median (max) validation accuracy = 0.982 (0.990)\n",
      "  Subject 3: median (max) validation accuracy = 0.982 (0.990)\n",
      "  Subject 4: median (max) validation accuracy = 0.979 (0.993)\n",
      "  Subject 5: median (max) validation accuracy = 0.980 (0.992)\n",
      "  Subject 6: median (max) validation accuracy = 0.979 (0.993)\n",
      "  Subject 7: median (max) validation accuracy = 0.976 (0.989)\n",
      "  Subject 8: median (max) validation accuracy = 0.985 (0.992)\n",
      "** Saving params with joint score = 0.981 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 33 of 50 took       39.932s\n",
      "  training loss:               0.093437\n",
      "  holdout loss (batch):        0.050697\n",
      "  Subject 1: median (max) validation accuracy = 0.981 (0.992)\n",
      "  Subject 2: median (max) validation accuracy = 0.983 (0.991)\n",
      "  Subject 3: median (max) validation accuracy = 0.983 (0.991)\n",
      "  Subject 4: median (max) validation accuracy = 0.980 (0.993)\n",
      "  Subject 5: median (max) validation accuracy = 0.981 (0.992)\n",
      "  Subject 6: median (max) validation accuracy = 0.981 (0.994)\n",
      "  Subject 7: median (max) validation accuracy = 0.978 (0.990)\n",
      "  Subject 8: median (max) validation accuracy = 0.986 (0.993)\n",
      "** Saving params with joint score = 0.982 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 34 of 50 took       39.932s\n",
      "  training loss:               0.092366\n",
      "  holdout loss (batch):        0.052799\n",
      "  Subject 1: median (max) validation accuracy = 0.981 (0.992)\n",
      "  Subject 2: median (max) validation accuracy = 0.982 (0.991)\n",
      "  Subject 3: median (max) validation accuracy = 0.982 (0.991)\n",
      "  Subject 4: median (max) validation accuracy = 0.980 (0.993)\n",
      "  Subject 5: median (max) validation accuracy = 0.980 (0.992)\n",
      "  Subject 6: median (max) validation accuracy = 0.981 (0.994)\n",
      "  Subject 7: median (max) validation accuracy = 0.978 (0.989)\n",
      "  Subject 8: median (max) validation accuracy = 0.986 (0.993)\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 35 of 50 took       40.061s\n",
      "  training loss:               0.091131\n",
      "  holdout loss (batch):        0.051529\n",
      "  Subject 1: median (max) validation accuracy = 0.981 (0.992)\n",
      "  Subject 2: median (max) validation accuracy = 0.983 (0.991)\n",
      "  Subject 3: median (max) validation accuracy = 0.982 (0.991)\n",
      "  Subject 4: median (max) validation accuracy = 0.980 (0.993)\n",
      "  Subject 5: median (max) validation accuracy = 0.980 (0.992)\n",
      "  Subject 6: median (max) validation accuracy = 0.981 (0.994)\n",
      "  Subject 7: median (max) validation accuracy = 0.978 (0.989)\n",
      "  Subject 8: median (max) validation accuracy = 0.986 (0.993)\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 36 of 50 took       40.616s\n",
      "  training loss:               0.091081\n",
      "  holdout loss (batch):        0.056464\n",
      "  Subject 1: median (max) validation accuracy = 0.982 (0.992)\n",
      "  Subject 2: median (max) validation accuracy = 0.983 (0.991)\n",
      "  Subject 3: median (max) validation accuracy = 0.983 (0.991)\n",
      "  Subject 4: median (max) validation accuracy = 0.981 (0.993)\n",
      "  Subject 5: median (max) validation accuracy = 0.981 (0.992)\n",
      "  Subject 6: median (max) validation accuracy = 0.981 (0.994)\n",
      "  Subject 7: median (max) validation accuracy = 0.978 (0.990)\n",
      "  Subject 8: median (max) validation accuracy = 0.986 (0.993)\n",
      "** Saving params with joint score = 0.982 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 37 of 50 took       40.623s\n",
      "  training loss:               0.089489\n",
      "  holdout loss (batch):        0.056933\n",
      "  Subject 1: median (max) validation accuracy = 0.981 (0.992)\n",
      "  Subject 2: median (max) validation accuracy = 0.983 (0.990)\n",
      "  Subject 3: median (max) validation accuracy = 0.983 (0.991)\n",
      "  Subject 4: median (max) validation accuracy = 0.980 (0.993)\n",
      "  Subject 5: median (max) validation accuracy = 0.980 (0.992)\n",
      "  Subject 6: median (max) validation accuracy = 0.981 (0.993)\n",
      "  Subject 7: median (max) validation accuracy = 0.977 (0.989)\n",
      "  Subject 8: median (max) validation accuracy = 0.986 (0.992)\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 38 of 50 took       40.100s\n",
      "  training loss:               0.088885\n",
      "  holdout loss (batch):        0.065696\n",
      "  Subject 1: median (max) validation accuracy = 0.981 (0.991)\n",
      "  Subject 2: median (max) validation accuracy = 0.984 (0.991)\n",
      "  Subject 3: median (max) validation accuracy = 0.983 (0.991)\n",
      "  Subject 4: median (max) validation accuracy = 0.980 (0.993)\n",
      "  Subject 5: median (max) validation accuracy = 0.980 (0.992)\n",
      "  Subject 6: median (max) validation accuracy = 0.981 (0.994)\n",
      "  Subject 7: median (max) validation accuracy = 0.977 (0.989)\n",
      "  Subject 8: median (max) validation accuracy = 0.986 (0.992)\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 39 of 50 took       39.904s\n",
      "  training loss:               0.088400\n",
      "  holdout loss (batch):        0.051293\n",
      "  Subject 1: median (max) validation accuracy = 0.982 (0.993)\n",
      "  Subject 2: median (max) validation accuracy = 0.984 (0.991)\n",
      "  Subject 3: median (max) validation accuracy = 0.983 (0.991)\n",
      "  Subject 4: median (max) validation accuracy = 0.981 (0.993)\n",
      "  Subject 5: median (max) validation accuracy = 0.981 (0.992)\n",
      "  Subject 6: median (max) validation accuracy = 0.981 (0.994)\n",
      "  Subject 7: median (max) validation accuracy = 0.978 (0.990)\n",
      "  Subject 8: median (max) validation accuracy = 0.987 (0.993)\n",
      "** Saving params with joint score = 0.982 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 40 of 50 took       39.918s\n",
      "  training loss:               0.087785\n",
      "  holdout loss (batch):        0.055840\n",
      "  Subject 1: median (max) validation accuracy = 0.982 (0.992)\n",
      "  Subject 2: median (max) validation accuracy = 0.984 (0.991)\n",
      "  Subject 3: median (max) validation accuracy = 0.984 (0.991)\n",
      "  Subject 4: median (max) validation accuracy = 0.981 (0.993)\n",
      "  Subject 5: median (max) validation accuracy = 0.982 (0.992)\n",
      "  Subject 6: median (max) validation accuracy = 0.982 (0.994)\n",
      "  Subject 7: median (max) validation accuracy = 0.979 (0.990)\n",
      "  Subject 8: median (max) validation accuracy = 0.987 (0.993)\n",
      "** Saving params with joint score = 0.983 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 41 of 50 took       40.332s\n",
      "  training loss:               0.087043\n",
      "  holdout loss (batch):        0.051650\n",
      "  Subject 1: median (max) validation accuracy = 0.983 (0.992)\n",
      "  Subject 2: median (max) validation accuracy = 0.985 (0.991)\n",
      "  Subject 3: median (max) validation accuracy = 0.984 (0.991)\n",
      "  Subject 4: median (max) validation accuracy = 0.981 (0.994)\n",
      "  Subject 5: median (max) validation accuracy = 0.982 (0.992)\n",
      "  Subject 6: median (max) validation accuracy = 0.982 (0.994)\n",
      "  Subject 7: median (max) validation accuracy = 0.979 (0.990)\n",
      "  Subject 8: median (max) validation accuracy = 0.987 (0.993)\n",
      "** Saving params with joint score = 0.983 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 42 of 50 took       40.247s\n",
      "  training loss:               0.087208\n",
      "  holdout loss (batch):        0.055647\n",
      "  Subject 1: median (max) validation accuracy = 0.983 (0.993)\n",
      "  Subject 2: median (max) validation accuracy = 0.984 (0.992)\n",
      "  Subject 3: median (max) validation accuracy = 0.984 (0.991)\n",
      "  Subject 4: median (max) validation accuracy = 0.981 (0.993)\n",
      "  Subject 5: median (max) validation accuracy = 0.982 (0.992)\n",
      "  Subject 6: median (max) validation accuracy = 0.982 (0.994)\n",
      "  Subject 7: median (max) validation accuracy = 0.979 (0.990)\n",
      "  Subject 8: median (max) validation accuracy = 0.987 (0.993)\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 43 of 50 took       40.150s\n",
      "  training loss:               0.086181\n",
      "  holdout loss (batch):        0.050565\n",
      "  Subject 1: median (max) validation accuracy = 0.983 (0.993)\n",
      "  Subject 2: median (max) validation accuracy = 0.985 (0.991)\n",
      "  Subject 3: median (max) validation accuracy = 0.984 (0.992)\n",
      "  Subject 4: median (max) validation accuracy = 0.981 (0.994)\n",
      "  Subject 5: median (max) validation accuracy = 0.982 (0.992)\n",
      "  Subject 6: median (max) validation accuracy = 0.982 (0.994)\n",
      "  Subject 7: median (max) validation accuracy = 0.979 (0.990)\n",
      "  Subject 8: median (max) validation accuracy = 0.987 (0.993)\n",
      "** Saving params with joint score = 0.983 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 44 of 50 took       40.179s\n",
      "  training loss:               0.085759\n",
      "  holdout loss (batch):        0.050213\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "  Subject 1: median (max) validation accuracy = 0.983 (0.992)\n",
      "  Subject 2: median (max) validation accuracy = 0.985 (0.991)\n",
      "  Subject 3: median (max) validation accuracy = 0.984 (0.991)\n",
      "  Subject 4: median (max) validation accuracy = 0.982 (0.993)\n",
      "  Subject 5: median (max) validation accuracy = 0.982 (0.993)\n",
      "  Subject 6: median (max) validation accuracy = 0.983 (0.994)\n",
      "  Subject 7: median (max) validation accuracy = 0.979 (0.991)\n",
      "  Subject 8: median (max) validation accuracy = 0.987 (0.993)\n",
      "** Saving params with joint score = 0.983 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 45 of 50 took       39.975s\n",
      "  training loss:               0.084923\n",
      "  holdout loss (batch):        0.049339\n",
      "  Subject 1: median (max) validation accuracy = 0.983 (0.992)\n",
      "  Subject 2: median (max) validation accuracy = 0.984 (0.991)\n",
      "  Subject 3: median (max) validation accuracy = 0.984 (0.991)\n",
      "  Subject 4: median (max) validation accuracy = 0.981 (0.993)\n",
      "  Subject 5: median (max) validation accuracy = 0.982 (0.993)\n",
      "  Subject 6: median (max) validation accuracy = 0.982 (0.994)\n",
      "  Subject 7: median (max) validation accuracy = 0.979 (0.990)\n",
      "  Subject 8: median (max) validation accuracy = 0.987 (0.993)\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 46 of 50 took       39.877s\n",
      "  training loss:               0.084999\n",
      "  holdout loss (batch):        0.045745\n",
      "  Subject 1: median (max) validation accuracy = 0.983 (0.993)\n",
      "  Subject 2: median (max) validation accuracy = 0.985 (0.992)\n",
      "  Subject 3: median (max) validation accuracy = 0.985 (0.992)\n",
      "  Subject 4: median (max) validation accuracy = 0.982 (0.994)\n",
      "  Subject 5: median (max) validation accuracy = 0.982 (0.993)\n",
      "  Subject 6: median (max) validation accuracy = 0.983 (0.994)\n",
      "  Subject 7: median (max) validation accuracy = 0.980 (0.991)\n",
      "  Subject 8: median (max) validation accuracy = 0.988 (0.993)\n",
      "** Saving params with joint score = 0.984 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 47 of 50 took       39.943s\n",
      "  training loss:               0.084219\n",
      "  holdout loss (batch):        0.046362\n",
      "  Subject 1: median (max) validation accuracy = 0.983 (0.993)\n",
      "  Subject 2: median (max) validation accuracy = 0.986 (0.992)\n",
      "  Subject 3: median (max) validation accuracy = 0.984 (0.992)\n",
      "  Subject 4: median (max) validation accuracy = 0.982 (0.994)\n",
      "  Subject 5: median (max) validation accuracy = 0.983 (0.993)\n",
      "  Subject 6: median (max) validation accuracy = 0.983 (0.994)\n",
      "  Subject 7: median (max) validation accuracy = 0.980 (0.990)\n",
      "  Subject 8: median (max) validation accuracy = 0.987 (0.993)\n",
      "** Saving params with joint score = 0.984 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 48 of 50 took       39.851s\n",
      "  training loss:               0.083924\n",
      "  holdout loss (batch):        0.049886\n",
      "  Subject 1: median (max) validation accuracy = 0.983 (0.992)\n",
      "  Subject 2: median (max) validation accuracy = 0.985 (0.992)\n",
      "  Subject 3: median (max) validation accuracy = 0.985 (0.992)\n",
      "  Subject 4: median (max) validation accuracy = 0.982 (0.993)\n",
      "  Subject 5: median (max) validation accuracy = 0.982 (0.992)\n",
      "  Subject 6: median (max) validation accuracy = 0.983 (0.994)\n",
      "  Subject 7: median (max) validation accuracy = 0.980 (0.991)\n",
      "  Subject 8: median (max) validation accuracy = 0.987 (0.993)\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 49 of 50 took       39.812s\n",
      "  training loss:               0.083105\n",
      "  holdout loss (batch):        0.055037\n",
      "  Subject 1: median (max) validation accuracy = 0.983 (0.993)\n",
      "  Subject 2: median (max) validation accuracy = 0.985 (0.991)\n",
      "  Subject 3: median (max) validation accuracy = 0.985 (0.992)\n",
      "  Subject 4: median (max) validation accuracy = 0.982 (0.994)\n",
      "  Subject 5: median (max) validation accuracy = 0.983 (0.993)\n",
      "  Subject 6: median (max) validation accuracy = 0.983 (0.994)\n",
      "  Subject 7: median (max) validation accuracy = 0.980 (0.990)\n",
      "  Subject 8: median (max) validation accuracy = 0.988 (0.993)\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 50 of 50 took       39.947s\n",
      "  training loss:               0.082924\n",
      "  holdout loss (batch):        0.047833\n",
      "  Subject 1: median (max) validation accuracy = 0.984 (0.993)\n",
      "  Subject 2: median (max) validation accuracy = 0.985 (0.992)\n",
      "  Subject 3: median (max) validation accuracy = 0.985 (0.992)\n",
      "  Subject 4: median (max) validation accuracy = 0.983 (0.994)\n",
      "  Subject 5: median (max) validation accuracy = 0.983 (0.993)\n",
      "  Subject 6: median (max) validation accuracy = 0.983 (0.994)\n",
      "  Subject 7: median (max) validation accuracy = 0.980 (0.991)\n",
      "  Subject 8: median (max) validation accuracy = 0.987 (0.993)\n",
      "** Saving params with joint score = 0.984 **\n",
      "\n",
      "***************************************\n",
      "best joint val cc = 0.985\n",
      "subject 1: val cc = 0.985\n",
      "subject 2: val cc = 0.987\n",
      "subject 3: val cc = 0.985\n",
      "subject 4: val cc = 0.983\n",
      "subject 5: val cc = 0.985\n",
      "subject 6: val cc = 0.985\n",
      "subject 7: val cc = 0.983\n",
      "subject 8: val cc = 0.989\n",
      "***************************************\n"
     ]
    }
   ],
   "source": [
    "batch_size = 100\n",
    "num_epochs = 50\n",
    "holdout_frac = .1\n",
    "\n",
    "\n",
    "#subject_image_pred = {s: np.zeros(shape=(len(image_data[s]), len(r)), dtype=np.float32) for s,r in voxel_roi.items()}\n",
    "for roi_name, roi_masks in block_voxel_masks.items(): \n",
    "    \n",
    "    roi_nv = {s: np.sum(roi_mask) for s, roi_mask in roi_masks.items()}\n",
    "    \n",
    "    print ('***************************************')\n",
    "    print ('***        %12s       ***'%roi_name)\n",
    "    print ({s: np.sum(m) for s,m in roi_masks.items()})\n",
    "    print ('***************************************') \n",
    "    \n",
    "    # create model\n",
    "    shared_model = Encoder(trn_stim_mean, trunk_width=N).to(device)\n",
    "    rec, fmaps, h = shared_model(T.from_numpy(stim_data[subjects[0]][:20]).to(device))\n",
    "\n",
    "    subject_fwrfs = {s: Torch_LayerwiseFWRF(fmaps, nv=roi_nv[s], pre_nl=_log_act_fn, \\\n",
    "                     post_nl=_log_act_fn, dtype=np.float32).to(device) for s in subjects}    \n",
    "    \n",
    "    ##################################\n",
    "    try:\n",
    "        from torch.hub import load_state_dict_from_url\n",
    "    except ImportError:\n",
    "        from torch.utils.model_zoo import load_url as load_state_dict_from_url\n",
    "\n",
    "    state_dict = load_state_dict_from_url('https://download.pytorch.org/models/alexnet-owt-4df8aa71.pth', progress=True)\n",
    "            ### Rename dictionary keys to match new breakdown\n",
    "    pre_state_dict = {}\n",
    "    pre_state_dict['conv1.0.weight'] = state_dict.pop('features.0.weight')\n",
    "    pre_state_dict['conv1.0.bias'] = state_dict.pop('features.0.bias')\n",
    "    pre_state_dict['conv2.0.weight'] = state_dict.pop('features.3.weight')\n",
    "    pre_state_dict['conv2.0.bias'] = state_dict.pop('features.3.bias')\n",
    "\n",
    "\n",
    "    shared_model.pre.load_state_dict(pre_state_dict)\n",
    "    ###################################\n",
    "    \n",
    "    optimizer_net = optim.Adam([\n",
    "            #{'params': shared_model.pre.parameters()},\n",
    "            {'params': shared_model.enc.parameters()},\n",
    "        ], lr=1e-3, betas=(0.9, 0.999), eps=1e-08)\n",
    "    subject_optimizer = {s: optim.Adam([\n",
    "            {'params': sp.parameters()}\n",
    "        ], lr=1e-3, betas=(0.9, 0.999), eps=1e-08) for s,sp in subject_fwrfs.items()}\n",
    "    subject_opts = {s: [optimizer_net, subject_optimizer[s]] for s in subject_optimizer.keys()}   \n",
    "    ###################################\n",
    "    \n",
    "\n",
    "    best_params, final_params, hold_cc_hist, hold_hist, trn_hist, best_epoch, best_joint_cc_score = \\\n",
    "        learn_params_(_training_fn, _holdout_fn, _pred_fn, shared_model, subject_fwrfs, subject_opts, \n",
    "            stim_data, trn_voxel_data, trn_stim_ordering,\n",
    "            num_epochs=num_epochs, batch_size=batch_size, holdout_frac=holdout_frac, masks=roi_masks, randomize=False)\n",
    "    \n",
    "    \n",
    "    ###################################\n",
    "    #val_voxel = {s: val_voxel_data[s] for s in val_voxel_data.keys()}\n",
    "    shared_model.load_state_dict(best_params['enc'])\n",
    "    shared_model.eval() \n",
    "    for s,sd in subject_fwrfs.items():\n",
    "        sd.load_state_dict(best_params['fwrfs'][s])\n",
    "        sd.eval() \n",
    "\n",
    "    subject_val_cc = validation_(_pred_fn, shared_model, subject_fwrfs, stim_data, val_voxel_data, val_stim_ordering, batch_size, masks=roi_masks)\n",
    "    joined_val_cc = np.concatenate(list(subject_val_cc.values()), axis=0)\n",
    "    \n",
    "    print ('***************************************')\n",
    "    print (\"best joint val cc = %.3f\"% np.median(joined_val_cc))\n",
    "    for s,v in subject_val_cc.items():\n",
    "        print (\"subject %s: val cc = %.3f\"%(s, np.median(v)))\n",
    " \n",
    "    print ('***************************************')\n",
    "    ###################################\n",
    "    torch.save({\n",
    "            'group_name': roi_name,\n",
    "            'num_epochs': num_epochs,\n",
    "            'batch_size': batch_size,\n",
    "            'holdout_frac': holdout_frac,   \n",
    "            'best_params': best_params, \n",
    "            'final_params': final_params,\n",
    "            'trn_loss_history': trn_hist,\n",
    "            'hold_loss_history': hold_hist,\n",
    "            'hold_cc_history': hold_cc_hist,\n",
    "            'best_epoch': best_epoch,\n",
    "            'best_joint_cc_score': best_joint_cc_score,\n",
    "            'val_cc': subject_val_cc,\n",
    "            'input_mean': trn_stim_mean,\n",
    "            'voxel_roi': voxel_roi, \n",
    "            'voxel_mask': voxel_mask,\n",
    "            'voxel_index': voxel_idx,\n",
    "            'group_mask': roi_masks\n",
    "            }, output_dir+'model_params_%s_%s'% (roi_name, noise_model))\n",
    "    \n",
    "#    ######\n",
    "#    image_pred[roi_name] = {}\n",
    "#    for s,sd in subject_fwrfs.items():\n",
    "#        subject_image_pred[s][:,roi_mask] = subject_pred_pass(_pred_fn, shared_model, sd, stim_data[s], batch_size)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Compare joined vs ROI-wise prediction accuracy"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 43,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "dict_keys(['num_epochs', 'batch_size', 'holdout_frac', 'best_params', 'final_params', 'trn_loss_history', 'hold_loss_history', 'hold_cc_history', 'best_epoch', 'best_joint_cc_score', 'val_cc', 'input_mean', 'voxel_roi', 'voxel_mask', 'voxel_index'])\n"
     ]
    }
   ],
   "source": [
    "joined_timestamp_dir = 'model_params_%s'%noise_model\n",
    "\n",
    "joined_checkpoint = torch.load(output_dir + joined_timestamp_dir)\n",
    "print (joined_checkpoint.keys())"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 44,
   "metadata": {},
   "outputs": [],
   "source": [
    "gnetfj_val_cc = joined_checkpoint['val_cc']"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 45,
   "metadata": {},
   "outputs": [],
   "source": [
    "remodel_dir = output_dir\n",
    "roi_timestamp_dir = {\n",
    "    'V1': '%s/model_params_V1_%s'%(remodel_dir, noise_model),\n",
    "    'V2': '%s/model_params_V2_%s'%(remodel_dir, noise_model),\n",
    "    'V3': '%s/model_params_V3_%s'%(remodel_dir, noise_model),\n",
    "    'hV4': '%s/model_params_hV4_%s'%(remodel_dir, noise_model),\n",
    "    }"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 46,
   "metadata": {},
   "outputs": [],
   "source": [
    "roi_checkpoint = {}\n",
    "for r,t in roi_timestamp_dir.items():\n",
    "    roi_checkpoint[r] = torch.load(t)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 47,
   "metadata": {},
   "outputs": [],
   "source": [
    "gnetfr_val_cc = {s: np.zeros_like(v) for s,v in gnetfj_val_cc.items()}\n",
    "for k,(roi_name, cp) in enumerate(roi_checkpoint.items()):\n",
    "    for s in gnetfj_val_cc.keys():\n",
    "        \n",
    "        group_mask = cp['group_mask'][s]\n",
    "        gnetfr_val_cc[s][group_mask] = cp['val_cc'][s]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 48,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/usr/local/lib/python3.6/site-packages/seaborn/_decorators.py:43: FutureWarning: Pass the following variables as keyword args: x, y. From version 0.12, the only valid positional argument will be `data`, and passing other arguments without an explicit keyword will result in an error or misinterpretation.\n",
      "  FutureWarning\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "Text(0.5, 40.49999999999994, 'joined gnet8 <--> roiwise gnet8')"
      ]
     },
     "execution_count": 48,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "text/plain": [
       "<Figure size 576x576 with 0 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 576x576 with 3 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "from src.plots import plot_fwrf_paper_compare\n",
    "\n",
    "x = np.concatenate([v for s,v in gnetfj_val_cc.items()])\n",
    "y = np.concatenate([v for s,v in gnetfr_val_cc.items()])\n",
    "\n",
    "fig2b=plt.figure(figsize=(8, 8))\n",
    "fig2b=plot_fwrf_paper_compare(x, y, 0.055, [-.2, .2], [-.1, 1.])\n",
    "ax=fig2b.ax_joint\n",
    "ax.text(x=-.2+.01, y=.84, s='med = %.3f'%np.median(np.nan_to_num(y)), fontsize=18)\n",
    "ax.text(x=-.2+.01, y=.77, s='max = %.3f'%np.max(np.nan_to_num(y)), fontsize=18)\n",
    "ax.text(x=.2 - .13, y=.84, s='med = %.3f'%np.median(np.nan_to_num(x)), fontsize=18)\n",
    "ax.text(x=.2 - .13, y=.77, s='max = %.3f'%np.max(np.nan_to_num(x)), fontsize=18)\n",
    "plt.xlabel('joined gnet8 <--> roiwise gnet8')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 49,
   "metadata": {},
   "outputs": [],
   "source": [
    "filename = output_dir + 'model_advantage_gnet8j_v_gnet8r_fake_%s%s'  % (noise_model, saveext)\n",
    "fig2b.savefig(filename, **savearg)\n",
    "plt.close()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Cross-ROI model "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 50,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "/export/raid1/home/styvesg/code/nsd_gnet8x/output/multisubject/gnet8r_pred_Feb-07-2022_1721/V1/\n",
      "***************************************\n",
      "***                  V1       ***\n",
      "from ==> /export/raid1/home/styvesg/code/nsd_gnet8x/output/multisubject/gnet8r_pred_Feb-07-2022_1721//model_params_V1_no_noise\n",
      "to   ==> /export/raid1/home/styvesg/code/nsd_gnet8x/output/multisubject/gnet8r_pred_Feb-07-2022_1721/V1/\n",
      "***************************************\n",
      "subject 1 masked 4627 of 4627\n",
      "subject 1 training/holdout 8100 900\n",
      "subject 2 masked 3765 of 3765\n",
      "subject 2 training/holdout 8100 900\n",
      "subject 3 masked 3778 of 3778\n",
      "subject 3 training/holdout 8100 900\n",
      "subject 4 masked 3027 of 3027\n",
      "subject 4 training/holdout 8100 900\n",
      "subject 5 masked 3685 of 3685\n",
      "subject 5 training/holdout 8100 900\n",
      "subject 6 masked 4024 of 4024\n",
      "subject 6 training/holdout 8100 900\n",
      "subject 7 masked 3201 of 3201\n",
      "subject 7 training/holdout 8100 900\n",
      "subject 8 masked 3490 of 3490\n",
      "subject 8 training/holdout 8100 900\n",
      "\\ : 100.0 %\n",
      "  Epoch 1 of 50 took       40.793s\n",
      "  training loss:               0.585621\n",
      "  holdout loss (batch):        0.431161\n",
      "  Subject 1: median (max) validation accuracy = 0.667 (0.940)\n",
      "  Subject 2: median (max) validation accuracy = 0.681 (0.941)\n",
      "  Subject 3: median (max) validation accuracy = 0.772 (0.957)\n",
      "  Subject 4: median (max) validation accuracy = 0.790 (0.954)\n",
      "  Subject 5: median (max) validation accuracy = 0.862 (0.956)\n",
      "  Subject 6: median (max) validation accuracy = 0.781 (0.950)\n",
      "  Subject 7: median (max) validation accuracy = 0.762 (0.939)\n",
      "  Subject 8: median (max) validation accuracy = 0.819 (0.949)\n",
      "** Saving params with joint score = 0.768 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 2 of 50 took       40.787s\n",
      "  training loss:               0.409979\n",
      "  holdout loss (batch):        0.352577\n",
      "  Subject 1: median (max) validation accuracy = 0.727 (0.958)\n",
      "  Subject 2: median (max) validation accuracy = 0.741 (0.961)\n",
      "  Subject 3: median (max) validation accuracy = 0.820 (0.967)\n",
      "  Subject 4: median (max) validation accuracy = 0.835 (0.965)\n",
      "  Subject 5: median (max) validation accuracy = 0.895 (0.972)\n",
      "  Subject 6: median (max) validation accuracy = 0.828 (0.966)\n",
      "  Subject 7: median (max) validation accuracy = 0.816 (0.957)\n",
      "  Subject 8: median (max) validation accuracy = 0.861 (0.964)\n",
      "** Saving params with joint score = 0.818 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 3 of 50 took       40.874s\n",
      "  training loss:               0.352875\n",
      "  holdout loss (batch):        0.305581\n",
      "  Subject 1: median (max) validation accuracy = 0.762 (0.963)\n",
      "  Subject 2: median (max) validation accuracy = 0.779 (0.968)\n",
      "  Subject 3: median (max) validation accuracy = 0.848 (0.972)\n",
      "  Subject 4: median (max) validation accuracy = 0.861 (0.970)\n",
      "  Subject 5: median (max) validation accuracy = 0.912 (0.978)\n",
      "  Subject 6: median (max) validation accuracy = 0.854 (0.975)\n",
      "  Subject 7: median (max) validation accuracy = 0.847 (0.968)\n",
      "  Subject 8: median (max) validation accuracy = 0.884 (0.973)\n",
      "** Saving params with joint score = 0.847 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 4 of 50 took       40.884s\n",
      "  training loss:               0.313477\n",
      "  holdout loss (batch):        0.273697\n",
      "  Subject 1: median (max) validation accuracy = 0.791 (0.971)\n",
      "  Subject 2: median (max) validation accuracy = 0.808 (0.975)\n",
      "  Subject 3: median (max) validation accuracy = 0.867 (0.976)\n",
      "  Subject 4: median (max) validation accuracy = 0.882 (0.975)\n",
      "  Subject 5: median (max) validation accuracy = 0.924 (0.980)\n",
      "  Subject 6: median (max) validation accuracy = 0.873 (0.980)\n",
      "  Subject 7: median (max) validation accuracy = 0.870 (0.974)\n",
      "  Subject 8: median (max) validation accuracy = 0.900 (0.978)\n",
      "** Saving params with joint score = 0.867 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 5 of 50 took       40.871s\n",
      "  training loss:               0.282299\n",
      "  holdout loss (batch):        0.245413\n",
      "  Subject 1: median (max) validation accuracy = 0.816 (0.976)\n",
      "  Subject 2: median (max) validation accuracy = 0.831 (0.979)\n",
      "  Subject 3: median (max) validation accuracy = 0.883 (0.979)\n",
      "  Subject 4: median (max) validation accuracy = 0.897 (0.980)\n",
      "  Subject 5: median (max) validation accuracy = 0.933 (0.982)\n",
      "  Subject 6: median (max) validation accuracy = 0.889 (0.984)\n",
      "  Subject 7: median (max) validation accuracy = 0.888 (0.978)\n",
      "  Subject 8: median (max) validation accuracy = 0.912 (0.981)\n",
      "** Saving params with joint score = 0.883 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 6 of 50 took       40.894s\n",
      "  training loss:               0.254618\n",
      "  holdout loss (batch):        0.218946\n",
      "  Subject 1: median (max) validation accuracy = 0.839 (0.979)\n",
      "  Subject 2: median (max) validation accuracy = 0.852 (0.982)\n",
      "  Subject 3: median (max) validation accuracy = 0.899 (0.981)\n",
      "  Subject 4: median (max) validation accuracy = 0.910 (0.984)\n",
      "  Subject 5: median (max) validation accuracy = 0.940 (0.986)\n",
      "  Subject 6: median (max) validation accuracy = 0.901 (0.987)\n",
      "  Subject 7: median (max) validation accuracy = 0.902 (0.982)\n",
      "  Subject 8: median (max) validation accuracy = 0.922 (0.985)\n",
      "** Saving params with joint score = 0.897 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 7 of 50 took       40.891s\n",
      "  training loss:               0.230411\n",
      "  holdout loss (batch):        0.195089\n",
      "  Subject 1: median (max) validation accuracy = 0.859 (0.981)\n",
      "  Subject 2: median (max) validation accuracy = 0.868 (0.984)\n",
      "  Subject 3: median (max) validation accuracy = 0.911 (0.984)\n",
      "  Subject 4: median (max) validation accuracy = 0.922 (0.986)\n",
      "  Subject 5: median (max) validation accuracy = 0.946 (0.988)\n",
      "  Subject 6: median (max) validation accuracy = 0.913 (0.989)\n",
      "  Subject 7: median (max) validation accuracy = 0.916 (0.984)\n",
      "  Subject 8: median (max) validation accuracy = 0.931 (0.987)\n",
      "** Saving params with joint score = 0.909 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 8 of 50 took       40.876s\n",
      "  training loss:               0.208895\n",
      "  holdout loss (batch):        0.176562\n",
      "  Subject 1: median (max) validation accuracy = 0.876 (0.983)\n",
      "  Subject 2: median (max) validation accuracy = 0.884 (0.986)\n",
      "  Subject 3: median (max) validation accuracy = 0.921 (0.985)\n",
      "  Subject 4: median (max) validation accuracy = 0.931 (0.988)\n",
      "  Subject 5: median (max) validation accuracy = 0.951 (0.989)\n",
      "  Subject 6: median (max) validation accuracy = 0.922 (0.990)\n",
      "  Subject 7: median (max) validation accuracy = 0.926 (0.985)\n",
      "  Subject 8: median (max) validation accuracy = 0.938 (0.989)\n",
      "** Saving params with joint score = 0.919 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 9 of 50 took       40.926s\n",
      "  training loss:               0.190433\n",
      "  holdout loss (batch):        0.158927\n",
      "  Subject 1: median (max) validation accuracy = 0.890 (0.984)\n",
      "  Subject 2: median (max) validation accuracy = 0.896 (0.987)\n",
      "  Subject 3: median (max) validation accuracy = 0.930 (0.987)\n",
      "  Subject 4: median (max) validation accuracy = 0.938 (0.990)\n",
      "  Subject 5: median (max) validation accuracy = 0.955 (0.991)\n",
      "  Subject 6: median (max) validation accuracy = 0.930 (0.991)\n",
      "  Subject 7: median (max) validation accuracy = 0.935 (0.987)\n",
      "  Subject 8: median (max) validation accuracy = 0.945 (0.991)\n",
      "** Saving params with joint score = 0.928 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 10 of 50 took       41.080s\n",
      "  training loss:               0.174568\n",
      "  holdout loss (batch):        0.145377\n",
      "  Subject 1: median (max) validation accuracy = 0.902 (0.985)\n",
      "  Subject 2: median (max) validation accuracy = 0.907 (0.988)\n",
      "  Subject 3: median (max) validation accuracy = 0.938 (0.988)\n",
      "  Subject 4: median (max) validation accuracy = 0.944 (0.991)\n",
      "  Subject 5: median (max) validation accuracy = 0.959 (0.992)\n",
      "  Subject 6: median (max) validation accuracy = 0.936 (0.992)\n",
      "  Subject 7: median (max) validation accuracy = 0.942 (0.988)\n",
      "  Subject 8: median (max) validation accuracy = 0.950 (0.992)\n",
      "** Saving params with joint score = 0.935 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 11 of 50 took       41.134s\n",
      "  training loss:               0.161429\n",
      "  holdout loss (batch):        0.132928\n",
      "  Subject 1: median (max) validation accuracy = 0.913 (0.987)\n",
      "  Subject 2: median (max) validation accuracy = 0.917 (0.990)\n",
      "  Subject 3: median (max) validation accuracy = 0.944 (0.990)\n",
      "  Subject 4: median (max) validation accuracy = 0.950 (0.992)\n",
      "  Subject 5: median (max) validation accuracy = 0.962 (0.993)\n",
      "  Subject 6: median (max) validation accuracy = 0.942 (0.993)\n",
      "  Subject 7: median (max) validation accuracy = 0.947 (0.989)\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "  Subject 8: median (max) validation accuracy = 0.954 (0.993)\n",
      "** Saving params with joint score = 0.940 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 12 of 50 took       41.046s\n",
      "  training loss:               0.149475\n",
      "  holdout loss (batch):        0.121714\n",
      "  Subject 1: median (max) validation accuracy = 0.922 (0.988)\n",
      "  Subject 2: median (max) validation accuracy = 0.925 (0.990)\n",
      "  Subject 3: median (max) validation accuracy = 0.950 (0.990)\n",
      "  Subject 4: median (max) validation accuracy = 0.954 (0.993)\n",
      "  Subject 5: median (max) validation accuracy = 0.963 (0.993)\n",
      "  Subject 6: median (max) validation accuracy = 0.947 (0.993)\n",
      "  Subject 7: median (max) validation accuracy = 0.952 (0.990)\n",
      "  Subject 8: median (max) validation accuracy = 0.958 (0.993)\n",
      "** Saving params with joint score = 0.946 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 13 of 50 took       40.957s\n",
      "  training loss:               0.139743\n",
      "  holdout loss (batch):        0.111357\n",
      "  Subject 1: median (max) validation accuracy = 0.930 (0.988)\n",
      "  Subject 2: median (max) validation accuracy = 0.932 (0.991)\n",
      "  Subject 3: median (max) validation accuracy = 0.955 (0.991)\n",
      "  Subject 4: median (max) validation accuracy = 0.959 (0.993)\n",
      "  Subject 5: median (max) validation accuracy = 0.966 (0.994)\n",
      "  Subject 6: median (max) validation accuracy = 0.952 (0.994)\n",
      "  Subject 7: median (max) validation accuracy = 0.957 (0.991)\n",
      "  Subject 8: median (max) validation accuracy = 0.961 (0.994)\n",
      "** Saving params with joint score = 0.950 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 14 of 50 took       40.988s\n",
      "  training loss:               0.130892\n",
      "  holdout loss (batch):        0.104854\n",
      "  Subject 1: median (max) validation accuracy = 0.937 (0.989)\n",
      "  Subject 2: median (max) validation accuracy = 0.939 (0.992)\n",
      "  Subject 3: median (max) validation accuracy = 0.959 (0.992)\n",
      "  Subject 4: median (max) validation accuracy = 0.962 (0.994)\n",
      "  Subject 5: median (max) validation accuracy = 0.967 (0.994)\n",
      "  Subject 6: median (max) validation accuracy = 0.956 (0.994)\n",
      "  Subject 7: median (max) validation accuracy = 0.961 (0.992)\n",
      "  Subject 8: median (max) validation accuracy = 0.964 (0.994)\n",
      "** Saving params with joint score = 0.955 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 15 of 50 took       40.892s\n",
      "  training loss:               0.123342\n",
      "  holdout loss (batch):        0.099328\n",
      "  Subject 1: median (max) validation accuracy = 0.943 (0.990)\n",
      "  Subject 2: median (max) validation accuracy = 0.944 (0.992)\n",
      "  Subject 3: median (max) validation accuracy = 0.962 (0.992)\n",
      "  Subject 4: median (max) validation accuracy = 0.965 (0.994)\n",
      "  Subject 5: median (max) validation accuracy = 0.968 (0.995)\n",
      "  Subject 6: median (max) validation accuracy = 0.959 (0.995)\n",
      "  Subject 7: median (max) validation accuracy = 0.964 (0.992)\n",
      "  Subject 8: median (max) validation accuracy = 0.966 (0.994)\n",
      "** Saving params with joint score = 0.958 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 16 of 50 took       40.873s\n",
      "  training loss:               0.117211\n",
      "  holdout loss (batch):        0.092374\n",
      "  Subject 1: median (max) validation accuracy = 0.949 (0.990)\n",
      "  Subject 2: median (max) validation accuracy = 0.949 (0.993)\n",
      "  Subject 3: median (max) validation accuracy = 0.965 (0.993)\n",
      "  Subject 4: median (max) validation accuracy = 0.967 (0.995)\n",
      "  Subject 5: median (max) validation accuracy = 0.969 (0.995)\n",
      "  Subject 6: median (max) validation accuracy = 0.962 (0.995)\n",
      "  Subject 7: median (max) validation accuracy = 0.966 (0.993)\n",
      "  Subject 8: median (max) validation accuracy = 0.968 (0.995)\n",
      "** Saving params with joint score = 0.961 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 17 of 50 took       40.878s\n",
      "  training loss:               0.112937\n",
      "  holdout loss (batch):        0.089833\n",
      "  Subject 1: median (max) validation accuracy = 0.952 (0.991)\n",
      "  Subject 2: median (max) validation accuracy = 0.953 (0.993)\n",
      "  Subject 3: median (max) validation accuracy = 0.967 (0.994)\n",
      "  Subject 4: median (max) validation accuracy = 0.968 (0.995)\n",
      "  Subject 5: median (max) validation accuracy = 0.970 (0.995)\n",
      "  Subject 6: median (max) validation accuracy = 0.964 (0.995)\n",
      "  Subject 7: median (max) validation accuracy = 0.968 (0.993)\n",
      "  Subject 8: median (max) validation accuracy = 0.969 (0.995)\n",
      "** Saving params with joint score = 0.963 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 18 of 50 took       40.876s\n",
      "  training loss:               0.109059\n",
      "  holdout loss (batch):        0.086268\n",
      "  Subject 1: median (max) validation accuracy = 0.955 (0.992)\n",
      "  Subject 2: median (max) validation accuracy = 0.956 (0.994)\n",
      "  Subject 3: median (max) validation accuracy = 0.968 (0.994)\n",
      "  Subject 4: median (max) validation accuracy = 0.970 (0.995)\n",
      "  Subject 5: median (max) validation accuracy = 0.971 (0.995)\n",
      "  Subject 6: median (max) validation accuracy = 0.966 (0.996)\n",
      "  Subject 7: median (max) validation accuracy = 0.970 (0.993)\n",
      "  Subject 8: median (max) validation accuracy = 0.970 (0.995)\n",
      "** Saving params with joint score = 0.965 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 19 of 50 took       40.883s\n",
      "  training loss:               0.106059\n",
      "  holdout loss (batch):        0.083544\n",
      "  Subject 1: median (max) validation accuracy = 0.958 (0.992)\n",
      "  Subject 2: median (max) validation accuracy = 0.958 (0.994)\n",
      "  Subject 3: median (max) validation accuracy = 0.970 (0.994)\n",
      "  Subject 4: median (max) validation accuracy = 0.971 (0.995)\n",
      "  Subject 5: median (max) validation accuracy = 0.971 (0.996)\n",
      "  Subject 6: median (max) validation accuracy = 0.968 (0.996)\n",
      "  Subject 7: median (max) validation accuracy = 0.971 (0.994)\n",
      "  Subject 8: median (max) validation accuracy = 0.971 (0.995)\n",
      "** Saving params with joint score = 0.966 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 20 of 50 took       40.849s\n",
      "  training loss:               0.103786\n",
      "  holdout loss (batch):        0.084130\n",
      "  Subject 1: median (max) validation accuracy = 0.959 (0.992)\n",
      "  Subject 2: median (max) validation accuracy = 0.959 (0.994)\n",
      "  Subject 3: median (max) validation accuracy = 0.971 (0.994)\n",
      "  Subject 4: median (max) validation accuracy = 0.972 (0.995)\n",
      "  Subject 5: median (max) validation accuracy = 0.971 (0.995)\n",
      "  Subject 6: median (max) validation accuracy = 0.968 (0.996)\n",
      "  Subject 7: median (max) validation accuracy = 0.971 (0.994)\n",
      "  Subject 8: median (max) validation accuracy = 0.972 (0.995)\n",
      "** Saving params with joint score = 0.967 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 21 of 50 took       40.832s\n",
      "  training loss:               0.101926\n",
      "  holdout loss (batch):        0.079019\n",
      "  Subject 1: median (max) validation accuracy = 0.961 (0.993)\n",
      "  Subject 2: median (max) validation accuracy = 0.961 (0.994)\n",
      "  Subject 3: median (max) validation accuracy = 0.972 (0.995)\n",
      "  Subject 4: median (max) validation accuracy = 0.972 (0.996)\n",
      "  Subject 5: median (max) validation accuracy = 0.972 (0.996)\n",
      "  Subject 6: median (max) validation accuracy = 0.970 (0.996)\n",
      "  Subject 7: median (max) validation accuracy = 0.973 (0.994)\n",
      "  Subject 8: median (max) validation accuracy = 0.973 (0.995)\n",
      "** Saving params with joint score = 0.969 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 22 of 50 took       40.830s\n",
      "  training loss:               0.100161\n",
      "  holdout loss (batch):        0.077667\n",
      "  Subject 1: median (max) validation accuracy = 0.963 (0.993)\n",
      "  Subject 2: median (max) validation accuracy = 0.962 (0.995)\n",
      "  Subject 3: median (max) validation accuracy = 0.972 (0.995)\n",
      "  Subject 4: median (max) validation accuracy = 0.973 (0.996)\n",
      "  Subject 5: median (max) validation accuracy = 0.973 (0.996)\n",
      "  Subject 6: median (max) validation accuracy = 0.970 (0.996)\n",
      "  Subject 7: median (max) validation accuracy = 0.973 (0.994)\n",
      "  Subject 8: median (max) validation accuracy = 0.973 (0.995)\n",
      "** Saving params with joint score = 0.969 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 23 of 50 took       40.798s\n",
      "  training loss:               0.098954\n",
      "  holdout loss (batch):        0.076359\n",
      "  Subject 1: median (max) validation accuracy = 0.964 (0.993)\n",
      "  Subject 2: median (max) validation accuracy = 0.963 (0.995)\n",
      "  Subject 3: median (max) validation accuracy = 0.973 (0.995)\n",
      "  Subject 4: median (max) validation accuracy = 0.974 (0.996)\n",
      "  Subject 5: median (max) validation accuracy = 0.973 (0.996)\n",
      "  Subject 6: median (max) validation accuracy = 0.971 (0.996)\n",
      "  Subject 7: median (max) validation accuracy = 0.974 (0.995)\n",
      "  Subject 8: median (max) validation accuracy = 0.974 (0.995)\n",
      "** Saving params with joint score = 0.970 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 24 of 50 took       40.821s\n",
      "  training loss:               0.097551\n",
      "  holdout loss (batch):        0.074525\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "  Subject 1: median (max) validation accuracy = 0.965 (0.994)\n",
      "  Subject 2: median (max) validation accuracy = 0.964 (0.995)\n",
      "  Subject 3: median (max) validation accuracy = 0.974 (0.995)\n",
      "  Subject 4: median (max) validation accuracy = 0.974 (0.996)\n",
      "  Subject 5: median (max) validation accuracy = 0.974 (0.996)\n",
      "  Subject 6: median (max) validation accuracy = 0.972 (0.996)\n",
      "  Subject 7: median (max) validation accuracy = 0.975 (0.995)\n",
      "  Subject 8: median (max) validation accuracy = 0.974 (0.996)\n",
      "** Saving params with joint score = 0.971 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 25 of 50 took       40.790s\n",
      "  training loss:               0.096190\n",
      "  holdout loss (batch):        0.073839\n",
      "  Subject 1: median (max) validation accuracy = 0.966 (0.994)\n",
      "  Subject 2: median (max) validation accuracy = 0.965 (0.995)\n",
      "  Subject 3: median (max) validation accuracy = 0.974 (0.995)\n",
      "  Subject 4: median (max) validation accuracy = 0.974 (0.996)\n",
      "  Subject 5: median (max) validation accuracy = 0.974 (0.996)\n",
      "  Subject 6: median (max) validation accuracy = 0.972 (0.996)\n",
      "  Subject 7: median (max) validation accuracy = 0.975 (0.995)\n",
      "  Subject 8: median (max) validation accuracy = 0.974 (0.996)\n",
      "** Saving params with joint score = 0.971 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 26 of 50 took       40.809s\n",
      "  training loss:               0.095318\n",
      "  holdout loss (batch):        0.071466\n",
      "  Subject 1: median (max) validation accuracy = 0.967 (0.994)\n",
      "  Subject 2: median (max) validation accuracy = 0.966 (0.995)\n",
      "  Subject 3: median (max) validation accuracy = 0.975 (0.996)\n",
      "  Subject 4: median (max) validation accuracy = 0.975 (0.996)\n",
      "  Subject 5: median (max) validation accuracy = 0.974 (0.996)\n",
      "  Subject 6: median (max) validation accuracy = 0.973 (0.997)\n",
      "  Subject 7: median (max) validation accuracy = 0.975 (0.995)\n",
      "  Subject 8: median (max) validation accuracy = 0.975 (0.996)\n",
      "** Saving params with joint score = 0.972 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 27 of 50 took       40.848s\n",
      "  training loss:               0.094793\n",
      "  holdout loss (batch):        0.073950\n",
      "  Subject 1: median (max) validation accuracy = 0.967 (0.994)\n",
      "  Subject 2: median (max) validation accuracy = 0.966 (0.995)\n",
      "  Subject 3: median (max) validation accuracy = 0.975 (0.996)\n",
      "  Subject 4: median (max) validation accuracy = 0.975 (0.996)\n",
      "  Subject 5: median (max) validation accuracy = 0.974 (0.996)\n",
      "  Subject 6: median (max) validation accuracy = 0.973 (0.996)\n",
      "  Subject 7: median (max) validation accuracy = 0.975 (0.995)\n",
      "  Subject 8: median (max) validation accuracy = 0.975 (0.996)\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 28 of 50 took       40.903s\n",
      "  training loss:               0.093640\n",
      "  holdout loss (batch):        0.071772\n",
      "  Subject 1: median (max) validation accuracy = 0.968 (0.995)\n",
      "  Subject 2: median (max) validation accuracy = 0.966 (0.995)\n",
      "  Subject 3: median (max) validation accuracy = 0.975 (0.996)\n",
      "  Subject 4: median (max) validation accuracy = 0.975 (0.996)\n",
      "  Subject 5: median (max) validation accuracy = 0.974 (0.996)\n",
      "  Subject 6: median (max) validation accuracy = 0.973 (0.997)\n",
      "  Subject 7: median (max) validation accuracy = 0.976 (0.995)\n",
      "  Subject 8: median (max) validation accuracy = 0.975 (0.996)\n",
      "** Saving params with joint score = 0.972 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 29 of 50 took       40.848s\n",
      "  training loss:               0.093425\n",
      "  holdout loss (batch):        0.071730\n",
      "  Subject 1: median (max) validation accuracy = 0.969 (0.994)\n",
      "  Subject 2: median (max) validation accuracy = 0.967 (0.995)\n",
      "  Subject 3: median (max) validation accuracy = 0.975 (0.996)\n",
      "  Subject 4: median (max) validation accuracy = 0.976 (0.996)\n",
      "  Subject 5: median (max) validation accuracy = 0.974 (0.996)\n",
      "  Subject 6: median (max) validation accuracy = 0.973 (0.997)\n",
      "  Subject 7: median (max) validation accuracy = 0.976 (0.995)\n",
      "  Subject 8: median (max) validation accuracy = 0.976 (0.996)\n",
      "** Saving params with joint score = 0.973 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 30 of 50 took       40.947s\n",
      "  training loss:               0.092598\n",
      "  holdout loss (batch):        0.070260\n",
      "  Subject 1: median (max) validation accuracy = 0.969 (0.995)\n",
      "  Subject 2: median (max) validation accuracy = 0.967 (0.995)\n",
      "  Subject 3: median (max) validation accuracy = 0.976 (0.996)\n",
      "  Subject 4: median (max) validation accuracy = 0.976 (0.996)\n",
      "  Subject 5: median (max) validation accuracy = 0.975 (0.996)\n",
      "  Subject 6: median (max) validation accuracy = 0.974 (0.997)\n",
      "  Subject 7: median (max) validation accuracy = 0.976 (0.995)\n",
      "  Subject 8: median (max) validation accuracy = 0.976 (0.996)\n",
      "** Saving params with joint score = 0.973 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 31 of 50 took       40.906s\n",
      "  training loss:               0.092113\n",
      "  holdout loss (batch):        0.068488\n",
      "  Subject 1: median (max) validation accuracy = 0.970 (0.995)\n",
      "  Subject 2: median (max) validation accuracy = 0.968 (0.995)\n",
      "  Subject 3: median (max) validation accuracy = 0.976 (0.996)\n",
      "  Subject 4: median (max) validation accuracy = 0.976 (0.996)\n",
      "  Subject 5: median (max) validation accuracy = 0.975 (0.996)\n",
      "  Subject 6: median (max) validation accuracy = 0.974 (0.997)\n",
      "  Subject 7: median (max) validation accuracy = 0.976 (0.996)\n",
      "  Subject 8: median (max) validation accuracy = 0.976 (0.996)\n",
      "** Saving params with joint score = 0.974 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 32 of 50 took       40.967s\n",
      "  training loss:               0.091610\n",
      "  holdout loss (batch):        0.070496\n",
      "  Subject 1: median (max) validation accuracy = 0.970 (0.995)\n",
      "  Subject 2: median (max) validation accuracy = 0.968 (0.996)\n",
      "  Subject 3: median (max) validation accuracy = 0.976 (0.996)\n",
      "  Subject 4: median (max) validation accuracy = 0.976 (0.996)\n",
      "  Subject 5: median (max) validation accuracy = 0.975 (0.996)\n",
      "  Subject 6: median (max) validation accuracy = 0.974 (0.997)\n",
      "  Subject 7: median (max) validation accuracy = 0.976 (0.995)\n",
      "  Subject 8: median (max) validation accuracy = 0.976 (0.996)\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 33 of 50 took       41.170s\n",
      "  training loss:               0.091014\n",
      "  holdout loss (batch):        0.069685\n",
      "  Subject 1: median (max) validation accuracy = 0.970 (0.995)\n",
      "  Subject 2: median (max) validation accuracy = 0.969 (0.996)\n",
      "  Subject 3: median (max) validation accuracy = 0.976 (0.996)\n",
      "  Subject 4: median (max) validation accuracy = 0.976 (0.996)\n",
      "  Subject 5: median (max) validation accuracy = 0.975 (0.996)\n",
      "  Subject 6: median (max) validation accuracy = 0.974 (0.997)\n",
      "  Subject 7: median (max) validation accuracy = 0.976 (0.995)\n",
      "  Subject 8: median (max) validation accuracy = 0.976 (0.996)\n",
      "** Saving params with joint score = 0.974 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 34 of 50 took       40.889s\n",
      "  training loss:               0.090716\n",
      "  holdout loss (batch):        0.069781\n",
      "  Subject 1: median (max) validation accuracy = 0.971 (0.995)\n",
      "  Subject 2: median (max) validation accuracy = 0.969 (0.996)\n",
      "  Subject 3: median (max) validation accuracy = 0.976 (0.996)\n",
      "  Subject 4: median (max) validation accuracy = 0.976 (0.996)\n",
      "  Subject 5: median (max) validation accuracy = 0.975 (0.996)\n",
      "  Subject 6: median (max) validation accuracy = 0.974 (0.997)\n",
      "  Subject 7: median (max) validation accuracy = 0.977 (0.996)\n",
      "  Subject 8: median (max) validation accuracy = 0.976 (0.996)\n",
      "** Saving params with joint score = 0.974 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 35 of 50 took       40.880s\n",
      "  training loss:               0.090737\n",
      "  holdout loss (batch):        0.069643\n",
      "  Subject 1: median (max) validation accuracy = 0.971 (0.995)\n",
      "  Subject 2: median (max) validation accuracy = 0.969 (0.996)\n",
      "  Subject 3: median (max) validation accuracy = 0.976 (0.996)\n",
      "  Subject 4: median (max) validation accuracy = 0.976 (0.996)\n",
      "  Subject 5: median (max) validation accuracy = 0.975 (0.996)\n",
      "  Subject 6: median (max) validation accuracy = 0.974 (0.997)\n",
      "  Subject 7: median (max) validation accuracy = 0.977 (0.996)\n",
      "  Subject 8: median (max) validation accuracy = 0.976 (0.996)\n",
      "** Saving params with joint score = 0.974 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 36 of 50 took       40.876s\n",
      "  training loss:               0.089874\n",
      "  holdout loss (batch):        0.068277\n",
      "  Subject 1: median (max) validation accuracy = 0.972 (0.996)\n",
      "  Subject 2: median (max) validation accuracy = 0.970 (0.996)\n",
      "  Subject 3: median (max) validation accuracy = 0.977 (0.996)\n",
      "  Subject 4: median (max) validation accuracy = 0.977 (0.997)\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "  Subject 5: median (max) validation accuracy = 0.976 (0.996)\n",
      "  Subject 6: median (max) validation accuracy = 0.975 (0.997)\n",
      "  Subject 7: median (max) validation accuracy = 0.977 (0.996)\n",
      "  Subject 8: median (max) validation accuracy = 0.976 (0.996)\n",
      "** Saving params with joint score = 0.975 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 37 of 50 took       40.859s\n",
      "  training loss:               0.089579\n",
      "  holdout loss (batch):        0.066125\n",
      "  Subject 1: median (max) validation accuracy = 0.972 (0.996)\n",
      "  Subject 2: median (max) validation accuracy = 0.970 (0.996)\n",
      "  Subject 3: median (max) validation accuracy = 0.977 (0.996)\n",
      "  Subject 4: median (max) validation accuracy = 0.977 (0.997)\n",
      "  Subject 5: median (max) validation accuracy = 0.975 (0.996)\n",
      "  Subject 6: median (max) validation accuracy = 0.975 (0.997)\n",
      "  Subject 7: median (max) validation accuracy = 0.977 (0.996)\n",
      "  Subject 8: median (max) validation accuracy = 0.977 (0.996)\n",
      "** Saving params with joint score = 0.975 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 38 of 50 took       40.821s\n",
      "  training loss:               0.089752\n",
      "  holdout loss (batch):        0.068373\n",
      "  Subject 1: median (max) validation accuracy = 0.972 (0.996)\n",
      "  Subject 2: median (max) validation accuracy = 0.970 (0.996)\n",
      "  Subject 3: median (max) validation accuracy = 0.977 (0.996)\n",
      "  Subject 4: median (max) validation accuracy = 0.977 (0.997)\n",
      "  Subject 5: median (max) validation accuracy = 0.975 (0.996)\n",
      "  Subject 6: median (max) validation accuracy = 0.975 (0.997)\n",
      "  Subject 7: median (max) validation accuracy = 0.977 (0.996)\n",
      "  Subject 8: median (max) validation accuracy = 0.976 (0.996)\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 39 of 50 took       40.832s\n",
      "  training loss:               0.089164\n",
      "  holdout loss (batch):        0.066622\n",
      "  Subject 1: median (max) validation accuracy = 0.972 (0.996)\n",
      "  Subject 2: median (max) validation accuracy = 0.970 (0.996)\n",
      "  Subject 3: median (max) validation accuracy = 0.977 (0.996)\n",
      "  Subject 4: median (max) validation accuracy = 0.977 (0.997)\n",
      "  Subject 5: median (max) validation accuracy = 0.975 (0.997)\n",
      "  Subject 6: median (max) validation accuracy = 0.975 (0.997)\n",
      "  Subject 7: median (max) validation accuracy = 0.977 (0.996)\n",
      "  Subject 8: median (max) validation accuracy = 0.977 (0.996)\n",
      "** Saving params with joint score = 0.975 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 40 of 50 took       40.819s\n",
      "  training loss:               0.089036\n",
      "  holdout loss (batch):        0.067093\n",
      "  Subject 1: median (max) validation accuracy = 0.972 (0.996)\n",
      "  Subject 2: median (max) validation accuracy = 0.970 (0.996)\n",
      "  Subject 3: median (max) validation accuracy = 0.977 (0.996)\n",
      "  Subject 4: median (max) validation accuracy = 0.977 (0.997)\n",
      "  Subject 5: median (max) validation accuracy = 0.976 (0.996)\n",
      "  Subject 6: median (max) validation accuracy = 0.975 (0.997)\n",
      "  Subject 7: median (max) validation accuracy = 0.977 (0.996)\n",
      "  Subject 8: median (max) validation accuracy = 0.977 (0.996)\n",
      "** Saving params with joint score = 0.975 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 41 of 50 took       40.873s\n",
      "  training loss:               0.088598\n",
      "  holdout loss (batch):        0.065062\n",
      "  Subject 1: median (max) validation accuracy = 0.973 (0.996)\n",
      "  Subject 2: median (max) validation accuracy = 0.971 (0.996)\n",
      "  Subject 3: median (max) validation accuracy = 0.977 (0.996)\n",
      "  Subject 4: median (max) validation accuracy = 0.977 (0.997)\n",
      "  Subject 5: median (max) validation accuracy = 0.976 (0.996)\n",
      "  Subject 6: median (max) validation accuracy = 0.976 (0.997)\n",
      "  Subject 7: median (max) validation accuracy = 0.978 (0.996)\n",
      "  Subject 8: median (max) validation accuracy = 0.977 (0.996)\n",
      "** Saving params with joint score = 0.975 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 42 of 50 took       40.880s\n",
      "  training loss:               0.088433\n",
      "  holdout loss (batch):        0.067602\n",
      "  Subject 1: median (max) validation accuracy = 0.972 (0.996)\n",
      "  Subject 2: median (max) validation accuracy = 0.970 (0.996)\n",
      "  Subject 3: median (max) validation accuracy = 0.977 (0.996)\n",
      "  Subject 4: median (max) validation accuracy = 0.977 (0.997)\n",
      "  Subject 5: median (max) validation accuracy = 0.976 (0.996)\n",
      "  Subject 6: median (max) validation accuracy = 0.975 (0.997)\n",
      "  Subject 7: median (max) validation accuracy = 0.977 (0.996)\n",
      "  Subject 8: median (max) validation accuracy = 0.977 (0.996)\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 43 of 50 took       40.908s\n",
      "  training loss:               0.088483\n",
      "  holdout loss (batch):        0.068339\n",
      "  Subject 1: median (max) validation accuracy = 0.973 (0.996)\n",
      "  Subject 2: median (max) validation accuracy = 0.971 (0.996)\n",
      "  Subject 3: median (max) validation accuracy = 0.977 (0.996)\n",
      "  Subject 4: median (max) validation accuracy = 0.977 (0.997)\n",
      "  Subject 5: median (max) validation accuracy = 0.976 (0.996)\n",
      "  Subject 6: median (max) validation accuracy = 0.975 (0.997)\n",
      "  Subject 7: median (max) validation accuracy = 0.977 (0.996)\n",
      "  Subject 8: median (max) validation accuracy = 0.977 (0.996)\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 44 of 50 took       40.825s\n",
      "  training loss:               0.087905\n",
      "  holdout loss (batch):        0.066625\n",
      "  Subject 1: median (max) validation accuracy = 0.973 (0.996)\n",
      "  Subject 2: median (max) validation accuracy = 0.971 (0.996)\n",
      "  Subject 3: median (max) validation accuracy = 0.977 (0.996)\n",
      "  Subject 4: median (max) validation accuracy = 0.977 (0.997)\n",
      "  Subject 5: median (max) validation accuracy = 0.976 (0.996)\n",
      "  Subject 6: median (max) validation accuracy = 0.975 (0.997)\n",
      "  Subject 7: median (max) validation accuracy = 0.978 (0.996)\n",
      "  Subject 8: median (max) validation accuracy = 0.977 (0.996)\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 45 of 50 took       40.818s\n",
      "  training loss:               0.087791\n",
      "  holdout loss (batch):        0.068036\n",
      "  Subject 1: median (max) validation accuracy = 0.973 (0.996)\n",
      "  Subject 2: median (max) validation accuracy = 0.971 (0.996)\n",
      "  Subject 3: median (max) validation accuracy = 0.977 (0.996)\n",
      "  Subject 4: median (max) validation accuracy = 0.977 (0.997)\n",
      "  Subject 5: median (max) validation accuracy = 0.976 (0.996)\n",
      "  Subject 6: median (max) validation accuracy = 0.975 (0.997)\n",
      "  Subject 7: median (max) validation accuracy = 0.978 (0.996)\n",
      "  Subject 8: median (max) validation accuracy = 0.977 (0.996)\n",
      "** Saving params with joint score = 0.975 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 46 of 50 took       40.804s\n",
      "  training loss:               0.087690\n",
      "  holdout loss (batch):        0.065544\n",
      "  Subject 1: median (max) validation accuracy = 0.973 (0.996)\n",
      "  Subject 2: median (max) validation accuracy = 0.971 (0.996)\n",
      "  Subject 3: median (max) validation accuracy = 0.978 (0.996)\n",
      "  Subject 4: median (max) validation accuracy = 0.977 (0.997)\n",
      "  Subject 5: median (max) validation accuracy = 0.976 (0.996)\n",
      "  Subject 6: median (max) validation accuracy = 0.976 (0.997)\n",
      "  Subject 7: median (max) validation accuracy = 0.978 (0.996)\n",
      "  Subject 8: median (max) validation accuracy = 0.977 (0.996)\n",
      "** Saving params with joint score = 0.976 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 47 of 50 took       40.821s\n",
      "  training loss:               0.087624\n",
      "  holdout loss (batch):        0.065552\n",
      "  Subject 1: median (max) validation accuracy = 0.973 (0.996)\n",
      "  Subject 2: median (max) validation accuracy = 0.971 (0.996)\n",
      "  Subject 3: median (max) validation accuracy = 0.977 (0.996)\n",
      "  Subject 4: median (max) validation accuracy = 0.977 (0.997)\n",
      "  Subject 5: median (max) validation accuracy = 0.976 (0.996)\n",
      "  Subject 6: median (max) validation accuracy = 0.976 (0.997)\n",
      "  Subject 7: median (max) validation accuracy = 0.978 (0.996)\n",
      "  Subject 8: median (max) validation accuracy = 0.977 (0.996)\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 48 of 50 took       40.921s\n",
      "  training loss:               0.087300\n",
      "  holdout loss (batch):        0.064988\n",
      "  Subject 1: median (max) validation accuracy = 0.973 (0.996)\n",
      "  Subject 2: median (max) validation accuracy = 0.972 (0.996)\n",
      "  Subject 3: median (max) validation accuracy = 0.978 (0.996)\n",
      "  Subject 4: median (max) validation accuracy = 0.978 (0.997)\n",
      "  Subject 5: median (max) validation accuracy = 0.976 (0.997)\n",
      "  Subject 6: median (max) validation accuracy = 0.976 (0.997)\n",
      "  Subject 7: median (max) validation accuracy = 0.978 (0.996)\n",
      "  Subject 8: median (max) validation accuracy = 0.977 (0.996)\n",
      "** Saving params with joint score = 0.976 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 49 of 50 took       40.866s\n",
      "  training loss:               0.087276\n",
      "  holdout loss (batch):        0.065984\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "  Subject 1: median (max) validation accuracy = 0.973 (0.996)\n",
      "  Subject 2: median (max) validation accuracy = 0.971 (0.996)\n",
      "  Subject 3: median (max) validation accuracy = 0.977 (0.996)\n",
      "  Subject 4: median (max) validation accuracy = 0.977 (0.997)\n",
      "  Subject 5: median (max) validation accuracy = 0.976 (0.996)\n",
      "  Subject 6: median (max) validation accuracy = 0.975 (0.997)\n",
      "  Subject 7: median (max) validation accuracy = 0.978 (0.996)\n",
      "  Subject 8: median (max) validation accuracy = 0.977 (0.996)\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 50 of 50 took       40.812s\n",
      "  training loss:               0.087078\n",
      "  holdout loss (batch):        0.063090\n",
      "  Subject 1: median (max) validation accuracy = 0.974 (0.996)\n",
      "  Subject 2: median (max) validation accuracy = 0.972 (0.996)\n",
      "  Subject 3: median (max) validation accuracy = 0.978 (0.996)\n",
      "  Subject 4: median (max) validation accuracy = 0.978 (0.997)\n",
      "  Subject 5: median (max) validation accuracy = 0.976 (0.997)\n",
      "  Subject 6: median (max) validation accuracy = 0.976 (0.997)\n",
      "  Subject 7: median (max) validation accuracy = 0.978 (0.996)\n",
      "  Subject 8: median (max) validation accuracy = 0.978 (0.996)\n",
      "** Saving params with joint score = 0.976 **\n",
      "\n",
      "***************************************\n",
      "best joint val cc = 0.977\n",
      "subject 1: val cc = 0.976\n",
      "subject 2: val cc = 0.973\n",
      "subject 3: val cc = 0.979\n",
      "subject 4: val cc = 0.978\n",
      "subject 5: val cc = 0.978\n",
      "subject 6: val cc = 0.977\n",
      "subject 7: val cc = 0.979\n",
      "subject 8: val cc = 0.978\n",
      "***************************************\n",
      "/export/raid1/home/styvesg/code/nsd_gnet8x/output/multisubject/gnet8r_pred_Feb-07-2022_1721/V2/\n",
      "***************************************\n",
      "***                  V2       ***\n",
      "from ==> /export/raid1/home/styvesg/code/nsd_gnet8x/output/multisubject/gnet8r_pred_Feb-07-2022_1721//model_params_V2_no_noise\n",
      "to   ==> /export/raid1/home/styvesg/code/nsd_gnet8x/output/multisubject/gnet8r_pred_Feb-07-2022_1721/V2/\n",
      "***************************************\n",
      "subject 1 masked 4627 of 4627\n",
      "subject 1 training/holdout 8100 900\n",
      "subject 2 masked 3765 of 3765\n",
      "subject 2 training/holdout 8100 900\n",
      "subject 3 masked 3778 of 3778\n",
      "subject 3 training/holdout 8100 900\n",
      "subject 4 masked 3027 of 3027\n",
      "subject 4 training/holdout 8100 900\n",
      "subject 5 masked 3685 of 3685\n",
      "subject 5 training/holdout 8100 900\n",
      "subject 6 masked 4024 of 4024\n",
      "subject 6 training/holdout 8100 900\n",
      "subject 7 masked 3201 of 3201\n",
      "subject 7 training/holdout 8100 900\n",
      "subject 8 masked 3490 of 3490\n",
      "subject 8 training/holdout 8100 900\n",
      "\\ : 100.0 %\n",
      "  Epoch 1 of 50 took       40.769s\n",
      "  training loss:               0.570142\n",
      "  holdout loss (batch):        0.397842\n",
      "  Subject 1: median (max) validation accuracy = 0.679 (0.951)\n",
      "  Subject 2: median (max) validation accuracy = 0.709 (0.947)\n",
      "  Subject 3: median (max) validation accuracy = 0.789 (0.962)\n",
      "  Subject 4: median (max) validation accuracy = 0.812 (0.941)\n",
      "  Subject 5: median (max) validation accuracy = 0.878 (0.959)\n",
      "  Subject 6: median (max) validation accuracy = 0.801 (0.940)\n",
      "  Subject 7: median (max) validation accuracy = 0.778 (0.934)\n",
      "  Subject 8: median (max) validation accuracy = 0.836 (0.951)\n",
      "** Saving params with joint score = 0.791 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 2 of 50 took       40.808s\n",
      "  training loss:               0.382221\n",
      "  holdout loss (batch):        0.317585\n",
      "  Subject 1: median (max) validation accuracy = 0.738 (0.967)\n",
      "  Subject 2: median (max) validation accuracy = 0.773 (0.966)\n",
      "  Subject 3: median (max) validation accuracy = 0.834 (0.971)\n",
      "  Subject 4: median (max) validation accuracy = 0.857 (0.956)\n",
      "  Subject 5: median (max) validation accuracy = 0.908 (0.971)\n",
      "  Subject 6: median (max) validation accuracy = 0.847 (0.960)\n",
      "  Subject 7: median (max) validation accuracy = 0.831 (0.951)\n",
      "  Subject 8: median (max) validation accuracy = 0.873 (0.965)\n",
      "** Saving params with joint score = 0.840 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 3 of 50 took       40.812s\n",
      "  training loss:               0.323691\n",
      "  holdout loss (batch):        0.276534\n",
      "  Subject 1: median (max) validation accuracy = 0.773 (0.974)\n",
      "  Subject 2: median (max) validation accuracy = 0.809 (0.974)\n",
      "  Subject 3: median (max) validation accuracy = 0.858 (0.975)\n",
      "  Subject 4: median (max) validation accuracy = 0.883 (0.964)\n",
      "  Subject 5: median (max) validation accuracy = 0.925 (0.978)\n",
      "  Subject 6: median (max) validation accuracy = 0.875 (0.969)\n",
      "  Subject 7: median (max) validation accuracy = 0.858 (0.966)\n",
      "  Subject 8: median (max) validation accuracy = 0.894 (0.972)\n",
      "** Saving params with joint score = 0.867 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 4 of 50 took       40.810s\n",
      "  training loss:               0.287685\n",
      "  holdout loss (batch):        0.245764\n",
      "  Subject 1: median (max) validation accuracy = 0.800 (0.977)\n",
      "  Subject 2: median (max) validation accuracy = 0.831 (0.979)\n",
      "  Subject 3: median (max) validation accuracy = 0.876 (0.978)\n",
      "  Subject 4: median (max) validation accuracy = 0.899 (0.970)\n",
      "  Subject 5: median (max) validation accuracy = 0.935 (0.983)\n",
      "  Subject 6: median (max) validation accuracy = 0.891 (0.974)\n",
      "  Subject 7: median (max) validation accuracy = 0.878 (0.973)\n",
      "  Subject 8: median (max) validation accuracy = 0.908 (0.975)\n",
      "** Saving params with joint score = 0.885 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 5 of 50 took       40.834s\n",
      "  training loss:               0.260302\n",
      "  holdout loss (batch):        0.222014\n",
      "  Subject 1: median (max) validation accuracy = 0.821 (0.979)\n",
      "  Subject 2: median (max) validation accuracy = 0.850 (0.983)\n",
      "  Subject 3: median (max) validation accuracy = 0.890 (0.980)\n",
      "  Subject 4: median (max) validation accuracy = 0.911 (0.974)\n",
      "  Subject 5: median (max) validation accuracy = 0.942 (0.985)\n",
      "  Subject 6: median (max) validation accuracy = 0.904 (0.977)\n",
      "  Subject 7: median (max) validation accuracy = 0.893 (0.977)\n",
      "  Subject 8: median (max) validation accuracy = 0.919 (0.979)\n",
      "** Saving params with joint score = 0.898 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 6 of 50 took       40.830s\n",
      "  training loss:               0.238148\n",
      "  holdout loss (batch):        0.200864\n",
      "  Subject 1: median (max) validation accuracy = 0.841 (0.981)\n",
      "  Subject 2: median (max) validation accuracy = 0.866 (0.984)\n",
      "  Subject 3: median (max) validation accuracy = 0.903 (0.981)\n",
      "  Subject 4: median (max) validation accuracy = 0.921 (0.978)\n",
      "  Subject 5: median (max) validation accuracy = 0.948 (0.987)\n",
      "  Subject 6: median (max) validation accuracy = 0.915 (0.981)\n",
      "  Subject 7: median (max) validation accuracy = 0.905 (0.980)\n",
      "  Subject 8: median (max) validation accuracy = 0.928 (0.982)\n",
      "** Saving params with joint score = 0.909 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 7 of 50 took       40.806s\n",
      "  training loss:               0.218594\n",
      "  holdout loss (batch):        0.183267\n",
      "  Subject 1: median (max) validation accuracy = 0.858 (0.982)\n",
      "  Subject 2: median (max) validation accuracy = 0.881 (0.986)\n",
      "  Subject 3: median (max) validation accuracy = 0.913 (0.982)\n",
      "  Subject 4: median (max) validation accuracy = 0.929 (0.981)\n",
      "  Subject 5: median (max) validation accuracy = 0.953 (0.989)\n",
      "  Subject 6: median (max) validation accuracy = 0.923 (0.983)\n",
      "  Subject 7: median (max) validation accuracy = 0.916 (0.981)\n",
      "  Subject 8: median (max) validation accuracy = 0.936 (0.983)\n",
      "** Saving params with joint score = 0.919 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 8 of 50 took       40.862s\n",
      "  training loss:               0.200463\n",
      "  holdout loss (batch):        0.166731\n",
      "  Subject 1: median (max) validation accuracy = 0.876 (0.983)\n",
      "  Subject 2: median (max) validation accuracy = 0.894 (0.987)\n",
      "  Subject 3: median (max) validation accuracy = 0.923 (0.983)\n",
      "  Subject 4: median (max) validation accuracy = 0.936 (0.983)\n",
      "  Subject 5: median (max) validation accuracy = 0.958 (0.990)\n",
      "  Subject 6: median (max) validation accuracy = 0.931 (0.985)\n",
      "  Subject 7: median (max) validation accuracy = 0.926 (0.982)\n",
      "  Subject 8: median (max) validation accuracy = 0.943 (0.985)\n",
      "** Saving params with joint score = 0.928 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 9 of 50 took       40.874s\n",
      "  training loss:               0.183213\n",
      "  holdout loss (batch):        0.149449\n",
      "  Subject 1: median (max) validation accuracy = 0.892 (0.983)\n",
      "  Subject 2: median (max) validation accuracy = 0.907 (0.988)\n",
      "  Subject 3: median (max) validation accuracy = 0.933 (0.984)\n",
      "  Subject 4: median (max) validation accuracy = 0.944 (0.985)\n",
      "  Subject 5: median (max) validation accuracy = 0.961 (0.990)\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "  Subject 6: median (max) validation accuracy = 0.939 (0.986)\n",
      "  Subject 7: median (max) validation accuracy = 0.935 (0.983)\n",
      "  Subject 8: median (max) validation accuracy = 0.949 (0.986)\n",
      "** Saving params with joint score = 0.936 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 10 of 50 took       40.794s\n",
      "  training loss:               0.166171\n",
      "  holdout loss (batch):        0.133213\n",
      "  Subject 1: median (max) validation accuracy = 0.907 (0.984)\n",
      "  Subject 2: median (max) validation accuracy = 0.919 (0.989)\n",
      "  Subject 3: median (max) validation accuracy = 0.942 (0.986)\n",
      "  Subject 4: median (max) validation accuracy = 0.951 (0.987)\n",
      "  Subject 5: median (max) validation accuracy = 0.965 (0.991)\n",
      "  Subject 6: median (max) validation accuracy = 0.945 (0.987)\n",
      "  Subject 7: median (max) validation accuracy = 0.943 (0.984)\n",
      "  Subject 8: median (max) validation accuracy = 0.955 (0.987)\n",
      "** Saving params with joint score = 0.943 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 11 of 50 took       40.818s\n",
      "  training loss:               0.150117\n",
      "  holdout loss (batch):        0.115872\n",
      "  Subject 1: median (max) validation accuracy = 0.922 (0.986)\n",
      "  Subject 2: median (max) validation accuracy = 0.930 (0.990)\n",
      "  Subject 3: median (max) validation accuracy = 0.951 (0.987)\n",
      "  Subject 4: median (max) validation accuracy = 0.958 (0.988)\n",
      "  Subject 5: median (max) validation accuracy = 0.968 (0.992)\n",
      "  Subject 6: median (max) validation accuracy = 0.952 (0.988)\n",
      "  Subject 7: median (max) validation accuracy = 0.952 (0.986)\n",
      "  Subject 8: median (max) validation accuracy = 0.960 (0.988)\n",
      "** Saving params with joint score = 0.951 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 12 of 50 took       40.835s\n",
      "  training loss:               0.135835\n",
      "  holdout loss (batch):        0.103972\n",
      "  Subject 1: median (max) validation accuracy = 0.934 (0.987)\n",
      "  Subject 2: median (max) validation accuracy = 0.939 (0.991)\n",
      "  Subject 3: median (max) validation accuracy = 0.958 (0.988)\n",
      "  Subject 4: median (max) validation accuracy = 0.962 (0.988)\n",
      "  Subject 5: median (max) validation accuracy = 0.970 (0.992)\n",
      "  Subject 6: median (max) validation accuracy = 0.957 (0.989)\n",
      "  Subject 7: median (max) validation accuracy = 0.958 (0.988)\n",
      "  Subject 8: median (max) validation accuracy = 0.964 (0.990)\n",
      "** Saving params with joint score = 0.956 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 13 of 50 took       40.795s\n",
      "  training loss:               0.123714\n",
      "  holdout loss (batch):        0.091221\n",
      "  Subject 1: median (max) validation accuracy = 0.944 (0.989)\n",
      "  Subject 2: median (max) validation accuracy = 0.948 (0.991)\n",
      "  Subject 3: median (max) validation accuracy = 0.965 (0.990)\n",
      "  Subject 4: median (max) validation accuracy = 0.968 (0.990)\n",
      "  Subject 5: median (max) validation accuracy = 0.972 (0.993)\n",
      "  Subject 6: median (max) validation accuracy = 0.963 (0.990)\n",
      "  Subject 7: median (max) validation accuracy = 0.965 (0.989)\n",
      "  Subject 8: median (max) validation accuracy = 0.969 (0.991)\n",
      "** Saving params with joint score = 0.962 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 14 of 50 took       40.808s\n",
      "  training loss:               0.114165\n",
      "  holdout loss (batch):        0.083738\n",
      "  Subject 1: median (max) validation accuracy = 0.951 (0.990)\n",
      "  Subject 2: median (max) validation accuracy = 0.954 (0.991)\n",
      "  Subject 3: median (max) validation accuracy = 0.969 (0.990)\n",
      "  Subject 4: median (max) validation accuracy = 0.971 (0.991)\n",
      "  Subject 5: median (max) validation accuracy = 0.974 (0.993)\n",
      "  Subject 6: median (max) validation accuracy = 0.967 (0.992)\n",
      "  Subject 7: median (max) validation accuracy = 0.969 (0.990)\n",
      "  Subject 8: median (max) validation accuracy = 0.972 (0.992)\n",
      "** Saving params with joint score = 0.966 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 15 of 50 took       40.820s\n",
      "  training loss:               0.106667\n",
      "  holdout loss (batch):        0.077075\n",
      "  Subject 1: median (max) validation accuracy = 0.956 (0.991)\n",
      "  Subject 2: median (max) validation accuracy = 0.959 (0.992)\n",
      "  Subject 3: median (max) validation accuracy = 0.972 (0.991)\n",
      "  Subject 4: median (max) validation accuracy = 0.974 (0.992)\n",
      "  Subject 5: median (max) validation accuracy = 0.975 (0.993)\n",
      "  Subject 6: median (max) validation accuracy = 0.970 (0.993)\n",
      "  Subject 7: median (max) validation accuracy = 0.972 (0.991)\n",
      "  Subject 8: median (max) validation accuracy = 0.974 (0.993)\n",
      "** Saving params with joint score = 0.969 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 16 of 50 took       40.816s\n",
      "  training loss:               0.101151\n",
      "  holdout loss (batch):        0.074203\n",
      "  Subject 1: median (max) validation accuracy = 0.960 (0.991)\n",
      "  Subject 2: median (max) validation accuracy = 0.963 (0.992)\n",
      "  Subject 3: median (max) validation accuracy = 0.975 (0.991)\n",
      "  Subject 4: median (max) validation accuracy = 0.976 (0.992)\n",
      "  Subject 5: median (max) validation accuracy = 0.976 (0.994)\n",
      "  Subject 6: median (max) validation accuracy = 0.973 (0.993)\n",
      "  Subject 7: median (max) validation accuracy = 0.974 (0.992)\n",
      "  Subject 8: median (max) validation accuracy = 0.976 (0.993)\n",
      "** Saving params with joint score = 0.972 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 17 of 50 took       40.841s\n",
      "  training loss:               0.097070\n",
      "  holdout loss (batch):        0.069067\n",
      "  Subject 1: median (max) validation accuracy = 0.964 (0.992)\n",
      "  Subject 2: median (max) validation accuracy = 0.966 (0.992)\n",
      "  Subject 3: median (max) validation accuracy = 0.977 (0.992)\n",
      "  Subject 4: median (max) validation accuracy = 0.977 (0.993)\n",
      "  Subject 5: median (max) validation accuracy = 0.977 (0.994)\n",
      "  Subject 6: median (max) validation accuracy = 0.975 (0.994)\n",
      "  Subject 7: median (max) validation accuracy = 0.976 (0.993)\n",
      "  Subject 8: median (max) validation accuracy = 0.978 (0.994)\n",
      "** Saving params with joint score = 0.974 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 18 of 50 took       40.809s\n",
      "  training loss:               0.093784\n",
      "  holdout loss (batch):        0.066179\n",
      "  Subject 1: median (max) validation accuracy = 0.966 (0.992)\n",
      "  Subject 2: median (max) validation accuracy = 0.968 (0.993)\n",
      "  Subject 3: median (max) validation accuracy = 0.978 (0.993)\n",
      "  Subject 4: median (max) validation accuracy = 0.978 (0.993)\n",
      "  Subject 5: median (max) validation accuracy = 0.977 (0.994)\n",
      "  Subject 6: median (max) validation accuracy = 0.976 (0.994)\n",
      "  Subject 7: median (max) validation accuracy = 0.977 (0.993)\n",
      "  Subject 8: median (max) validation accuracy = 0.979 (0.994)\n",
      "** Saving params with joint score = 0.975 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 19 of 50 took       40.800s\n",
      "  training loss:               0.091427\n",
      "  holdout loss (batch):        0.064970\n",
      "  Subject 1: median (max) validation accuracy = 0.968 (0.993)\n",
      "  Subject 2: median (max) validation accuracy = 0.970 (0.993)\n",
      "  Subject 3: median (max) validation accuracy = 0.979 (0.993)\n",
      "  Subject 4: median (max) validation accuracy = 0.979 (0.994)\n",
      "  Subject 5: median (max) validation accuracy = 0.978 (0.994)\n",
      "  Subject 6: median (max) validation accuracy = 0.977 (0.994)\n",
      "  Subject 7: median (max) validation accuracy = 0.978 (0.993)\n",
      "  Subject 8: median (max) validation accuracy = 0.979 (0.994)\n",
      "** Saving params with joint score = 0.976 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 20 of 50 took       40.823s\n",
      "  training loss:               0.089517\n",
      "  holdout loss (batch):        0.062259\n",
      "  Subject 1: median (max) validation accuracy = 0.970 (0.993)\n",
      "  Subject 2: median (max) validation accuracy = 0.971 (0.993)\n",
      "  Subject 3: median (max) validation accuracy = 0.980 (0.993)\n",
      "  Subject 4: median (max) validation accuracy = 0.980 (0.994)\n",
      "  Subject 5: median (max) validation accuracy = 0.979 (0.994)\n",
      "  Subject 6: median (max) validation accuracy = 0.978 (0.994)\n",
      "  Subject 7: median (max) validation accuracy = 0.979 (0.994)\n",
      "  Subject 8: median (max) validation accuracy = 0.980 (0.994)\n",
      "** Saving params with joint score = 0.977 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 21 of 50 took       40.848s\n",
      "  training loss:               0.087684\n",
      "  holdout loss (batch):        0.060118\n",
      "  Subject 1: median (max) validation accuracy = 0.971 (0.994)\n",
      "  Subject 2: median (max) validation accuracy = 0.972 (0.993)\n",
      "  Subject 3: median (max) validation accuracy = 0.981 (0.994)\n",
      "  Subject 4: median (max) validation accuracy = 0.981 (0.994)\n",
      "  Subject 5: median (max) validation accuracy = 0.979 (0.994)\n",
      "  Subject 6: median (max) validation accuracy = 0.979 (0.994)\n",
      "  Subject 7: median (max) validation accuracy = 0.980 (0.994)\n",
      "  Subject 8: median (max) validation accuracy = 0.981 (0.995)\n",
      "** Saving params with joint score = 0.978 **\n",
      "\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\\ : 100.0 %\n",
      "  Epoch 22 of 50 took       40.808s\n",
      "  training loss:               0.086319\n",
      "  holdout loss (batch):        0.059515\n",
      "  Subject 1: median (max) validation accuracy = 0.973 (0.994)\n",
      "  Subject 2: median (max) validation accuracy = 0.973 (0.994)\n",
      "  Subject 3: median (max) validation accuracy = 0.981 (0.994)\n",
      "  Subject 4: median (max) validation accuracy = 0.981 (0.994)\n",
      "  Subject 5: median (max) validation accuracy = 0.979 (0.995)\n",
      "  Subject 6: median (max) validation accuracy = 0.979 (0.995)\n",
      "  Subject 7: median (max) validation accuracy = 0.980 (0.994)\n",
      "  Subject 8: median (max) validation accuracy = 0.981 (0.995)\n",
      "** Saving params with joint score = 0.978 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 23 of 50 took       40.844s\n",
      "  training loss:               0.084800\n",
      "  holdout loss (batch):        0.056548\n",
      "  Subject 1: median (max) validation accuracy = 0.974 (0.994)\n",
      "  Subject 2: median (max) validation accuracy = 0.974 (0.994)\n",
      "  Subject 3: median (max) validation accuracy = 0.982 (0.994)\n",
      "  Subject 4: median (max) validation accuracy = 0.982 (0.995)\n",
      "  Subject 5: median (max) validation accuracy = 0.979 (0.995)\n",
      "  Subject 6: median (max) validation accuracy = 0.980 (0.995)\n",
      "  Subject 7: median (max) validation accuracy = 0.981 (0.994)\n",
      "  Subject 8: median (max) validation accuracy = 0.981 (0.995)\n",
      "** Saving params with joint score = 0.979 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 24 of 50 took       40.835s\n",
      "  training loss:               0.083862\n",
      "  holdout loss (batch):        0.056071\n",
      "  Subject 1: median (max) validation accuracy = 0.975 (0.995)\n",
      "  Subject 2: median (max) validation accuracy = 0.975 (0.994)\n",
      "  Subject 3: median (max) validation accuracy = 0.982 (0.994)\n",
      "  Subject 4: median (max) validation accuracy = 0.982 (0.995)\n",
      "  Subject 5: median (max) validation accuracy = 0.980 (0.995)\n",
      "  Subject 6: median (max) validation accuracy = 0.980 (0.995)\n",
      "  Subject 7: median (max) validation accuracy = 0.981 (0.994)\n",
      "  Subject 8: median (max) validation accuracy = 0.982 (0.995)\n",
      "** Saving params with joint score = 0.979 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 25 of 50 took       40.831s\n",
      "  training loss:               0.083021\n",
      "  holdout loss (batch):        0.058095\n",
      "  Subject 1: median (max) validation accuracy = 0.975 (0.995)\n",
      "  Subject 2: median (max) validation accuracy = 0.976 (0.994)\n",
      "  Subject 3: median (max) validation accuracy = 0.982 (0.994)\n",
      "  Subject 4: median (max) validation accuracy = 0.982 (0.995)\n",
      "  Subject 5: median (max) validation accuracy = 0.980 (0.995)\n",
      "  Subject 6: median (max) validation accuracy = 0.980 (0.995)\n",
      "  Subject 7: median (max) validation accuracy = 0.981 (0.995)\n",
      "  Subject 8: median (max) validation accuracy = 0.982 (0.995)\n",
      "** Saving params with joint score = 0.980 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 26 of 50 took       40.832s\n",
      "  training loss:               0.081906\n",
      "  holdout loss (batch):        0.056375\n",
      "  Subject 1: median (max) validation accuracy = 0.976 (0.995)\n",
      "  Subject 2: median (max) validation accuracy = 0.976 (0.994)\n",
      "  Subject 3: median (max) validation accuracy = 0.982 (0.994)\n",
      "  Subject 4: median (max) validation accuracy = 0.983 (0.995)\n",
      "  Subject 5: median (max) validation accuracy = 0.980 (0.995)\n",
      "  Subject 6: median (max) validation accuracy = 0.980 (0.995)\n",
      "  Subject 7: median (max) validation accuracy = 0.982 (0.995)\n",
      "  Subject 8: median (max) validation accuracy = 0.982 (0.995)\n",
      "** Saving params with joint score = 0.980 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 27 of 50 took       40.844s\n",
      "  training loss:               0.081253\n",
      "  holdout loss (batch):        0.055362\n",
      "  Subject 1: median (max) validation accuracy = 0.976 (0.994)\n",
      "  Subject 2: median (max) validation accuracy = 0.977 (0.993)\n",
      "  Subject 3: median (max) validation accuracy = 0.983 (0.995)\n",
      "  Subject 4: median (max) validation accuracy = 0.983 (0.995)\n",
      "  Subject 5: median (max) validation accuracy = 0.980 (0.995)\n",
      "  Subject 6: median (max) validation accuracy = 0.981 (0.995)\n",
      "  Subject 7: median (max) validation accuracy = 0.982 (0.995)\n",
      "  Subject 8: median (max) validation accuracy = 0.982 (0.995)\n",
      "** Saving params with joint score = 0.980 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 28 of 50 took       40.925s\n",
      "  training loss:               0.080582\n",
      "  holdout loss (batch):        0.054385\n",
      "  Subject 1: median (max) validation accuracy = 0.977 (0.995)\n",
      "  Subject 2: median (max) validation accuracy = 0.977 (0.994)\n",
      "  Subject 3: median (max) validation accuracy = 0.983 (0.995)\n",
      "  Subject 4: median (max) validation accuracy = 0.983 (0.995)\n",
      "  Subject 5: median (max) validation accuracy = 0.980 (0.995)\n",
      "  Subject 6: median (max) validation accuracy = 0.981 (0.995)\n",
      "  Subject 7: median (max) validation accuracy = 0.982 (0.995)\n",
      "  Subject 8: median (max) validation accuracy = 0.982 (0.995)\n",
      "** Saving params with joint score = 0.981 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 29 of 50 took       40.948s\n",
      "  training loss:               0.080302\n",
      "  holdout loss (batch):        0.055057\n",
      "  Subject 1: median (max) validation accuracy = 0.978 (0.995)\n",
      "  Subject 2: median (max) validation accuracy = 0.977 (0.994)\n",
      "  Subject 3: median (max) validation accuracy = 0.983 (0.995)\n",
      "  Subject 4: median (max) validation accuracy = 0.983 (0.995)\n",
      "  Subject 5: median (max) validation accuracy = 0.980 (0.995)\n",
      "  Subject 6: median (max) validation accuracy = 0.981 (0.995)\n",
      "  Subject 7: median (max) validation accuracy = 0.982 (0.995)\n",
      "  Subject 8: median (max) validation accuracy = 0.983 (0.995)\n",
      "** Saving params with joint score = 0.981 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 30 of 50 took       40.895s\n",
      "  training loss:               0.079538\n",
      "  holdout loss (batch):        0.054024\n",
      "  Subject 1: median (max) validation accuracy = 0.978 (0.995)\n",
      "  Subject 2: median (max) validation accuracy = 0.978 (0.994)\n",
      "  Subject 3: median (max) validation accuracy = 0.983 (0.995)\n",
      "  Subject 4: median (max) validation accuracy = 0.983 (0.995)\n",
      "  Subject 5: median (max) validation accuracy = 0.981 (0.995)\n",
      "  Subject 6: median (max) validation accuracy = 0.982 (0.995)\n",
      "  Subject 7: median (max) validation accuracy = 0.982 (0.995)\n",
      "  Subject 8: median (max) validation accuracy = 0.983 (0.995)\n",
      "** Saving params with joint score = 0.981 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 31 of 50 took       40.824s\n",
      "  training loss:               0.079196\n",
      "  holdout loss (batch):        0.054232\n",
      "  Subject 1: median (max) validation accuracy = 0.979 (0.995)\n",
      "  Subject 2: median (max) validation accuracy = 0.978 (0.994)\n",
      "  Subject 3: median (max) validation accuracy = 0.983 (0.995)\n",
      "  Subject 4: median (max) validation accuracy = 0.983 (0.995)\n",
      "  Subject 5: median (max) validation accuracy = 0.981 (0.995)\n",
      "  Subject 6: median (max) validation accuracy = 0.982 (0.995)\n",
      "  Subject 7: median (max) validation accuracy = 0.982 (0.995)\n",
      "  Subject 8: median (max) validation accuracy = 0.983 (0.995)\n",
      "** Saving params with joint score = 0.981 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 32 of 50 took       40.865s\n",
      "  training loss:               0.078660\n",
      "  holdout loss (batch):        0.052790\n",
      "  Subject 1: median (max) validation accuracy = 0.979 (0.995)\n",
      "  Subject 2: median (max) validation accuracy = 0.979 (0.994)\n",
      "  Subject 3: median (max) validation accuracy = 0.983 (0.995)\n",
      "  Subject 4: median (max) validation accuracy = 0.984 (0.995)\n",
      "  Subject 5: median (max) validation accuracy = 0.981 (0.995)\n",
      "  Subject 6: median (max) validation accuracy = 0.982 (0.995)\n",
      "  Subject 7: median (max) validation accuracy = 0.983 (0.995)\n",
      "  Subject 8: median (max) validation accuracy = 0.983 (0.995)\n",
      "** Saving params with joint score = 0.981 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 33 of 50 took       40.869s\n",
      "  training loss:               0.078403\n",
      "  holdout loss (batch):        0.051837\n",
      "  Subject 1: median (max) validation accuracy = 0.979 (0.995)\n",
      "  Subject 2: median (max) validation accuracy = 0.979 (0.994)\n",
      "  Subject 3: median (max) validation accuracy = 0.984 (0.995)\n",
      "  Subject 4: median (max) validation accuracy = 0.984 (0.995)\n",
      "  Subject 5: median (max) validation accuracy = 0.981 (0.995)\n",
      "  Subject 6: median (max) validation accuracy = 0.982 (0.995)\n",
      "  Subject 7: median (max) validation accuracy = 0.983 (0.995)\n",
      "  Subject 8: median (max) validation accuracy = 0.983 (0.995)\n",
      "** Saving params with joint score = 0.982 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 34 of 50 took       40.826s\n",
      "  training loss:               0.077985\n",
      "  holdout loss (batch):        0.052769\n",
      "  Subject 1: median (max) validation accuracy = 0.980 (0.995)\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "  Subject 2: median (max) validation accuracy = 0.979 (0.994)\n",
      "  Subject 3: median (max) validation accuracy = 0.984 (0.995)\n",
      "  Subject 4: median (max) validation accuracy = 0.984 (0.995)\n",
      "  Subject 5: median (max) validation accuracy = 0.981 (0.995)\n",
      "  Subject 6: median (max) validation accuracy = 0.982 (0.995)\n",
      "  Subject 7: median (max) validation accuracy = 0.983 (0.995)\n",
      "  Subject 8: median (max) validation accuracy = 0.983 (0.995)\n",
      "** Saving params with joint score = 0.982 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 35 of 50 took       40.855s\n",
      "  training loss:               0.077779\n",
      "  holdout loss (batch):        0.051114\n",
      "  Subject 1: median (max) validation accuracy = 0.979 (0.995)\n",
      "  Subject 2: median (max) validation accuracy = 0.979 (0.994)\n",
      "  Subject 3: median (max) validation accuracy = 0.984 (0.995)\n",
      "  Subject 4: median (max) validation accuracy = 0.984 (0.995)\n",
      "  Subject 5: median (max) validation accuracy = 0.981 (0.995)\n",
      "  Subject 6: median (max) validation accuracy = 0.982 (0.995)\n",
      "  Subject 7: median (max) validation accuracy = 0.983 (0.995)\n",
      "  Subject 8: median (max) validation accuracy = 0.983 (0.995)\n",
      "** Saving params with joint score = 0.982 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 36 of 50 took       40.859s\n",
      "  training loss:               0.077225\n",
      "  holdout loss (batch):        0.052126\n",
      "  Subject 1: median (max) validation accuracy = 0.980 (0.995)\n",
      "  Subject 2: median (max) validation accuracy = 0.980 (0.994)\n",
      "  Subject 3: median (max) validation accuracy = 0.984 (0.995)\n",
      "  Subject 4: median (max) validation accuracy = 0.984 (0.995)\n",
      "  Subject 5: median (max) validation accuracy = 0.981 (0.995)\n",
      "  Subject 6: median (max) validation accuracy = 0.982 (0.995)\n",
      "  Subject 7: median (max) validation accuracy = 0.983 (0.995)\n",
      "  Subject 8: median (max) validation accuracy = 0.983 (0.995)\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 37 of 50 took       40.907s\n",
      "  training loss:               0.077110\n",
      "  holdout loss (batch):        0.051394\n",
      "  Subject 1: median (max) validation accuracy = 0.980 (0.995)\n",
      "  Subject 2: median (max) validation accuracy = 0.980 (0.994)\n",
      "  Subject 3: median (max) validation accuracy = 0.984 (0.995)\n",
      "  Subject 4: median (max) validation accuracy = 0.984 (0.995)\n",
      "  Subject 5: median (max) validation accuracy = 0.981 (0.995)\n",
      "  Subject 6: median (max) validation accuracy = 0.982 (0.996)\n",
      "  Subject 7: median (max) validation accuracy = 0.983 (0.995)\n",
      "  Subject 8: median (max) validation accuracy = 0.984 (0.996)\n",
      "** Saving params with joint score = 0.982 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 38 of 50 took       40.911s\n",
      "  training loss:               0.077112\n",
      "  holdout loss (batch):        0.051325\n",
      "  Subject 1: median (max) validation accuracy = 0.980 (0.995)\n",
      "  Subject 2: median (max) validation accuracy = 0.980 (0.994)\n",
      "  Subject 3: median (max) validation accuracy = 0.984 (0.995)\n",
      "  Subject 4: median (max) validation accuracy = 0.984 (0.995)\n",
      "  Subject 5: median (max) validation accuracy = 0.981 (0.995)\n",
      "  Subject 6: median (max) validation accuracy = 0.982 (0.995)\n",
      "  Subject 7: median (max) validation accuracy = 0.983 (0.995)\n",
      "  Subject 8: median (max) validation accuracy = 0.983 (0.995)\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 39 of 50 took       40.921s\n",
      "  training loss:               0.076688\n",
      "  holdout loss (batch):        0.052556\n",
      "  Subject 1: median (max) validation accuracy = 0.980 (0.995)\n",
      "  Subject 2: median (max) validation accuracy = 0.980 (0.994)\n",
      "  Subject 3: median (max) validation accuracy = 0.984 (0.995)\n",
      "  Subject 4: median (max) validation accuracy = 0.984 (0.995)\n",
      "  Subject 5: median (max) validation accuracy = 0.981 (0.995)\n",
      "  Subject 6: median (max) validation accuracy = 0.982 (0.995)\n",
      "  Subject 7: median (max) validation accuracy = 0.983 (0.995)\n",
      "  Subject 8: median (max) validation accuracy = 0.983 (0.995)\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 40 of 50 took       40.929s\n",
      "  training loss:               0.076521\n",
      "  holdout loss (batch):        0.050802\n",
      "  Subject 1: median (max) validation accuracy = 0.980 (0.996)\n",
      "  Subject 2: median (max) validation accuracy = 0.980 (0.994)\n",
      "  Subject 3: median (max) validation accuracy = 0.984 (0.995)\n",
      "  Subject 4: median (max) validation accuracy = 0.984 (0.995)\n",
      "  Subject 5: median (max) validation accuracy = 0.981 (0.995)\n",
      "  Subject 6: median (max) validation accuracy = 0.982 (0.995)\n",
      "  Subject 7: median (max) validation accuracy = 0.983 (0.995)\n",
      "  Subject 8: median (max) validation accuracy = 0.984 (0.996)\n",
      "** Saving params with joint score = 0.982 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 41 of 50 took       40.886s\n",
      "  training loss:               0.076179\n",
      "  holdout loss (batch):        0.051290\n",
      "  Subject 1: median (max) validation accuracy = 0.981 (0.996)\n",
      "  Subject 2: median (max) validation accuracy = 0.980 (0.994)\n",
      "  Subject 3: median (max) validation accuracy = 0.984 (0.995)\n",
      "  Subject 4: median (max) validation accuracy = 0.984 (0.995)\n",
      "  Subject 5: median (max) validation accuracy = 0.981 (0.995)\n",
      "  Subject 6: median (max) validation accuracy = 0.983 (0.995)\n",
      "  Subject 7: median (max) validation accuracy = 0.983 (0.995)\n",
      "  Subject 8: median (max) validation accuracy = 0.984 (0.996)\n",
      "** Saving params with joint score = 0.982 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 42 of 50 took       40.834s\n",
      "  training loss:               0.076263\n",
      "  holdout loss (batch):        0.050138\n",
      "  Subject 1: median (max) validation accuracy = 0.981 (0.996)\n",
      "  Subject 2: median (max) validation accuracy = 0.981 (0.994)\n",
      "  Subject 3: median (max) validation accuracy = 0.984 (0.995)\n",
      "  Subject 4: median (max) validation accuracy = 0.984 (0.995)\n",
      "  Subject 5: median (max) validation accuracy = 0.982 (0.995)\n",
      "  Subject 6: median (max) validation accuracy = 0.983 (0.995)\n",
      "  Subject 7: median (max) validation accuracy = 0.984 (0.995)\n",
      "  Subject 8: median (max) validation accuracy = 0.984 (0.996)\n",
      "** Saving params with joint score = 0.983 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 43 of 50 took       40.843s\n",
      "  training loss:               0.075774\n",
      "  holdout loss (batch):        0.051557\n",
      "  Subject 1: median (max) validation accuracy = 0.981 (0.996)\n",
      "  Subject 2: median (max) validation accuracy = 0.981 (0.994)\n",
      "  Subject 3: median (max) validation accuracy = 0.984 (0.995)\n",
      "  Subject 4: median (max) validation accuracy = 0.984 (0.995)\n",
      "  Subject 5: median (max) validation accuracy = 0.981 (0.995)\n",
      "  Subject 6: median (max) validation accuracy = 0.983 (0.995)\n",
      "  Subject 7: median (max) validation accuracy = 0.983 (0.996)\n",
      "  Subject 8: median (max) validation accuracy = 0.984 (0.996)\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 44 of 50 took       40.853s\n",
      "  training loss:               0.075777\n",
      "  holdout loss (batch):        0.049284\n",
      "  Subject 1: median (max) validation accuracy = 0.981 (0.996)\n",
      "  Subject 2: median (max) validation accuracy = 0.981 (0.994)\n",
      "  Subject 3: median (max) validation accuracy = 0.984 (0.995)\n",
      "  Subject 4: median (max) validation accuracy = 0.985 (0.995)\n",
      "  Subject 5: median (max) validation accuracy = 0.982 (0.995)\n",
      "  Subject 6: median (max) validation accuracy = 0.983 (0.995)\n",
      "  Subject 7: median (max) validation accuracy = 0.984 (0.996)\n",
      "  Subject 8: median (max) validation accuracy = 0.984 (0.996)\n",
      "** Saving params with joint score = 0.983 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 45 of 50 took       41.132s\n",
      "  training loss:               0.075714\n",
      "  holdout loss (batch):        0.052079\n",
      "  Subject 1: median (max) validation accuracy = 0.981 (0.996)\n",
      "  Subject 2: median (max) validation accuracy = 0.981 (0.995)\n",
      "  Subject 3: median (max) validation accuracy = 0.984 (0.995)\n",
      "  Subject 4: median (max) validation accuracy = 0.984 (0.995)\n",
      "  Subject 5: median (max) validation accuracy = 0.982 (0.995)\n",
      "  Subject 6: median (max) validation accuracy = 0.983 (0.995)\n",
      "  Subject 7: median (max) validation accuracy = 0.984 (0.995)\n",
      "  Subject 8: median (max) validation accuracy = 0.984 (0.995)\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 46 of 50 took       41.231s\n",
      "  training loss:               0.075587\n",
      "  holdout loss (batch):        0.048631\n",
      "  Subject 1: median (max) validation accuracy = 0.981 (0.996)\n",
      "  Subject 2: median (max) validation accuracy = 0.981 (0.995)\n",
      "  Subject 3: median (max) validation accuracy = 0.985 (0.995)\n",
      "  Subject 4: median (max) validation accuracy = 0.985 (0.995)\n",
      "  Subject 5: median (max) validation accuracy = 0.982 (0.995)\n",
      "  Subject 6: median (max) validation accuracy = 0.983 (0.995)\n",
      "  Subject 7: median (max) validation accuracy = 0.984 (0.996)\n",
      "  Subject 8: median (max) validation accuracy = 0.984 (0.996)\n",
      "** Saving params with joint score = 0.983 **\n",
      "\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\\ : 100.0 %\n",
      "  Epoch 47 of 50 took       41.229s\n",
      "  training loss:               0.075342\n",
      "  holdout loss (batch):        0.051172\n",
      "  Subject 1: median (max) validation accuracy = 0.981 (0.995)\n",
      "  Subject 2: median (max) validation accuracy = 0.981 (0.994)\n",
      "  Subject 3: median (max) validation accuracy = 0.984 (0.995)\n",
      "  Subject 4: median (max) validation accuracy = 0.985 (0.995)\n",
      "  Subject 5: median (max) validation accuracy = 0.982 (0.995)\n",
      "  Subject 6: median (max) validation accuracy = 0.983 (0.996)\n",
      "  Subject 7: median (max) validation accuracy = 0.984 (0.995)\n",
      "  Subject 8: median (max) validation accuracy = 0.984 (0.995)\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 48 of 50 took       40.952s\n",
      "  training loss:               0.074873\n",
      "  holdout loss (batch):        0.048565\n",
      "  Subject 1: median (max) validation accuracy = 0.981 (0.996)\n",
      "  Subject 2: median (max) validation accuracy = 0.981 (0.994)\n",
      "  Subject 3: median (max) validation accuracy = 0.985 (0.995)\n",
      "  Subject 4: median (max) validation accuracy = 0.985 (0.995)\n",
      "  Subject 5: median (max) validation accuracy = 0.982 (0.995)\n",
      "  Subject 6: median (max) validation accuracy = 0.983 (0.995)\n",
      "  Subject 7: median (max) validation accuracy = 0.984 (0.996)\n",
      "  Subject 8: median (max) validation accuracy = 0.984 (0.995)\n",
      "** Saving params with joint score = 0.983 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 49 of 50 took       40.885s\n",
      "  training loss:               0.074936\n",
      "  holdout loss (batch):        0.050024\n",
      "  Subject 1: median (max) validation accuracy = 0.981 (0.996)\n",
      "  Subject 2: median (max) validation accuracy = 0.981 (0.994)\n",
      "  Subject 3: median (max) validation accuracy = 0.985 (0.995)\n",
      "  Subject 4: median (max) validation accuracy = 0.985 (0.995)\n",
      "  Subject 5: median (max) validation accuracy = 0.982 (0.995)\n",
      "  Subject 6: median (max) validation accuracy = 0.983 (0.995)\n",
      "  Subject 7: median (max) validation accuracy = 0.984 (0.995)\n",
      "  Subject 8: median (max) validation accuracy = 0.984 (0.995)\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 50 of 50 took       41.241s\n",
      "  training loss:               0.075040\n",
      "  holdout loss (batch):        0.049247\n",
      "  Subject 1: median (max) validation accuracy = 0.982 (0.996)\n",
      "  Subject 2: median (max) validation accuracy = 0.981 (0.994)\n",
      "  Subject 3: median (max) validation accuracy = 0.985 (0.995)\n",
      "  Subject 4: median (max) validation accuracy = 0.985 (0.995)\n",
      "  Subject 5: median (max) validation accuracy = 0.982 (0.995)\n",
      "  Subject 6: median (max) validation accuracy = 0.983 (0.995)\n",
      "  Subject 7: median (max) validation accuracy = 0.984 (0.996)\n",
      "  Subject 8: median (max) validation accuracy = 0.984 (0.995)\n",
      "** Saving params with joint score = 0.983 **\n",
      "\n",
      "***************************************\n",
      "best joint val cc = 0.984\n",
      "subject 1: val cc = 0.983\n",
      "subject 2: val cc = 0.983\n",
      "subject 3: val cc = 0.986\n",
      "subject 4: val cc = 0.985\n",
      "subject 5: val cc = 0.984\n",
      "subject 6: val cc = 0.984\n",
      "subject 7: val cc = 0.985\n",
      "subject 8: val cc = 0.984\n",
      "***************************************\n",
      "/export/raid1/home/styvesg/code/nsd_gnet8x/output/multisubject/gnet8r_pred_Feb-07-2022_1721/V3/\n",
      "***************************************\n",
      "***                  V3       ***\n",
      "from ==> /export/raid1/home/styvesg/code/nsd_gnet8x/output/multisubject/gnet8r_pred_Feb-07-2022_1721//model_params_V3_no_noise\n",
      "to   ==> /export/raid1/home/styvesg/code/nsd_gnet8x/output/multisubject/gnet8r_pred_Feb-07-2022_1721/V3/\n",
      "***************************************\n",
      "subject 1 masked 4627 of 4627\n",
      "subject 1 training/holdout 8100 900\n",
      "subject 2 masked 3765 of 3765\n",
      "subject 2 training/holdout 8100 900\n",
      "subject 3 masked 3778 of 3778\n",
      "subject 3 training/holdout 8100 900\n",
      "subject 4 masked 3027 of 3027\n",
      "subject 4 training/holdout 8100 900\n",
      "subject 5 masked 3685 of 3685\n",
      "subject 5 training/holdout 8100 900\n",
      "subject 6 masked 4024 of 4024\n",
      "subject 6 training/holdout 8100 900\n",
      "subject 7 masked 3201 of 3201\n",
      "subject 7 training/holdout 8100 900\n",
      "subject 8 masked 3490 of 3490\n",
      "subject 8 training/holdout 8100 900\n",
      "\\ : 100.0 %\n",
      "  Epoch 1 of 50 took       41.177s\n",
      "  training loss:               0.574547\n",
      "  holdout loss (batch):        0.408934\n",
      "  Subject 1: median (max) validation accuracy = 0.661 (0.956)\n",
      "  Subject 2: median (max) validation accuracy = 0.701 (0.942)\n",
      "  Subject 3: median (max) validation accuracy = 0.776 (0.949)\n",
      "  Subject 4: median (max) validation accuracy = 0.804 (0.944)\n",
      "  Subject 5: median (max) validation accuracy = 0.869 (0.957)\n",
      "  Subject 6: median (max) validation accuracy = 0.800 (0.941)\n",
      "  Subject 7: median (max) validation accuracy = 0.771 (0.938)\n",
      "  Subject 8: median (max) validation accuracy = 0.828 (0.956)\n",
      "** Saving params with joint score = 0.784 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 2 of 50 took       41.217s\n",
      "  training loss:               0.389731\n",
      "  holdout loss (batch):        0.330634\n",
      "  Subject 1: median (max) validation accuracy = 0.721 (0.968)\n",
      "  Subject 2: median (max) validation accuracy = 0.762 (0.963)\n",
      "  Subject 3: median (max) validation accuracy = 0.824 (0.960)\n",
      "  Subject 4: median (max) validation accuracy = 0.851 (0.962)\n",
      "  Subject 5: median (max) validation accuracy = 0.902 (0.971)\n",
      "  Subject 6: median (max) validation accuracy = 0.846 (0.959)\n",
      "  Subject 7: median (max) validation accuracy = 0.822 (0.956)\n",
      "  Subject 8: median (max) validation accuracy = 0.868 (0.972)\n",
      "** Saving params with joint score = 0.835 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 3 of 50 took       41.214s\n",
      "  training loss:               0.332298\n",
      "  holdout loss (batch):        0.289037\n",
      "  Subject 1: median (max) validation accuracy = 0.759 (0.973)\n",
      "  Subject 2: median (max) validation accuracy = 0.795 (0.971)\n",
      "  Subject 3: median (max) validation accuracy = 0.850 (0.965)\n",
      "  Subject 4: median (max) validation accuracy = 0.876 (0.969)\n",
      "  Subject 5: median (max) validation accuracy = 0.918 (0.976)\n",
      "  Subject 6: median (max) validation accuracy = 0.872 (0.969)\n",
      "  Subject 7: median (max) validation accuracy = 0.850 (0.963)\n",
      "  Subject 8: median (max) validation accuracy = 0.888 (0.978)\n",
      "** Saving params with joint score = 0.861 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 4 of 50 took       41.229s\n",
      "  training loss:               0.297111\n",
      "  holdout loss (batch):        0.258694\n",
      "  Subject 1: median (max) validation accuracy = 0.787 (0.977)\n",
      "  Subject 2: median (max) validation accuracy = 0.819 (0.975)\n",
      "  Subject 3: median (max) validation accuracy = 0.869 (0.970)\n",
      "  Subject 4: median (max) validation accuracy = 0.893 (0.975)\n",
      "  Subject 5: median (max) validation accuracy = 0.929 (0.980)\n",
      "  Subject 6: median (max) validation accuracy = 0.890 (0.976)\n",
      "  Subject 7: median (max) validation accuracy = 0.868 (0.967)\n",
      "  Subject 8: median (max) validation accuracy = 0.903 (0.982)\n",
      "** Saving params with joint score = 0.880 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 5 of 50 took       41.241s\n",
      "  training loss:               0.270140\n",
      "  holdout loss (batch):        0.235832\n",
      "  Subject 1: median (max) validation accuracy = 0.807 (0.979)\n",
      "  Subject 2: median (max) validation accuracy = 0.839 (0.979)\n",
      "  Subject 3: median (max) validation accuracy = 0.883 (0.972)\n",
      "  Subject 4: median (max) validation accuracy = 0.905 (0.978)\n",
      "  Subject 5: median (max) validation accuracy = 0.936 (0.982)\n",
      "  Subject 6: median (max) validation accuracy = 0.903 (0.980)\n",
      "  Subject 7: median (max) validation accuracy = 0.884 (0.972)\n",
      "  Subject 8: median (max) validation accuracy = 0.913 (0.984)\n",
      "** Saving params with joint score = 0.893 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 6 of 50 took       41.237s\n",
      "  training loss:               0.247956\n",
      "  holdout loss (batch):        0.213389\n",
      "  Subject 1: median (max) validation accuracy = 0.827 (0.981)\n",
      "  Subject 2: median (max) validation accuracy = 0.857 (0.981)\n",
      "  Subject 3: median (max) validation accuracy = 0.896 (0.977)\n",
      "  Subject 4: median (max) validation accuracy = 0.916 (0.980)\n",
      "  Subject 5: median (max) validation accuracy = 0.943 (0.985)\n",
      "  Subject 6: median (max) validation accuracy = 0.914 (0.983)\n",
      "  Subject 7: median (max) validation accuracy = 0.898 (0.975)\n",
      "  Subject 8: median (max) validation accuracy = 0.923 (0.986)\n",
      "** Saving params with joint score = 0.906 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 7 of 50 took       41.245s\n",
      "  training loss:               0.227553\n",
      "  holdout loss (batch):        0.195638\n",
      "  Subject 1: median (max) validation accuracy = 0.846 (0.983)\n",
      "  Subject 2: median (max) validation accuracy = 0.872 (0.983)\n",
      "  Subject 3: median (max) validation accuracy = 0.908 (0.979)\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "  Subject 4: median (max) validation accuracy = 0.925 (0.982)\n",
      "  Subject 5: median (max) validation accuracy = 0.949 (0.986)\n",
      "  Subject 6: median (max) validation accuracy = 0.923 (0.985)\n",
      "  Subject 7: median (max) validation accuracy = 0.910 (0.977)\n",
      "  Subject 8: median (max) validation accuracy = 0.931 (0.987)\n",
      "** Saving params with joint score = 0.916 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 8 of 50 took       41.243s\n",
      "  training loss:               0.208564\n",
      "  holdout loss (batch):        0.173544\n",
      "  Subject 1: median (max) validation accuracy = 0.866 (0.985)\n",
      "  Subject 2: median (max) validation accuracy = 0.887 (0.985)\n",
      "  Subject 3: median (max) validation accuracy = 0.920 (0.982)\n",
      "  Subject 4: median (max) validation accuracy = 0.934 (0.984)\n",
      "  Subject 5: median (max) validation accuracy = 0.954 (0.988)\n",
      "  Subject 6: median (max) validation accuracy = 0.931 (0.987)\n",
      "  Subject 7: median (max) validation accuracy = 0.920 (0.979)\n",
      "  Subject 8: median (max) validation accuracy = 0.938 (0.989)\n",
      "** Saving params with joint score = 0.925 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 9 of 50 took       41.243s\n",
      "  training loss:               0.190035\n",
      "  holdout loss (batch):        0.157664\n",
      "  Subject 1: median (max) validation accuracy = 0.883 (0.986)\n",
      "  Subject 2: median (max) validation accuracy = 0.901 (0.986)\n",
      "  Subject 3: median (max) validation accuracy = 0.929 (0.983)\n",
      "  Subject 4: median (max) validation accuracy = 0.941 (0.985)\n",
      "  Subject 5: median (max) validation accuracy = 0.958 (0.989)\n",
      "  Subject 6: median (max) validation accuracy = 0.938 (0.988)\n",
      "  Subject 7: median (max) validation accuracy = 0.930 (0.980)\n",
      "  Subject 8: median (max) validation accuracy = 0.944 (0.989)\n",
      "** Saving params with joint score = 0.933 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 10 of 50 took       41.231s\n",
      "  training loss:               0.172794\n",
      "  holdout loss (batch):        0.140905\n",
      "  Subject 1: median (max) validation accuracy = 0.899 (0.987)\n",
      "  Subject 2: median (max) validation accuracy = 0.914 (0.987)\n",
      "  Subject 3: median (max) validation accuracy = 0.939 (0.986)\n",
      "  Subject 4: median (max) validation accuracy = 0.948 (0.986)\n",
      "  Subject 5: median (max) validation accuracy = 0.961 (0.990)\n",
      "  Subject 6: median (max) validation accuracy = 0.945 (0.989)\n",
      "  Subject 7: median (max) validation accuracy = 0.938 (0.981)\n",
      "  Subject 8: median (max) validation accuracy = 0.950 (0.990)\n",
      "** Saving params with joint score = 0.941 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 11 of 50 took       41.234s\n",
      "  training loss:               0.156285\n",
      "  holdout loss (batch):        0.126295\n",
      "  Subject 1: median (max) validation accuracy = 0.915 (0.988)\n",
      "  Subject 2: median (max) validation accuracy = 0.927 (0.988)\n",
      "  Subject 3: median (max) validation accuracy = 0.947 (0.987)\n",
      "  Subject 4: median (max) validation accuracy = 0.954 (0.987)\n",
      "  Subject 5: median (max) validation accuracy = 0.965 (0.991)\n",
      "  Subject 6: median (max) validation accuracy = 0.951 (0.989)\n",
      "  Subject 7: median (max) validation accuracy = 0.947 (0.983)\n",
      "  Subject 8: median (max) validation accuracy = 0.956 (0.990)\n",
      "** Saving params with joint score = 0.948 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 12 of 50 took       41.225s\n",
      "  training loss:               0.141795\n",
      "  holdout loss (batch):        0.108949\n",
      "  Subject 1: median (max) validation accuracy = 0.928 (0.989)\n",
      "  Subject 2: median (max) validation accuracy = 0.938 (0.989)\n",
      "  Subject 3: median (max) validation accuracy = 0.955 (0.989)\n",
      "  Subject 4: median (max) validation accuracy = 0.960 (0.988)\n",
      "  Subject 5: median (max) validation accuracy = 0.967 (0.992)\n",
      "  Subject 6: median (max) validation accuracy = 0.957 (0.990)\n",
      "  Subject 7: median (max) validation accuracy = 0.954 (0.984)\n",
      "  Subject 8: median (max) validation accuracy = 0.962 (0.991)\n",
      "** Saving params with joint score = 0.954 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 13 of 50 took       41.245s\n",
      "  training loss:               0.129384\n",
      "  holdout loss (batch):        0.098155\n",
      "  Subject 1: median (max) validation accuracy = 0.938 (0.990)\n",
      "  Subject 2: median (max) validation accuracy = 0.946 (0.989)\n",
      "  Subject 3: median (max) validation accuracy = 0.961 (0.990)\n",
      "  Subject 4: median (max) validation accuracy = 0.965 (0.989)\n",
      "  Subject 5: median (max) validation accuracy = 0.969 (0.992)\n",
      "  Subject 6: median (max) validation accuracy = 0.962 (0.991)\n",
      "  Subject 7: median (max) validation accuracy = 0.960 (0.985)\n",
      "  Subject 8: median (max) validation accuracy = 0.965 (0.992)\n",
      "** Saving params with joint score = 0.959 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 14 of 50 took       41.201s\n",
      "  training loss:               0.119099\n",
      "  holdout loss (batch):        0.087289\n",
      "  Subject 1: median (max) validation accuracy = 0.946 (0.990)\n",
      "  Subject 2: median (max) validation accuracy = 0.953 (0.990)\n",
      "  Subject 3: median (max) validation accuracy = 0.965 (0.991)\n",
      "  Subject 4: median (max) validation accuracy = 0.968 (0.990)\n",
      "  Subject 5: median (max) validation accuracy = 0.971 (0.993)\n",
      "  Subject 6: median (max) validation accuracy = 0.966 (0.992)\n",
      "  Subject 7: median (max) validation accuracy = 0.964 (0.987)\n",
      "  Subject 8: median (max) validation accuracy = 0.969 (0.992)\n",
      "** Saving params with joint score = 0.964 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 15 of 50 took       41.214s\n",
      "  training loss:               0.111147\n",
      "  holdout loss (batch):        0.081316\n",
      "  Subject 1: median (max) validation accuracy = 0.952 (0.991)\n",
      "  Subject 2: median (max) validation accuracy = 0.957 (0.990)\n",
      "  Subject 3: median (max) validation accuracy = 0.969 (0.991)\n",
      "  Subject 4: median (max) validation accuracy = 0.971 (0.991)\n",
      "  Subject 5: median (max) validation accuracy = 0.973 (0.993)\n",
      "  Subject 6: median (max) validation accuracy = 0.969 (0.993)\n",
      "  Subject 7: median (max) validation accuracy = 0.967 (0.988)\n",
      "  Subject 8: median (max) validation accuracy = 0.971 (0.992)\n",
      "** Saving params with joint score = 0.967 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 16 of 50 took       41.212s\n",
      "  training loss:               0.104923\n",
      "  holdout loss (batch):        0.076108\n",
      "  Subject 1: median (max) validation accuracy = 0.957 (0.991)\n",
      "  Subject 2: median (max) validation accuracy = 0.961 (0.990)\n",
      "  Subject 3: median (max) validation accuracy = 0.971 (0.991)\n",
      "  Subject 4: median (max) validation accuracy = 0.973 (0.991)\n",
      "  Subject 5: median (max) validation accuracy = 0.973 (0.993)\n",
      "  Subject 6: median (max) validation accuracy = 0.972 (0.993)\n",
      "  Subject 7: median (max) validation accuracy = 0.970 (0.989)\n",
      "  Subject 8: median (max) validation accuracy = 0.973 (0.993)\n",
      "** Saving params with joint score = 0.969 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 17 of 50 took       41.247s\n",
      "  training loss:               0.100118\n",
      "  holdout loss (batch):        0.070375\n",
      "  Subject 1: median (max) validation accuracy = 0.960 (0.992)\n",
      "  Subject 2: median (max) validation accuracy = 0.965 (0.991)\n",
      "  Subject 3: median (max) validation accuracy = 0.973 (0.992)\n",
      "  Subject 4: median (max) validation accuracy = 0.975 (0.992)\n",
      "  Subject 5: median (max) validation accuracy = 0.975 (0.993)\n",
      "  Subject 6: median (max) validation accuracy = 0.974 (0.994)\n",
      "  Subject 7: median (max) validation accuracy = 0.972 (0.990)\n",
      "  Subject 8: median (max) validation accuracy = 0.975 (0.993)\n",
      "** Saving params with joint score = 0.971 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 18 of 50 took       40.955s\n",
      "  training loss:               0.096626\n",
      "  holdout loss (batch):        0.068176\n",
      "  Subject 1: median (max) validation accuracy = 0.962 (0.992)\n",
      "  Subject 2: median (max) validation accuracy = 0.967 (0.991)\n",
      "  Subject 3: median (max) validation accuracy = 0.975 (0.992)\n",
      "  Subject 4: median (max) validation accuracy = 0.976 (0.992)\n",
      "  Subject 5: median (max) validation accuracy = 0.975 (0.994)\n",
      "  Subject 6: median (max) validation accuracy = 0.975 (0.994)\n",
      "  Subject 7: median (max) validation accuracy = 0.974 (0.990)\n",
      "  Subject 8: median (max) validation accuracy = 0.976 (0.993)\n",
      "** Saving params with joint score = 0.973 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 19 of 50 took       40.898s\n",
      "  training loss:               0.093850\n",
      "  holdout loss (batch):        0.068850\n",
      "  Subject 1: median (max) validation accuracy = 0.964 (0.992)\n",
      "  Subject 2: median (max) validation accuracy = 0.969 (0.991)\n",
      "  Subject 3: median (max) validation accuracy = 0.976 (0.993)\n",
      "  Subject 4: median (max) validation accuracy = 0.977 (0.992)\n",
      "  Subject 5: median (max) validation accuracy = 0.976 (0.994)\n",
      "  Subject 6: median (max) validation accuracy = 0.976 (0.994)\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "  Subject 7: median (max) validation accuracy = 0.975 (0.990)\n",
      "  Subject 8: median (max) validation accuracy = 0.977 (0.994)\n",
      "** Saving params with joint score = 0.974 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 20 of 50 took       40.926s\n",
      "  training loss:               0.091470\n",
      "  holdout loss (batch):        0.064162\n",
      "  Subject 1: median (max) validation accuracy = 0.966 (0.993)\n",
      "  Subject 2: median (max) validation accuracy = 0.970 (0.991)\n",
      "  Subject 3: median (max) validation accuracy = 0.977 (0.993)\n",
      "  Subject 4: median (max) validation accuracy = 0.978 (0.992)\n",
      "  Subject 5: median (max) validation accuracy = 0.977 (0.994)\n",
      "  Subject 6: median (max) validation accuracy = 0.977 (0.995)\n",
      "  Subject 7: median (max) validation accuracy = 0.976 (0.991)\n",
      "  Subject 8: median (max) validation accuracy = 0.978 (0.994)\n",
      "** Saving params with joint score = 0.975 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 21 of 50 took       40.917s\n",
      "  training loss:               0.089335\n",
      "  holdout loss (batch):        0.061448\n",
      "  Subject 1: median (max) validation accuracy = 0.968 (0.993)\n",
      "  Subject 2: median (max) validation accuracy = 0.971 (0.991)\n",
      "  Subject 3: median (max) validation accuracy = 0.978 (0.993)\n",
      "  Subject 4: median (max) validation accuracy = 0.979 (0.993)\n",
      "  Subject 5: median (max) validation accuracy = 0.977 (0.994)\n",
      "  Subject 6: median (max) validation accuracy = 0.978 (0.995)\n",
      "  Subject 7: median (max) validation accuracy = 0.977 (0.992)\n",
      "  Subject 8: median (max) validation accuracy = 0.979 (0.994)\n",
      "** Saving params with joint score = 0.976 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 22 of 50 took       40.912s\n",
      "  training loss:               0.087840\n",
      "  holdout loss (batch):        0.059750\n",
      "  Subject 1: median (max) validation accuracy = 0.970 (0.993)\n",
      "  Subject 2: median (max) validation accuracy = 0.972 (0.992)\n",
      "  Subject 3: median (max) validation accuracy = 0.978 (0.993)\n",
      "  Subject 4: median (max) validation accuracy = 0.980 (0.993)\n",
      "  Subject 5: median (max) validation accuracy = 0.977 (0.994)\n",
      "  Subject 6: median (max) validation accuracy = 0.979 (0.995)\n",
      "  Subject 7: median (max) validation accuracy = 0.978 (0.992)\n",
      "  Subject 8: median (max) validation accuracy = 0.979 (0.995)\n",
      "** Saving params with joint score = 0.977 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 23 of 50 took       40.920s\n",
      "  training loss:               0.086377\n",
      "  holdout loss (batch):        0.061216\n",
      "  Subject 1: median (max) validation accuracy = 0.971 (0.993)\n",
      "  Subject 2: median (max) validation accuracy = 0.973 (0.991)\n",
      "  Subject 3: median (max) validation accuracy = 0.979 (0.993)\n",
      "  Subject 4: median (max) validation accuracy = 0.980 (0.992)\n",
      "  Subject 5: median (max) validation accuracy = 0.977 (0.994)\n",
      "  Subject 6: median (max) validation accuracy = 0.979 (0.995)\n",
      "  Subject 7: median (max) validation accuracy = 0.978 (0.992)\n",
      "  Subject 8: median (max) validation accuracy = 0.979 (0.994)\n",
      "** Saving params with joint score = 0.977 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 24 of 50 took       40.915s\n",
      "  training loss:               0.085227\n",
      "  holdout loss (batch):        0.057698\n",
      "  Subject 1: median (max) validation accuracy = 0.972 (0.993)\n",
      "  Subject 2: median (max) validation accuracy = 0.974 (0.992)\n",
      "  Subject 3: median (max) validation accuracy = 0.979 (0.993)\n",
      "  Subject 4: median (max) validation accuracy = 0.981 (0.993)\n",
      "  Subject 5: median (max) validation accuracy = 0.978 (0.994)\n",
      "  Subject 6: median (max) validation accuracy = 0.980 (0.995)\n",
      "  Subject 7: median (max) validation accuracy = 0.978 (0.993)\n",
      "  Subject 8: median (max) validation accuracy = 0.980 (0.994)\n",
      "** Saving params with joint score = 0.978 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 25 of 50 took       40.943s\n",
      "  training loss:               0.084309\n",
      "  holdout loss (batch):        0.058087\n",
      "  Subject 1: median (max) validation accuracy = 0.973 (0.993)\n",
      "  Subject 2: median (max) validation accuracy = 0.975 (0.992)\n",
      "  Subject 3: median (max) validation accuracy = 0.980 (0.993)\n",
      "  Subject 4: median (max) validation accuracy = 0.980 (0.993)\n",
      "  Subject 5: median (max) validation accuracy = 0.978 (0.994)\n",
      "  Subject 6: median (max) validation accuracy = 0.980 (0.995)\n",
      "  Subject 7: median (max) validation accuracy = 0.979 (0.993)\n",
      "  Subject 8: median (max) validation accuracy = 0.980 (0.994)\n",
      "** Saving params with joint score = 0.978 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 26 of 50 took       40.955s\n",
      "  training loss:               0.083092\n",
      "  holdout loss (batch):        0.055007\n",
      "  Subject 1: median (max) validation accuracy = 0.973 (0.994)\n",
      "  Subject 2: median (max) validation accuracy = 0.975 (0.992)\n",
      "  Subject 3: median (max) validation accuracy = 0.980 (0.994)\n",
      "  Subject 4: median (max) validation accuracy = 0.981 (0.993)\n",
      "  Subject 5: median (max) validation accuracy = 0.979 (0.995)\n",
      "  Subject 6: median (max) validation accuracy = 0.980 (0.995)\n",
      "  Subject 7: median (max) validation accuracy = 0.979 (0.993)\n",
      "  Subject 8: median (max) validation accuracy = 0.980 (0.995)\n",
      "** Saving params with joint score = 0.979 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 27 of 50 took       40.919s\n",
      "  training loss:               0.082321\n",
      "  holdout loss (batch):        0.056024\n",
      "  Subject 1: median (max) validation accuracy = 0.974 (0.994)\n",
      "  Subject 2: median (max) validation accuracy = 0.976 (0.992)\n",
      "  Subject 3: median (max) validation accuracy = 0.981 (0.994)\n",
      "  Subject 4: median (max) validation accuracy = 0.981 (0.993)\n",
      "  Subject 5: median (max) validation accuracy = 0.979 (0.994)\n",
      "  Subject 6: median (max) validation accuracy = 0.981 (0.995)\n",
      "  Subject 7: median (max) validation accuracy = 0.979 (0.993)\n",
      "  Subject 8: median (max) validation accuracy = 0.981 (0.995)\n",
      "** Saving params with joint score = 0.979 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 28 of 50 took       40.990s\n",
      "  training loss:               0.081650\n",
      "  holdout loss (batch):        0.053612\n",
      "  Subject 1: median (max) validation accuracy = 0.975 (0.994)\n",
      "  Subject 2: median (max) validation accuracy = 0.976 (0.992)\n",
      "  Subject 3: median (max) validation accuracy = 0.981 (0.994)\n",
      "  Subject 4: median (max) validation accuracy = 0.982 (0.994)\n",
      "  Subject 5: median (max) validation accuracy = 0.979 (0.994)\n",
      "  Subject 6: median (max) validation accuracy = 0.981 (0.995)\n",
      "  Subject 7: median (max) validation accuracy = 0.980 (0.993)\n",
      "  Subject 8: median (max) validation accuracy = 0.981 (0.995)\n",
      "** Saving params with joint score = 0.979 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 29 of 50 took       40.946s\n",
      "  training loss:               0.080838\n",
      "  holdout loss (batch):        0.053239\n",
      "  Subject 1: median (max) validation accuracy = 0.975 (0.994)\n",
      "  Subject 2: median (max) validation accuracy = 0.977 (0.993)\n",
      "  Subject 3: median (max) validation accuracy = 0.981 (0.994)\n",
      "  Subject 4: median (max) validation accuracy = 0.982 (0.994)\n",
      "  Subject 5: median (max) validation accuracy = 0.979 (0.995)\n",
      "  Subject 6: median (max) validation accuracy = 0.981 (0.995)\n",
      "  Subject 7: median (max) validation accuracy = 0.980 (0.994)\n",
      "  Subject 8: median (max) validation accuracy = 0.981 (0.994)\n",
      "** Saving params with joint score = 0.980 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 30 of 50 took       40.928s\n",
      "  training loss:               0.080391\n",
      "  holdout loss (batch):        0.053988\n",
      "  Subject 1: median (max) validation accuracy = 0.975 (0.994)\n",
      "  Subject 2: median (max) validation accuracy = 0.977 (0.993)\n",
      "  Subject 3: median (max) validation accuracy = 0.981 (0.994)\n",
      "  Subject 4: median (max) validation accuracy = 0.982 (0.994)\n",
      "  Subject 5: median (max) validation accuracy = 0.979 (0.995)\n",
      "  Subject 6: median (max) validation accuracy = 0.981 (0.995)\n",
      "  Subject 7: median (max) validation accuracy = 0.980 (0.993)\n",
      "  Subject 8: median (max) validation accuracy = 0.981 (0.995)\n",
      "** Saving params with joint score = 0.980 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 31 of 50 took       41.006s\n",
      "  training loss:               0.079668\n",
      "  holdout loss (batch):        0.054021\n",
      "  Subject 1: median (max) validation accuracy = 0.976 (0.994)\n",
      "  Subject 2: median (max) validation accuracy = 0.978 (0.993)\n",
      "  Subject 3: median (max) validation accuracy = 0.982 (0.994)\n",
      "  Subject 4: median (max) validation accuracy = 0.982 (0.994)\n",
      "  Subject 5: median (max) validation accuracy = 0.979 (0.995)\n",
      "  Subject 6: median (max) validation accuracy = 0.982 (0.996)\n",
      "  Subject 7: median (max) validation accuracy = 0.980 (0.993)\n",
      "  Subject 8: median (max) validation accuracy = 0.981 (0.995)\n",
      "** Saving params with joint score = 0.980 **\n",
      "\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\\ : 100.0 %\n",
      "  Epoch 32 of 50 took       40.966s\n",
      "  training loss:               0.079236\n",
      "  holdout loss (batch):        0.052198\n",
      "  Subject 1: median (max) validation accuracy = 0.977 (0.994)\n",
      "  Subject 2: median (max) validation accuracy = 0.978 (0.993)\n",
      "  Subject 3: median (max) validation accuracy = 0.982 (0.994)\n",
      "  Subject 4: median (max) validation accuracy = 0.982 (0.994)\n",
      "  Subject 5: median (max) validation accuracy = 0.979 (0.995)\n",
      "  Subject 6: median (max) validation accuracy = 0.982 (0.996)\n",
      "  Subject 7: median (max) validation accuracy = 0.981 (0.994)\n",
      "  Subject 8: median (max) validation accuracy = 0.981 (0.995)\n",
      "** Saving params with joint score = 0.980 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 33 of 50 took       41.053s\n",
      "  training loss:               0.078912\n",
      "  holdout loss (batch):        0.051554\n",
      "  Subject 1: median (max) validation accuracy = 0.977 (0.994)\n",
      "  Subject 2: median (max) validation accuracy = 0.978 (0.993)\n",
      "  Subject 3: median (max) validation accuracy = 0.982 (0.994)\n",
      "  Subject 4: median (max) validation accuracy = 0.982 (0.994)\n",
      "  Subject 5: median (max) validation accuracy = 0.979 (0.995)\n",
      "  Subject 6: median (max) validation accuracy = 0.982 (0.996)\n",
      "  Subject 7: median (max) validation accuracy = 0.981 (0.993)\n",
      "  Subject 8: median (max) validation accuracy = 0.982 (0.995)\n",
      "** Saving params with joint score = 0.980 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 34 of 50 took       40.973s\n",
      "  training loss:               0.078747\n",
      "  holdout loss (batch):        0.050968\n",
      "  Subject 1: median (max) validation accuracy = 0.977 (0.994)\n",
      "  Subject 2: median (max) validation accuracy = 0.979 (0.993)\n",
      "  Subject 3: median (max) validation accuracy = 0.982 (0.994)\n",
      "  Subject 4: median (max) validation accuracy = 0.982 (0.994)\n",
      "  Subject 5: median (max) validation accuracy = 0.980 (0.995)\n",
      "  Subject 6: median (max) validation accuracy = 0.982 (0.996)\n",
      "  Subject 7: median (max) validation accuracy = 0.981 (0.994)\n",
      "  Subject 8: median (max) validation accuracy = 0.982 (0.995)\n",
      "** Saving params with joint score = 0.981 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 35 of 50 took       40.998s\n",
      "  training loss:               0.078129\n",
      "  holdout loss (batch):        0.051277\n",
      "  Subject 1: median (max) validation accuracy = 0.977 (0.994)\n",
      "  Subject 2: median (max) validation accuracy = 0.979 (0.993)\n",
      "  Subject 3: median (max) validation accuracy = 0.982 (0.994)\n",
      "  Subject 4: median (max) validation accuracy = 0.983 (0.994)\n",
      "  Subject 5: median (max) validation accuracy = 0.979 (0.995)\n",
      "  Subject 6: median (max) validation accuracy = 0.982 (0.996)\n",
      "  Subject 7: median (max) validation accuracy = 0.981 (0.994)\n",
      "  Subject 8: median (max) validation accuracy = 0.982 (0.995)\n",
      "** Saving params with joint score = 0.981 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 36 of 50 took       40.945s\n",
      "  training loss:               0.077655\n",
      "  holdout loss (batch):        0.050166\n",
      "  Subject 1: median (max) validation accuracy = 0.978 (0.994)\n",
      "  Subject 2: median (max) validation accuracy = 0.979 (0.993)\n",
      "  Subject 3: median (max) validation accuracy = 0.982 (0.994)\n",
      "  Subject 4: median (max) validation accuracy = 0.983 (0.994)\n",
      "  Subject 5: median (max) validation accuracy = 0.980 (0.995)\n",
      "  Subject 6: median (max) validation accuracy = 0.982 (0.996)\n",
      "  Subject 7: median (max) validation accuracy = 0.981 (0.994)\n",
      "  Subject 8: median (max) validation accuracy = 0.982 (0.995)\n",
      "** Saving params with joint score = 0.981 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 37 of 50 took       40.985s\n",
      "  training loss:               0.077806\n",
      "  holdout loss (batch):        0.050660\n",
      "  Subject 1: median (max) validation accuracy = 0.978 (0.994)\n",
      "  Subject 2: median (max) validation accuracy = 0.979 (0.993)\n",
      "  Subject 3: median (max) validation accuracy = 0.982 (0.994)\n",
      "  Subject 4: median (max) validation accuracy = 0.983 (0.994)\n",
      "  Subject 5: median (max) validation accuracy = 0.980 (0.995)\n",
      "  Subject 6: median (max) validation accuracy = 0.982 (0.996)\n",
      "  Subject 7: median (max) validation accuracy = 0.981 (0.994)\n",
      "  Subject 8: median (max) validation accuracy = 0.982 (0.995)\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 38 of 50 took       40.935s\n",
      "  training loss:               0.077277\n",
      "  holdout loss (batch):        0.050322\n",
      "  Subject 1: median (max) validation accuracy = 0.978 (0.994)\n",
      "  Subject 2: median (max) validation accuracy = 0.979 (0.993)\n",
      "  Subject 3: median (max) validation accuracy = 0.983 (0.994)\n",
      "  Subject 4: median (max) validation accuracy = 0.983 (0.994)\n",
      "  Subject 5: median (max) validation accuracy = 0.980 (0.995)\n",
      "  Subject 6: median (max) validation accuracy = 0.982 (0.996)\n",
      "  Subject 7: median (max) validation accuracy = 0.981 (0.994)\n",
      "  Subject 8: median (max) validation accuracy = 0.982 (0.995)\n",
      "** Saving params with joint score = 0.981 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 39 of 50 took       40.921s\n",
      "  training loss:               0.076936\n",
      "  holdout loss (batch):        0.050597\n",
      "  Subject 1: median (max) validation accuracy = 0.978 (0.994)\n",
      "  Subject 2: median (max) validation accuracy = 0.979 (0.993)\n",
      "  Subject 3: median (max) validation accuracy = 0.982 (0.994)\n",
      "  Subject 4: median (max) validation accuracy = 0.983 (0.994)\n",
      "  Subject 5: median (max) validation accuracy = 0.980 (0.995)\n",
      "  Subject 6: median (max) validation accuracy = 0.983 (0.996)\n",
      "  Subject 7: median (max) validation accuracy = 0.981 (0.994)\n",
      "  Subject 8: median (max) validation accuracy = 0.982 (0.995)\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 40 of 50 took       40.918s\n",
      "  training loss:               0.076826\n",
      "  holdout loss (batch):        0.052917\n",
      "  Subject 1: median (max) validation accuracy = 0.978 (0.994)\n",
      "  Subject 2: median (max) validation accuracy = 0.979 (0.993)\n",
      "  Subject 3: median (max) validation accuracy = 0.983 (0.994)\n",
      "  Subject 4: median (max) validation accuracy = 0.983 (0.994)\n",
      "  Subject 5: median (max) validation accuracy = 0.980 (0.995)\n",
      "  Subject 6: median (max) validation accuracy = 0.982 (0.996)\n",
      "  Subject 7: median (max) validation accuracy = 0.981 (0.994)\n",
      "  Subject 8: median (max) validation accuracy = 0.982 (0.995)\n",
      "** Saving params with joint score = 0.981 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 41 of 50 took       40.913s\n",
      "  training loss:               0.076402\n",
      "  holdout loss (batch):        0.049640\n",
      "  Subject 1: median (max) validation accuracy = 0.978 (0.995)\n",
      "  Subject 2: median (max) validation accuracy = 0.980 (0.993)\n",
      "  Subject 3: median (max) validation accuracy = 0.983 (0.994)\n",
      "  Subject 4: median (max) validation accuracy = 0.983 (0.994)\n",
      "  Subject 5: median (max) validation accuracy = 0.980 (0.995)\n",
      "  Subject 6: median (max) validation accuracy = 0.983 (0.996)\n",
      "  Subject 7: median (max) validation accuracy = 0.981 (0.994)\n",
      "  Subject 8: median (max) validation accuracy = 0.982 (0.995)\n",
      "** Saving params with joint score = 0.981 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 42 of 50 took       40.915s\n",
      "  training loss:               0.076279\n",
      "  holdout loss (batch):        0.049084\n",
      "  Subject 1: median (max) validation accuracy = 0.979 (0.994)\n",
      "  Subject 2: median (max) validation accuracy = 0.979 (0.993)\n",
      "  Subject 3: median (max) validation accuracy = 0.983 (0.994)\n",
      "  Subject 4: median (max) validation accuracy = 0.983 (0.994)\n",
      "  Subject 5: median (max) validation accuracy = 0.980 (0.995)\n",
      "  Subject 6: median (max) validation accuracy = 0.983 (0.996)\n",
      "  Subject 7: median (max) validation accuracy = 0.981 (0.994)\n",
      "  Subject 8: median (max) validation accuracy = 0.982 (0.995)\n",
      "** Saving params with joint score = 0.981 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 43 of 50 took       40.939s\n",
      "  training loss:               0.076105\n",
      "  holdout loss (batch):        0.049719\n",
      "  Subject 1: median (max) validation accuracy = 0.979 (0.995)\n",
      "  Subject 2: median (max) validation accuracy = 0.980 (0.993)\n",
      "  Subject 3: median (max) validation accuracy = 0.983 (0.994)\n",
      "  Subject 4: median (max) validation accuracy = 0.983 (0.994)\n",
      "  Subject 5: median (max) validation accuracy = 0.980 (0.995)\n",
      "  Subject 6: median (max) validation accuracy = 0.983 (0.996)\n",
      "  Subject 7: median (max) validation accuracy = 0.981 (0.994)\n",
      "  Subject 8: median (max) validation accuracy = 0.982 (0.995)\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 44 of 50 took       40.948s\n",
      "  training loss:               0.075975\n",
      "  holdout loss (batch):        0.050137\n",
      "  Subject 1: median (max) validation accuracy = 0.979 (0.995)\n",
      "  Subject 2: median (max) validation accuracy = 0.980 (0.993)\n",
      "  Subject 3: median (max) validation accuracy = 0.983 (0.994)\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "  Subject 4: median (max) validation accuracy = 0.983 (0.994)\n",
      "  Subject 5: median (max) validation accuracy = 0.980 (0.995)\n",
      "  Subject 6: median (max) validation accuracy = 0.983 (0.996)\n",
      "  Subject 7: median (max) validation accuracy = 0.981 (0.994)\n",
      "  Subject 8: median (max) validation accuracy = 0.982 (0.995)\n",
      "** Saving params with joint score = 0.981 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 45 of 50 took       41.086s\n",
      "  training loss:               0.075907\n",
      "  holdout loss (batch):        0.051934\n",
      "  Subject 1: median (max) validation accuracy = 0.979 (0.994)\n",
      "  Subject 2: median (max) validation accuracy = 0.980 (0.993)\n",
      "  Subject 3: median (max) validation accuracy = 0.983 (0.994)\n",
      "  Subject 4: median (max) validation accuracy = 0.983 (0.993)\n",
      "  Subject 5: median (max) validation accuracy = 0.980 (0.995)\n",
      "  Subject 6: median (max) validation accuracy = 0.983 (0.996)\n",
      "  Subject 7: median (max) validation accuracy = 0.981 (0.994)\n",
      "  Subject 8: median (max) validation accuracy = 0.982 (0.995)\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 46 of 50 took       41.354s\n",
      "  training loss:               0.075620\n",
      "  holdout loss (batch):        0.047741\n",
      "  Subject 1: median (max) validation accuracy = 0.979 (0.995)\n",
      "  Subject 2: median (max) validation accuracy = 0.980 (0.993)\n",
      "  Subject 3: median (max) validation accuracy = 0.983 (0.994)\n",
      "  Subject 4: median (max) validation accuracy = 0.983 (0.994)\n",
      "  Subject 5: median (max) validation accuracy = 0.980 (0.995)\n",
      "  Subject 6: median (max) validation accuracy = 0.983 (0.996)\n",
      "  Subject 7: median (max) validation accuracy = 0.982 (0.994)\n",
      "  Subject 8: median (max) validation accuracy = 0.982 (0.995)\n",
      "** Saving params with joint score = 0.982 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 47 of 50 took       41.352s\n",
      "  training loss:               0.075526\n",
      "  holdout loss (batch):        0.048907\n",
      "  Subject 1: median (max) validation accuracy = 0.979 (0.995)\n",
      "  Subject 2: median (max) validation accuracy = 0.980 (0.993)\n",
      "  Subject 3: median (max) validation accuracy = 0.983 (0.994)\n",
      "  Subject 4: median (max) validation accuracy = 0.984 (0.994)\n",
      "  Subject 5: median (max) validation accuracy = 0.981 (0.995)\n",
      "  Subject 6: median (max) validation accuracy = 0.983 (0.996)\n",
      "  Subject 7: median (max) validation accuracy = 0.982 (0.994)\n",
      "  Subject 8: median (max) validation accuracy = 0.982 (0.995)\n",
      "** Saving params with joint score = 0.982 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 48 of 50 took       40.915s\n",
      "  training loss:               0.075488\n",
      "  holdout loss (batch):        0.049800\n",
      "  Subject 1: median (max) validation accuracy = 0.979 (0.995)\n",
      "  Subject 2: median (max) validation accuracy = 0.980 (0.993)\n",
      "  Subject 3: median (max) validation accuracy = 0.983 (0.994)\n",
      "  Subject 4: median (max) validation accuracy = 0.983 (0.993)\n",
      "  Subject 5: median (max) validation accuracy = 0.980 (0.995)\n",
      "  Subject 6: median (max) validation accuracy = 0.983 (0.996)\n",
      "  Subject 7: median (max) validation accuracy = 0.981 (0.994)\n",
      "  Subject 8: median (max) validation accuracy = 0.982 (0.994)\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 49 of 50 took       40.909s\n",
      "  training loss:               0.075264\n",
      "  holdout loss (batch):        0.049056\n",
      "  Subject 1: median (max) validation accuracy = 0.979 (0.994)\n",
      "  Subject 2: median (max) validation accuracy = 0.980 (0.994)\n",
      "  Subject 3: median (max) validation accuracy = 0.983 (0.994)\n",
      "  Subject 4: median (max) validation accuracy = 0.983 (0.994)\n",
      "  Subject 5: median (max) validation accuracy = 0.980 (0.995)\n",
      "  Subject 6: median (max) validation accuracy = 0.983 (0.996)\n",
      "  Subject 7: median (max) validation accuracy = 0.982 (0.994)\n",
      "  Subject 8: median (max) validation accuracy = 0.982 (0.995)\n",
      "** Saving params with joint score = 0.982 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 50 of 50 took       40.934s\n",
      "  training loss:               0.075504\n",
      "  holdout loss (batch):        0.047475\n",
      "  Subject 1: median (max) validation accuracy = 0.979 (0.995)\n",
      "  Subject 2: median (max) validation accuracy = 0.980 (0.994)\n",
      "  Subject 3: median (max) validation accuracy = 0.983 (0.994)\n",
      "  Subject 4: median (max) validation accuracy = 0.984 (0.994)\n",
      "  Subject 5: median (max) validation accuracy = 0.981 (0.995)\n",
      "  Subject 6: median (max) validation accuracy = 0.983 (0.996)\n",
      "  Subject 7: median (max) validation accuracy = 0.982 (0.994)\n",
      "  Subject 8: median (max) validation accuracy = 0.983 (0.995)\n",
      "** Saving params with joint score = 0.982 **\n",
      "\n",
      "***************************************\n",
      "best joint val cc = 0.983\n",
      "subject 1: val cc = 0.981\n",
      "subject 2: val cc = 0.981\n",
      "subject 3: val cc = 0.984\n",
      "subject 4: val cc = 0.984\n",
      "subject 5: val cc = 0.982\n",
      "subject 6: val cc = 0.984\n",
      "subject 7: val cc = 0.983\n",
      "subject 8: val cc = 0.983\n",
      "***************************************\n",
      "/export/raid1/home/styvesg/code/nsd_gnet8x/output/multisubject/gnet8r_pred_Feb-07-2022_1721/hV4/\n",
      "***************************************\n",
      "***                 hV4       ***\n",
      "from ==> /export/raid1/home/styvesg/code/nsd_gnet8x/output/multisubject/gnet8r_pred_Feb-07-2022_1721//model_params_hV4_no_noise\n",
      "to   ==> /export/raid1/home/styvesg/code/nsd_gnet8x/output/multisubject/gnet8r_pred_Feb-07-2022_1721/hV4/\n",
      "***************************************\n",
      "subject 1 masked 4627 of 4627\n",
      "subject 1 training/holdout 8100 900\n",
      "subject 2 masked 3765 of 3765\n",
      "subject 2 training/holdout 8100 900\n",
      "subject 3 masked 3778 of 3778\n",
      "subject 3 training/holdout 8100 900\n",
      "subject 4 masked 3027 of 3027\n",
      "subject 4 training/holdout 8100 900\n",
      "subject 5 masked 3685 of 3685\n",
      "subject 5 training/holdout 8100 900\n",
      "subject 6 masked 4024 of 4024\n",
      "subject 6 training/holdout 8100 900\n",
      "subject 7 masked 3201 of 3201\n",
      "subject 7 training/holdout 8100 900\n",
      "subject 8 masked 3490 of 3490\n",
      "subject 8 training/holdout 8100 900\n",
      "\\ : 100.0 %\n",
      "  Epoch 1 of 50 took       40.904s\n",
      "  training loss:               0.599022\n",
      "  holdout loss (batch):        0.454154\n",
      "  Subject 1: median (max) validation accuracy = 0.617 (0.950)\n",
      "  Subject 2: median (max) validation accuracy = 0.663 (0.947)\n",
      "  Subject 3: median (max) validation accuracy = 0.744 (0.946)\n",
      "  Subject 4: median (max) validation accuracy = 0.769 (0.954)\n",
      "  Subject 5: median (max) validation accuracy = 0.841 (0.966)\n",
      "  Subject 6: median (max) validation accuracy = 0.766 (0.961)\n",
      "  Subject 7: median (max) validation accuracy = 0.725 (0.932)\n",
      "  Subject 8: median (max) validation accuracy = 0.794 (0.968)\n",
      "** Saving params with joint score = 0.748 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 2 of 50 took       41.344s\n",
      "  training loss:               0.433855\n",
      "  holdout loss (batch):        0.375146\n",
      "  Subject 1: median (max) validation accuracy = 0.681 (0.964)\n",
      "  Subject 2: median (max) validation accuracy = 0.721 (0.966)\n",
      "  Subject 3: median (max) validation accuracy = 0.792 (0.965)\n",
      "  Subject 4: median (max) validation accuracy = 0.823 (0.968)\n",
      "  Subject 5: median (max) validation accuracy = 0.882 (0.975)\n",
      "  Subject 6: median (max) validation accuracy = 0.818 (0.972)\n",
      "  Subject 7: median (max) validation accuracy = 0.782 (0.949)\n",
      "  Subject 8: median (max) validation accuracy = 0.837 (0.978)\n",
      "** Saving params with joint score = 0.804 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 3 of 50 took       41.339s\n",
      "  training loss:               0.376784\n",
      "  holdout loss (batch):        0.334164\n",
      "  Subject 1: median (max) validation accuracy = 0.717 (0.972)\n",
      "  Subject 2: median (max) validation accuracy = 0.752 (0.972)\n",
      "  Subject 3: median (max) validation accuracy = 0.819 (0.971)\n",
      "  Subject 4: median (max) validation accuracy = 0.850 (0.973)\n",
      "  Subject 5: median (max) validation accuracy = 0.899 (0.981)\n",
      "  Subject 6: median (max) validation accuracy = 0.844 (0.976)\n",
      "  Subject 7: median (max) validation accuracy = 0.812 (0.956)\n",
      "  Subject 8: median (max) validation accuracy = 0.859 (0.983)\n",
      "** Saving params with joint score = 0.832 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 4 of 50 took       41.306s\n",
      "  training loss:               0.340680\n",
      "  holdout loss (batch):        0.303687\n",
      "  Subject 1: median (max) validation accuracy = 0.745 (0.976)\n",
      "  Subject 2: median (max) validation accuracy = 0.776 (0.977)\n",
      "  Subject 3: median (max) validation accuracy = 0.841 (0.975)\n",
      "  Subject 4: median (max) validation accuracy = 0.868 (0.978)\n",
      "  Subject 5: median (max) validation accuracy = 0.911 (0.982)\n",
      "  Subject 6: median (max) validation accuracy = 0.862 (0.978)\n",
      "  Subject 7: median (max) validation accuracy = 0.834 (0.960)\n",
      "  Subject 8: median (max) validation accuracy = 0.875 (0.985)\n",
      "** Saving params with joint score = 0.851 **\n",
      "\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\\ : 100.0 %\n",
      "  Epoch 5 of 50 took       41.377s\n",
      "  training loss:               0.312070\n",
      "  holdout loss (batch):        0.274401\n",
      "  Subject 1: median (max) validation accuracy = 0.771 (0.978)\n",
      "  Subject 2: median (max) validation accuracy = 0.797 (0.979)\n",
      "  Subject 3: median (max) validation accuracy = 0.859 (0.978)\n",
      "  Subject 4: median (max) validation accuracy = 0.883 (0.980)\n",
      "  Subject 5: median (max) validation accuracy = 0.921 (0.983)\n",
      "  Subject 6: median (max) validation accuracy = 0.877 (0.980)\n",
      "  Subject 7: median (max) validation accuracy = 0.852 (0.964)\n",
      "  Subject 8: median (max) validation accuracy = 0.888 (0.987)\n",
      "** Saving params with joint score = 0.868 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 6 of 50 took       41.374s\n",
      "  training loss:               0.287859\n",
      "  holdout loss (batch):        0.254111\n",
      "  Subject 1: median (max) validation accuracy = 0.792 (0.980)\n",
      "  Subject 2: median (max) validation accuracy = 0.815 (0.980)\n",
      "  Subject 3: median (max) validation accuracy = 0.872 (0.979)\n",
      "  Subject 4: median (max) validation accuracy = 0.894 (0.982)\n",
      "  Subject 5: median (max) validation accuracy = 0.928 (0.984)\n",
      "  Subject 6: median (max) validation accuracy = 0.888 (0.981)\n",
      "  Subject 7: median (max) validation accuracy = 0.867 (0.968)\n",
      "  Subject 8: median (max) validation accuracy = 0.899 (0.988)\n",
      "** Saving params with joint score = 0.881 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 7 of 50 took       41.367s\n",
      "  training loss:               0.265216\n",
      "  holdout loss (batch):        0.231387\n",
      "  Subject 1: median (max) validation accuracy = 0.813 (0.983)\n",
      "  Subject 2: median (max) validation accuracy = 0.834 (0.982)\n",
      "  Subject 3: median (max) validation accuracy = 0.886 (0.982)\n",
      "  Subject 4: median (max) validation accuracy = 0.906 (0.983)\n",
      "  Subject 5: median (max) validation accuracy = 0.936 (0.986)\n",
      "  Subject 6: median (max) validation accuracy = 0.900 (0.984)\n",
      "  Subject 7: median (max) validation accuracy = 0.882 (0.970)\n",
      "  Subject 8: median (max) validation accuracy = 0.909 (0.990)\n",
      "** Saving params with joint score = 0.893 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 8 of 50 took       41.225s\n",
      "  training loss:               0.245277\n",
      "  holdout loss (batch):        0.212315\n",
      "  Subject 1: median (max) validation accuracy = 0.834 (0.984)\n",
      "  Subject 2: median (max) validation accuracy = 0.851 (0.984)\n",
      "  Subject 3: median (max) validation accuracy = 0.897 (0.983)\n",
      "  Subject 4: median (max) validation accuracy = 0.915 (0.985)\n",
      "  Subject 5: median (max) validation accuracy = 0.941 (0.986)\n",
      "  Subject 6: median (max) validation accuracy = 0.910 (0.986)\n",
      "  Subject 7: median (max) validation accuracy = 0.894 (0.973)\n",
      "  Subject 8: median (max) validation accuracy = 0.918 (0.990)\n",
      "** Saving params with joint score = 0.904 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 9 of 50 took       41.235s\n",
      "  training loss:               0.225614\n",
      "  holdout loss (batch):        0.191768\n",
      "  Subject 1: median (max) validation accuracy = 0.853 (0.986)\n",
      "  Subject 2: median (max) validation accuracy = 0.868 (0.984)\n",
      "  Subject 3: median (max) validation accuracy = 0.909 (0.985)\n",
      "  Subject 4: median (max) validation accuracy = 0.925 (0.986)\n",
      "  Subject 5: median (max) validation accuracy = 0.947 (0.988)\n",
      "  Subject 6: median (max) validation accuracy = 0.919 (0.987)\n",
      "  Subject 7: median (max) validation accuracy = 0.906 (0.974)\n",
      "  Subject 8: median (max) validation accuracy = 0.926 (0.991)\n",
      "** Saving params with joint score = 0.914 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 10 of 50 took       41.255s\n",
      "  training loss:               0.207596\n",
      "  holdout loss (batch):        0.173967\n",
      "  Subject 1: median (max) validation accuracy = 0.871 (0.986)\n",
      "  Subject 2: median (max) validation accuracy = 0.883 (0.985)\n",
      "  Subject 3: median (max) validation accuracy = 0.919 (0.986)\n",
      "  Subject 4: median (max) validation accuracy = 0.932 (0.987)\n",
      "  Subject 5: median (max) validation accuracy = 0.951 (0.989)\n",
      "  Subject 6: median (max) validation accuracy = 0.926 (0.988)\n",
      "  Subject 7: median (max) validation accuracy = 0.916 (0.975)\n",
      "  Subject 8: median (max) validation accuracy = 0.933 (0.992)\n",
      "** Saving params with joint score = 0.923 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 11 of 50 took       40.973s\n",
      "  training loss:               0.190588\n",
      "  holdout loss (batch):        0.158604\n",
      "  Subject 1: median (max) validation accuracy = 0.887 (0.987)\n",
      "  Subject 2: median (max) validation accuracy = 0.897 (0.986)\n",
      "  Subject 3: median (max) validation accuracy = 0.928 (0.987)\n",
      "  Subject 4: median (max) validation accuracy = 0.940 (0.988)\n",
      "  Subject 5: median (max) validation accuracy = 0.954 (0.989)\n",
      "  Subject 6: median (max) validation accuracy = 0.933 (0.989)\n",
      "  Subject 7: median (max) validation accuracy = 0.925 (0.978)\n",
      "  Subject 8: median (max) validation accuracy = 0.939 (0.992)\n",
      "** Saving params with joint score = 0.930 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 12 of 50 took       40.892s\n",
      "  training loss:               0.175516\n",
      "  holdout loss (batch):        0.144191\n",
      "  Subject 1: median (max) validation accuracy = 0.901 (0.987)\n",
      "  Subject 2: median (max) validation accuracy = 0.910 (0.986)\n",
      "  Subject 3: median (max) validation accuracy = 0.937 (0.988)\n",
      "  Subject 4: median (max) validation accuracy = 0.946 (0.988)\n",
      "  Subject 5: median (max) validation accuracy = 0.957 (0.990)\n",
      "  Subject 6: median (max) validation accuracy = 0.939 (0.990)\n",
      "  Subject 7: median (max) validation accuracy = 0.933 (0.980)\n",
      "  Subject 8: median (max) validation accuracy = 0.945 (0.992)\n",
      "** Saving params with joint score = 0.937 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 13 of 50 took       40.852s\n",
      "  training loss:               0.161970\n",
      "  holdout loss (batch):        0.129236\n",
      "  Subject 1: median (max) validation accuracy = 0.914 (0.989)\n",
      "  Subject 2: median (max) validation accuracy = 0.921 (0.987)\n",
      "  Subject 3: median (max) validation accuracy = 0.944 (0.988)\n",
      "  Subject 4: median (max) validation accuracy = 0.952 (0.989)\n",
      "  Subject 5: median (max) validation accuracy = 0.960 (0.990)\n",
      "  Subject 6: median (max) validation accuracy = 0.946 (0.991)\n",
      "  Subject 7: median (max) validation accuracy = 0.940 (0.982)\n",
      "  Subject 8: median (max) validation accuracy = 0.951 (0.993)\n",
      "** Saving params with joint score = 0.944 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 14 of 50 took       40.869s\n",
      "  training loss:               0.150800\n",
      "  holdout loss (batch):        0.120924\n",
      "  Subject 1: median (max) validation accuracy = 0.923 (0.989)\n",
      "  Subject 2: median (max) validation accuracy = 0.929 (0.987)\n",
      "  Subject 3: median (max) validation accuracy = 0.949 (0.989)\n",
      "  Subject 4: median (max) validation accuracy = 0.956 (0.989)\n",
      "  Subject 5: median (max) validation accuracy = 0.961 (0.990)\n",
      "  Subject 6: median (max) validation accuracy = 0.950 (0.992)\n",
      "  Subject 7: median (max) validation accuracy = 0.945 (0.983)\n",
      "  Subject 8: median (max) validation accuracy = 0.954 (0.993)\n",
      "** Saving params with joint score = 0.948 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 15 of 50 took       40.962s\n",
      "  training loss:               0.141181\n",
      "  holdout loss (batch):        0.110792\n",
      "  Subject 1: median (max) validation accuracy = 0.931 (0.990)\n",
      "  Subject 2: median (max) validation accuracy = 0.936 (0.987)\n",
      "  Subject 3: median (max) validation accuracy = 0.955 (0.989)\n",
      "  Subject 4: median (max) validation accuracy = 0.960 (0.990)\n",
      "  Subject 5: median (max) validation accuracy = 0.964 (0.991)\n",
      "  Subject 6: median (max) validation accuracy = 0.954 (0.992)\n",
      "  Subject 7: median (max) validation accuracy = 0.950 (0.985)\n",
      "  Subject 8: median (max) validation accuracy = 0.958 (0.993)\n",
      "** Saving params with joint score = 0.953 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 16 of 50 took       40.913s\n",
      "  training loss:               0.133343\n",
      "  holdout loss (batch):        0.103615\n",
      "  Subject 1: median (max) validation accuracy = 0.937 (0.990)\n",
      "  Subject 2: median (max) validation accuracy = 0.942 (0.987)\n",
      "  Subject 3: median (max) validation accuracy = 0.958 (0.989)\n",
      "  Subject 4: median (max) validation accuracy = 0.962 (0.990)\n",
      "  Subject 5: median (max) validation accuracy = 0.964 (0.991)\n",
      "  Subject 6: median (max) validation accuracy = 0.957 (0.992)\n",
      "  Subject 7: median (max) validation accuracy = 0.954 (0.986)\n",
      "  Subject 8: median (max) validation accuracy = 0.960 (0.993)\n",
      "** Saving params with joint score = 0.956 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 17 of 50 took       41.092s\n",
      "  training loss:               0.127415\n",
      "  holdout loss (batch):        0.097062\n",
      "  Subject 1: median (max) validation accuracy = 0.942 (0.990)\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "  Subject 2: median (max) validation accuracy = 0.947 (0.988)\n",
      "  Subject 3: median (max) validation accuracy = 0.961 (0.990)\n",
      "  Subject 4: median (max) validation accuracy = 0.965 (0.991)\n",
      "  Subject 5: median (max) validation accuracy = 0.966 (0.991)\n",
      "  Subject 6: median (max) validation accuracy = 0.960 (0.993)\n",
      "  Subject 7: median (max) validation accuracy = 0.957 (0.987)\n",
      "  Subject 8: median (max) validation accuracy = 0.963 (0.993)\n",
      "** Saving params with joint score = 0.959 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 18 of 50 took       41.181s\n",
      "  training loss:               0.122278\n",
      "  holdout loss (batch):        0.093720\n",
      "  Subject 1: median (max) validation accuracy = 0.946 (0.991)\n",
      "  Subject 2: median (max) validation accuracy = 0.950 (0.988)\n",
      "  Subject 3: median (max) validation accuracy = 0.963 (0.991)\n",
      "  Subject 4: median (max) validation accuracy = 0.967 (0.992)\n",
      "  Subject 5: median (max) validation accuracy = 0.967 (0.991)\n",
      "  Subject 6: median (max) validation accuracy = 0.962 (0.993)\n",
      "  Subject 7: median (max) validation accuracy = 0.959 (0.988)\n",
      "  Subject 8: median (max) validation accuracy = 0.965 (0.993)\n",
      "** Saving params with joint score = 0.961 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 19 of 50 took       41.152s\n",
      "  training loss:               0.118453\n",
      "  holdout loss (batch):        0.090239\n",
      "  Subject 1: median (max) validation accuracy = 0.950 (0.990)\n",
      "  Subject 2: median (max) validation accuracy = 0.953 (0.988)\n",
      "  Subject 3: median (max) validation accuracy = 0.965 (0.991)\n",
      "  Subject 4: median (max) validation accuracy = 0.968 (0.992)\n",
      "  Subject 5: median (max) validation accuracy = 0.968 (0.991)\n",
      "  Subject 6: median (max) validation accuracy = 0.964 (0.993)\n",
      "  Subject 7: median (max) validation accuracy = 0.961 (0.988)\n",
      "  Subject 8: median (max) validation accuracy = 0.966 (0.993)\n",
      "** Saving params with joint score = 0.963 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 20 of 50 took       41.141s\n",
      "  training loss:               0.114803\n",
      "  holdout loss (batch):        0.086693\n",
      "  Subject 1: median (max) validation accuracy = 0.953 (0.991)\n",
      "  Subject 2: median (max) validation accuracy = 0.956 (0.989)\n",
      "  Subject 3: median (max) validation accuracy = 0.967 (0.991)\n",
      "  Subject 4: median (max) validation accuracy = 0.969 (0.992)\n",
      "  Subject 5: median (max) validation accuracy = 0.968 (0.991)\n",
      "  Subject 6: median (max) validation accuracy = 0.966 (0.994)\n",
      "  Subject 7: median (max) validation accuracy = 0.963 (0.989)\n",
      "  Subject 8: median (max) validation accuracy = 0.968 (0.993)\n",
      "** Saving params with joint score = 0.964 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 21 of 50 took       41.194s\n",
      "  training loss:               0.112165\n",
      "  holdout loss (batch):        0.085336\n",
      "  Subject 1: median (max) validation accuracy = 0.954 (0.991)\n",
      "  Subject 2: median (max) validation accuracy = 0.957 (0.990)\n",
      "  Subject 3: median (max) validation accuracy = 0.968 (0.991)\n",
      "  Subject 4: median (max) validation accuracy = 0.970 (0.992)\n",
      "  Subject 5: median (max) validation accuracy = 0.968 (0.992)\n",
      "  Subject 6: median (max) validation accuracy = 0.966 (0.993)\n",
      "  Subject 7: median (max) validation accuracy = 0.965 (0.988)\n",
      "  Subject 8: median (max) validation accuracy = 0.969 (0.993)\n",
      "** Saving params with joint score = 0.965 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 22 of 50 took       41.255s\n",
      "  training loss:               0.109595\n",
      "  holdout loss (batch):        0.081064\n",
      "  Subject 1: median (max) validation accuracy = 0.957 (0.991)\n",
      "  Subject 2: median (max) validation accuracy = 0.960 (0.990)\n",
      "  Subject 3: median (max) validation accuracy = 0.969 (0.991)\n",
      "  Subject 4: median (max) validation accuracy = 0.971 (0.993)\n",
      "  Subject 5: median (max) validation accuracy = 0.970 (0.992)\n",
      "  Subject 6: median (max) validation accuracy = 0.968 (0.994)\n",
      "  Subject 7: median (max) validation accuracy = 0.966 (0.990)\n",
      "  Subject 8: median (max) validation accuracy = 0.970 (0.994)\n",
      "** Saving params with joint score = 0.967 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 23 of 50 took       41.264s\n",
      "  training loss:               0.107674\n",
      "  holdout loss (batch):        0.081830\n",
      "  Subject 1: median (max) validation accuracy = 0.958 (0.991)\n",
      "  Subject 2: median (max) validation accuracy = 0.961 (0.990)\n",
      "  Subject 3: median (max) validation accuracy = 0.970 (0.992)\n",
      "  Subject 4: median (max) validation accuracy = 0.972 (0.993)\n",
      "  Subject 5: median (max) validation accuracy = 0.970 (0.992)\n",
      "  Subject 6: median (max) validation accuracy = 0.969 (0.994)\n",
      "  Subject 7: median (max) validation accuracy = 0.966 (0.989)\n",
      "  Subject 8: median (max) validation accuracy = 0.971 (0.994)\n",
      "** Saving params with joint score = 0.967 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 24 of 50 took       41.215s\n",
      "  training loss:               0.105727\n",
      "  holdout loss (batch):        0.075934\n",
      "  Subject 1: median (max) validation accuracy = 0.960 (0.992)\n",
      "  Subject 2: median (max) validation accuracy = 0.962 (0.991)\n",
      "  Subject 3: median (max) validation accuracy = 0.971 (0.992)\n",
      "  Subject 4: median (max) validation accuracy = 0.973 (0.993)\n",
      "  Subject 5: median (max) validation accuracy = 0.971 (0.992)\n",
      "  Subject 6: median (max) validation accuracy = 0.970 (0.994)\n",
      "  Subject 7: median (max) validation accuracy = 0.967 (0.990)\n",
      "  Subject 8: median (max) validation accuracy = 0.971 (0.994)\n",
      "** Saving params with joint score = 0.968 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 25 of 50 took       40.942s\n",
      "  training loss:               0.104227\n",
      "  holdout loss (batch):        0.074730\n",
      "  Subject 1: median (max) validation accuracy = 0.961 (0.992)\n",
      "  Subject 2: median (max) validation accuracy = 0.963 (0.991)\n",
      "  Subject 3: median (max) validation accuracy = 0.972 (0.992)\n",
      "  Subject 4: median (max) validation accuracy = 0.973 (0.993)\n",
      "  Subject 5: median (max) validation accuracy = 0.971 (0.992)\n",
      "  Subject 6: median (max) validation accuracy = 0.970 (0.994)\n",
      "  Subject 7: median (max) validation accuracy = 0.968 (0.990)\n",
      "  Subject 8: median (max) validation accuracy = 0.971 (0.994)\n",
      "** Saving params with joint score = 0.969 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 26 of 50 took       40.823s\n",
      "  training loss:               0.102555\n",
      "  holdout loss (batch):        0.075262\n",
      "  Subject 1: median (max) validation accuracy = 0.962 (0.992)\n",
      "  Subject 2: median (max) validation accuracy = 0.964 (0.991)\n",
      "  Subject 3: median (max) validation accuracy = 0.972 (0.992)\n",
      "  Subject 4: median (max) validation accuracy = 0.973 (0.993)\n",
      "  Subject 5: median (max) validation accuracy = 0.971 (0.993)\n",
      "  Subject 6: median (max) validation accuracy = 0.971 (0.994)\n",
      "  Subject 7: median (max) validation accuracy = 0.969 (0.990)\n",
      "  Subject 8: median (max) validation accuracy = 0.972 (0.994)\n",
      "** Saving params with joint score = 0.969 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 27 of 50 took       40.811s\n",
      "  training loss:               0.101513\n",
      "  holdout loss (batch):        0.072910\n",
      "  Subject 1: median (max) validation accuracy = 0.963 (0.992)\n",
      "  Subject 2: median (max) validation accuracy = 0.965 (0.991)\n",
      "  Subject 3: median (max) validation accuracy = 0.972 (0.992)\n",
      "  Subject 4: median (max) validation accuracy = 0.974 (0.994)\n",
      "  Subject 5: median (max) validation accuracy = 0.971 (0.993)\n",
      "  Subject 6: median (max) validation accuracy = 0.971 (0.994)\n",
      "  Subject 7: median (max) validation accuracy = 0.969 (0.990)\n",
      "  Subject 8: median (max) validation accuracy = 0.972 (0.994)\n",
      "** Saving params with joint score = 0.970 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 28 of 50 took       40.829s\n",
      "  training loss:               0.100381\n",
      "  holdout loss (batch):        0.071642\n",
      "  Subject 1: median (max) validation accuracy = 0.964 (0.992)\n",
      "  Subject 2: median (max) validation accuracy = 0.965 (0.991)\n",
      "  Subject 3: median (max) validation accuracy = 0.973 (0.992)\n",
      "  Subject 4: median (max) validation accuracy = 0.974 (0.994)\n",
      "  Subject 5: median (max) validation accuracy = 0.971 (0.993)\n",
      "  Subject 6: median (max) validation accuracy = 0.972 (0.994)\n",
      "  Subject 7: median (max) validation accuracy = 0.970 (0.990)\n",
      "  Subject 8: median (max) validation accuracy = 0.973 (0.994)\n",
      "** Saving params with joint score = 0.970 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 29 of 50 took       40.817s\n",
      "  training loss:               0.099609\n",
      "  holdout loss (batch):        0.071100\n",
      "  Subject 1: median (max) validation accuracy = 0.964 (0.992)\n",
      "  Subject 2: median (max) validation accuracy = 0.966 (0.991)\n",
      "  Subject 3: median (max) validation accuracy = 0.973 (0.992)\n",
      "  Subject 4: median (max) validation accuracy = 0.975 (0.994)\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "  Subject 5: median (max) validation accuracy = 0.971 (0.993)\n",
      "  Subject 6: median (max) validation accuracy = 0.972 (0.994)\n",
      "  Subject 7: median (max) validation accuracy = 0.970 (0.991)\n",
      "  Subject 8: median (max) validation accuracy = 0.973 (0.994)\n",
      "** Saving params with joint score = 0.971 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 30 of 50 took       40.814s\n",
      "  training loss:               0.098691\n",
      "  holdout loss (batch):        0.070693\n",
      "  Subject 1: median (max) validation accuracy = 0.965 (0.992)\n",
      "  Subject 2: median (max) validation accuracy = 0.967 (0.991)\n",
      "  Subject 3: median (max) validation accuracy = 0.974 (0.992)\n",
      "  Subject 4: median (max) validation accuracy = 0.975 (0.994)\n",
      "  Subject 5: median (max) validation accuracy = 0.971 (0.993)\n",
      "  Subject 6: median (max) validation accuracy = 0.972 (0.994)\n",
      "  Subject 7: median (max) validation accuracy = 0.970 (0.990)\n",
      "  Subject 8: median (max) validation accuracy = 0.973 (0.994)\n",
      "** Saving params with joint score = 0.971 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 31 of 50 took       40.850s\n",
      "  training loss:               0.098008\n",
      "  holdout loss (batch):        0.071708\n",
      "  Subject 1: median (max) validation accuracy = 0.966 (0.992)\n",
      "  Subject 2: median (max) validation accuracy = 0.967 (0.991)\n",
      "  Subject 3: median (max) validation accuracy = 0.974 (0.992)\n",
      "  Subject 4: median (max) validation accuracy = 0.975 (0.994)\n",
      "  Subject 5: median (max) validation accuracy = 0.971 (0.993)\n",
      "  Subject 6: median (max) validation accuracy = 0.972 (0.994)\n",
      "  Subject 7: median (max) validation accuracy = 0.971 (0.990)\n",
      "  Subject 8: median (max) validation accuracy = 0.973 (0.994)\n",
      "** Saving params with joint score = 0.971 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 32 of 50 took       40.825s\n",
      "  training loss:               0.097389\n",
      "  holdout loss (batch):        0.070647\n",
      "  Subject 1: median (max) validation accuracy = 0.966 (0.992)\n",
      "  Subject 2: median (max) validation accuracy = 0.967 (0.991)\n",
      "  Subject 3: median (max) validation accuracy = 0.974 (0.992)\n",
      "  Subject 4: median (max) validation accuracy = 0.975 (0.994)\n",
      "  Subject 5: median (max) validation accuracy = 0.972 (0.993)\n",
      "  Subject 6: median (max) validation accuracy = 0.973 (0.994)\n",
      "  Subject 7: median (max) validation accuracy = 0.971 (0.990)\n",
      "  Subject 8: median (max) validation accuracy = 0.974 (0.994)\n",
      "** Saving params with joint score = 0.971 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 33 of 50 took       40.813s\n",
      "  training loss:               0.096815\n",
      "  holdout loss (batch):        0.068895\n",
      "  Subject 1: median (max) validation accuracy = 0.966 (0.992)\n",
      "  Subject 2: median (max) validation accuracy = 0.967 (0.991)\n",
      "  Subject 3: median (max) validation accuracy = 0.974 (0.992)\n",
      "  Subject 4: median (max) validation accuracy = 0.975 (0.993)\n",
      "  Subject 5: median (max) validation accuracy = 0.972 (0.993)\n",
      "  Subject 6: median (max) validation accuracy = 0.973 (0.994)\n",
      "  Subject 7: median (max) validation accuracy = 0.971 (0.990)\n",
      "  Subject 8: median (max) validation accuracy = 0.974 (0.994)\n",
      "** Saving params with joint score = 0.971 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 34 of 50 took       41.003s\n",
      "  training loss:               0.096266\n",
      "  holdout loss (batch):        0.070175\n",
      "  Subject 1: median (max) validation accuracy = 0.967 (0.992)\n",
      "  Subject 2: median (max) validation accuracy = 0.968 (0.992)\n",
      "  Subject 3: median (max) validation accuracy = 0.974 (0.992)\n",
      "  Subject 4: median (max) validation accuracy = 0.975 (0.993)\n",
      "  Subject 5: median (max) validation accuracy = 0.972 (0.993)\n",
      "  Subject 6: median (max) validation accuracy = 0.973 (0.994)\n",
      "  Subject 7: median (max) validation accuracy = 0.971 (0.991)\n",
      "  Subject 8: median (max) validation accuracy = 0.974 (0.993)\n",
      "** Saving params with joint score = 0.972 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 35 of 50 took       41.003s\n",
      "  training loss:               0.095691\n",
      "  holdout loss (batch):        0.069154\n",
      "  Subject 1: median (max) validation accuracy = 0.967 (0.992)\n",
      "  Subject 2: median (max) validation accuracy = 0.968 (0.992)\n",
      "  Subject 3: median (max) validation accuracy = 0.974 (0.992)\n",
      "  Subject 4: median (max) validation accuracy = 0.976 (0.993)\n",
      "  Subject 5: median (max) validation accuracy = 0.972 (0.993)\n",
      "  Subject 6: median (max) validation accuracy = 0.973 (0.994)\n",
      "  Subject 7: median (max) validation accuracy = 0.971 (0.991)\n",
      "  Subject 8: median (max) validation accuracy = 0.974 (0.994)\n",
      "** Saving params with joint score = 0.972 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 36 of 50 took       40.992s\n",
      "  training loss:               0.095553\n",
      "  holdout loss (batch):        0.067534\n",
      "  Subject 1: median (max) validation accuracy = 0.968 (0.993)\n",
      "  Subject 2: median (max) validation accuracy = 0.969 (0.991)\n",
      "  Subject 3: median (max) validation accuracy = 0.975 (0.992)\n",
      "  Subject 4: median (max) validation accuracy = 0.976 (0.994)\n",
      "  Subject 5: median (max) validation accuracy = 0.972 (0.993)\n",
      "  Subject 6: median (max) validation accuracy = 0.973 (0.994)\n",
      "  Subject 7: median (max) validation accuracy = 0.972 (0.991)\n",
      "  Subject 8: median (max) validation accuracy = 0.974 (0.994)\n",
      "** Saving params with joint score = 0.972 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 37 of 50 took       41.023s\n",
      "  training loss:               0.094782\n",
      "  holdout loss (batch):        0.068597\n",
      "  Subject 1: median (max) validation accuracy = 0.968 (0.993)\n",
      "  Subject 2: median (max) validation accuracy = 0.969 (0.992)\n",
      "  Subject 3: median (max) validation accuracy = 0.975 (0.992)\n",
      "  Subject 4: median (max) validation accuracy = 0.976 (0.994)\n",
      "  Subject 5: median (max) validation accuracy = 0.972 (0.993)\n",
      "  Subject 6: median (max) validation accuracy = 0.973 (0.994)\n",
      "  Subject 7: median (max) validation accuracy = 0.972 (0.991)\n",
      "  Subject 8: median (max) validation accuracy = 0.974 (0.993)\n",
      "** Saving params with joint score = 0.972 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 38 of 50 took       41.006s\n",
      "  training loss:               0.094426\n",
      "  holdout loss (batch):        0.066832\n",
      "  Subject 1: median (max) validation accuracy = 0.969 (0.993)\n",
      "  Subject 2: median (max) validation accuracy = 0.969 (0.992)\n",
      "  Subject 3: median (max) validation accuracy = 0.975 (0.992)\n",
      "  Subject 4: median (max) validation accuracy = 0.976 (0.994)\n",
      "  Subject 5: median (max) validation accuracy = 0.972 (0.993)\n",
      "  Subject 6: median (max) validation accuracy = 0.974 (0.994)\n",
      "  Subject 7: median (max) validation accuracy = 0.972 (0.991)\n",
      "  Subject 8: median (max) validation accuracy = 0.975 (0.994)\n",
      "** Saving params with joint score = 0.973 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 39 of 50 took       40.945s\n",
      "  training loss:               0.094162\n",
      "  holdout loss (batch):        0.067699\n",
      "  Subject 1: median (max) validation accuracy = 0.968 (0.993)\n",
      "  Subject 2: median (max) validation accuracy = 0.969 (0.992)\n",
      "  Subject 3: median (max) validation accuracy = 0.975 (0.992)\n",
      "  Subject 4: median (max) validation accuracy = 0.976 (0.994)\n",
      "  Subject 5: median (max) validation accuracy = 0.972 (0.993)\n",
      "  Subject 6: median (max) validation accuracy = 0.974 (0.995)\n",
      "  Subject 7: median (max) validation accuracy = 0.972 (0.991)\n",
      "  Subject 8: median (max) validation accuracy = 0.975 (0.994)\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 40 of 50 took       41.069s\n",
      "  training loss:               0.094093\n",
      "  holdout loss (batch):        0.065449\n",
      "  Subject 1: median (max) validation accuracy = 0.969 (0.993)\n",
      "  Subject 2: median (max) validation accuracy = 0.970 (0.992)\n",
      "  Subject 3: median (max) validation accuracy = 0.975 (0.992)\n",
      "  Subject 4: median (max) validation accuracy = 0.976 (0.994)\n",
      "  Subject 5: median (max) validation accuracy = 0.972 (0.993)\n",
      "  Subject 6: median (max) validation accuracy = 0.974 (0.994)\n",
      "  Subject 7: median (max) validation accuracy = 0.972 (0.991)\n",
      "  Subject 8: median (max) validation accuracy = 0.975 (0.994)\n",
      "** Saving params with joint score = 0.973 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 41 of 50 took       41.207s\n",
      "  training loss:               0.093385\n",
      "  holdout loss (batch):        0.068873\n",
      "  Subject 1: median (max) validation accuracy = 0.969 (0.993)\n",
      "  Subject 2: median (max) validation accuracy = 0.970 (0.992)\n",
      "  Subject 3: median (max) validation accuracy = 0.975 (0.992)\n",
      "  Subject 4: median (max) validation accuracy = 0.976 (0.994)\n",
      "  Subject 5: median (max) validation accuracy = 0.972 (0.993)\n",
      "  Subject 6: median (max) validation accuracy = 0.974 (0.994)\n",
      "  Subject 7: median (max) validation accuracy = 0.972 (0.991)\n",
      "  Subject 8: median (max) validation accuracy = 0.975 (0.994)\n",
      "** Saving params with joint score = 0.973 **\n",
      "\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\\ : 100.0 %\n",
      "  Epoch 42 of 50 took       41.146s\n",
      "  training loss:               0.093381\n",
      "  holdout loss (batch):        0.066977\n",
      "  Subject 1: median (max) validation accuracy = 0.969 (0.993)\n",
      "  Subject 2: median (max) validation accuracy = 0.970 (0.992)\n",
      "  Subject 3: median (max) validation accuracy = 0.975 (0.992)\n",
      "  Subject 4: median (max) validation accuracy = 0.976 (0.994)\n",
      "  Subject 5: median (max) validation accuracy = 0.973 (0.993)\n",
      "  Subject 6: median (max) validation accuracy = 0.974 (0.994)\n",
      "  Subject 7: median (max) validation accuracy = 0.972 (0.991)\n",
      "  Subject 8: median (max) validation accuracy = 0.975 (0.994)\n",
      "** Saving params with joint score = 0.973 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 43 of 50 took       41.104s\n",
      "  training loss:               0.093162\n",
      "  holdout loss (batch):        0.066273\n",
      "  Subject 1: median (max) validation accuracy = 0.969 (0.993)\n",
      "  Subject 2: median (max) validation accuracy = 0.970 (0.992)\n",
      "  Subject 3: median (max) validation accuracy = 0.975 (0.992)\n",
      "  Subject 4: median (max) validation accuracy = 0.976 (0.994)\n",
      "  Subject 5: median (max) validation accuracy = 0.972 (0.993)\n",
      "  Subject 6: median (max) validation accuracy = 0.974 (0.994)\n",
      "  Subject 7: median (max) validation accuracy = 0.972 (0.991)\n",
      "  Subject 8: median (max) validation accuracy = 0.975 (0.994)\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 44 of 50 took       40.919s\n",
      "  training loss:               0.093093\n",
      "  holdout loss (batch):        0.064173\n",
      "  Subject 1: median (max) validation accuracy = 0.969 (0.992)\n",
      "  Subject 2: median (max) validation accuracy = 0.970 (0.992)\n",
      "  Subject 3: median (max) validation accuracy = 0.975 (0.992)\n",
      "  Subject 4: median (max) validation accuracy = 0.976 (0.994)\n",
      "  Subject 5: median (max) validation accuracy = 0.973 (0.993)\n",
      "  Subject 6: median (max) validation accuracy = 0.974 (0.994)\n",
      "  Subject 7: median (max) validation accuracy = 0.972 (0.991)\n",
      "  Subject 8: median (max) validation accuracy = 0.975 (0.994)\n",
      "** Saving params with joint score = 0.973 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 45 of 50 took       40.846s\n",
      "  training loss:               0.092798\n",
      "  holdout loss (batch):        0.066910\n",
      "  Subject 1: median (max) validation accuracy = 0.970 (0.993)\n",
      "  Subject 2: median (max) validation accuracy = 0.970 (0.992)\n",
      "  Subject 3: median (max) validation accuracy = 0.976 (0.992)\n",
      "  Subject 4: median (max) validation accuracy = 0.976 (0.994)\n",
      "  Subject 5: median (max) validation accuracy = 0.973 (0.993)\n",
      "  Subject 6: median (max) validation accuracy = 0.974 (0.994)\n",
      "  Subject 7: median (max) validation accuracy = 0.972 (0.991)\n",
      "  Subject 8: median (max) validation accuracy = 0.975 (0.994)\n",
      "** Saving params with joint score = 0.973 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 46 of 50 took       41.003s\n",
      "  training loss:               0.092652\n",
      "  holdout loss (batch):        0.067232\n",
      "  Subject 1: median (max) validation accuracy = 0.970 (0.993)\n",
      "  Subject 2: median (max) validation accuracy = 0.970 (0.992)\n",
      "  Subject 3: median (max) validation accuracy = 0.975 (0.992)\n",
      "  Subject 4: median (max) validation accuracy = 0.976 (0.994)\n",
      "  Subject 5: median (max) validation accuracy = 0.972 (0.993)\n",
      "  Subject 6: median (max) validation accuracy = 0.974 (0.994)\n",
      "  Subject 7: median (max) validation accuracy = 0.972 (0.991)\n",
      "  Subject 8: median (max) validation accuracy = 0.975 (0.994)\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 47 of 50 took       41.131s\n",
      "  training loss:               0.092233\n",
      "  holdout loss (batch):        0.065232\n",
      "  Subject 1: median (max) validation accuracy = 0.970 (0.992)\n",
      "  Subject 2: median (max) validation accuracy = 0.970 (0.992)\n",
      "  Subject 3: median (max) validation accuracy = 0.976 (0.992)\n",
      "  Subject 4: median (max) validation accuracy = 0.976 (0.994)\n",
      "  Subject 5: median (max) validation accuracy = 0.972 (0.993)\n",
      "  Subject 6: median (max) validation accuracy = 0.974 (0.994)\n",
      "  Subject 7: median (max) validation accuracy = 0.972 (0.991)\n",
      "  Subject 8: median (max) validation accuracy = 0.975 (0.994)\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 48 of 50 took       41.050s\n",
      "  training loss:               0.092236\n",
      "  holdout loss (batch):        0.064699\n",
      "  Subject 1: median (max) validation accuracy = 0.970 (0.993)\n",
      "  Subject 2: median (max) validation accuracy = 0.971 (0.992)\n",
      "  Subject 3: median (max) validation accuracy = 0.976 (0.992)\n",
      "  Subject 4: median (max) validation accuracy = 0.977 (0.994)\n",
      "  Subject 5: median (max) validation accuracy = 0.973 (0.993)\n",
      "  Subject 6: median (max) validation accuracy = 0.974 (0.995)\n",
      "  Subject 7: median (max) validation accuracy = 0.973 (0.991)\n",
      "  Subject 8: median (max) validation accuracy = 0.975 (0.994)\n",
      "** Saving params with joint score = 0.974 **\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 49 of 50 took       41.080s\n",
      "  training loss:               0.092034\n",
      "  holdout loss (batch):        0.066460\n",
      "  Subject 1: median (max) validation accuracy = 0.970 (0.993)\n",
      "  Subject 2: median (max) validation accuracy = 0.971 (0.992)\n",
      "  Subject 3: median (max) validation accuracy = 0.976 (0.993)\n",
      "  Subject 4: median (max) validation accuracy = 0.977 (0.994)\n",
      "  Subject 5: median (max) validation accuracy = 0.973 (0.993)\n",
      "  Subject 6: median (max) validation accuracy = 0.974 (0.994)\n",
      "  Subject 7: median (max) validation accuracy = 0.972 (0.991)\n",
      "  Subject 8: median (max) validation accuracy = 0.975 (0.994)\n",
      "\n",
      "\\ : 100.0 %\n",
      "  Epoch 50 of 50 took       41.037s\n",
      "  training loss:               0.092044\n",
      "  holdout loss (batch):        0.063582\n",
      "  Subject 1: median (max) validation accuracy = 0.970 (0.993)\n",
      "  Subject 2: median (max) validation accuracy = 0.971 (0.992)\n",
      "  Subject 3: median (max) validation accuracy = 0.976 (0.992)\n",
      "  Subject 4: median (max) validation accuracy = 0.977 (0.994)\n",
      "  Subject 5: median (max) validation accuracy = 0.973 (0.993)\n",
      "  Subject 6: median (max) validation accuracy = 0.975 (0.995)\n",
      "  Subject 7: median (max) validation accuracy = 0.973 (0.991)\n",
      "  Subject 8: median (max) validation accuracy = 0.976 (0.994)\n",
      "** Saving params with joint score = 0.974 **\n",
      "\n",
      "***************************************\n",
      "best joint val cc = 0.976\n",
      "subject 1: val cc = 0.973\n",
      "subject 2: val cc = 0.973\n",
      "subject 3: val cc = 0.977\n",
      "subject 4: val cc = 0.978\n",
      "subject 5: val cc = 0.977\n",
      "subject 6: val cc = 0.977\n",
      "subject 7: val cc = 0.975\n",
      "subject 8: val cc = 0.977\n",
      "***************************************\n"
     ]
    }
   ],
   "source": [
    "batch_size = 100\n",
    "num_epochs = 50\n",
    "holdout_frac = .1\n",
    "\n",
    "for roi_name, roi_file_dir in roi_timestamp_dir.items(): \n",
    "\n",
    "    # create subdir\n",
    "    model_dir     = roi_file_dir\n",
    "    output_subdir = output_dir + roi_name + '/'\n",
    "    print (output_subdir)\n",
    "    if not os.path.exists(output_subdir):\n",
    "        os.makedirs(output_subdir)\n",
    "\n",
    "    print ('***************************************')\n",
    "    print ('***        %12s       ***'%roi_name)\n",
    "    print ('from ==> %s'%model_dir)\n",
    "    print ('to   ==> %s'%output_subdir)\n",
    "    print ('***************************************') \n",
    "    \n",
    "    \n",
    "    checkpoint = torch.load(model_dir)\n",
    "    shared_params = checkpoint['best_params']\n",
    "    ###\n",
    "    \n",
    "    shared_model = Encoder(trn_stim_mean, trunk_width=N).to(device)\n",
    "    shared_model.load_state_dict(shared_params['enc'])\n",
    "    shared_model.eval()     \n",
    "\n",
    "    rec, fmaps, h = shared_model(T.from_numpy(stim_data[list(stim_data.keys())[0]][:20]).to(device))\n",
    "    \n",
    "    subject_fwrfs = {s: Torch_LayerwiseFWRF(fmaps, nv=trn_voxel_data[s].shape[1], pre_nl=_log_act_fn, \\\n",
    "                     post_nl=_log_act_fn, dtype=np.float32).to(device) for s in subjects}    \n",
    "    \n",
    "    subject_optimizer = {s: optim.Adam([\n",
    "            {'params': sp.parameters()}\n",
    "        ], lr=1e-3, betas=(0.9, 0.999), eps=1e-08) for s,sp in subject_fwrfs.items()}\n",
    "    subject_opts = {s: [subject_optimizer[s]] for s in subject_optimizer.keys()}   \n",
    "    ###################################\n",
    "    \n",
    "\n",
    "    best_params, final_params, hold_cc_hist, hold_hist, trn_hist, best_epoch, best_joint_cc_score = \\\n",
    "        learn_params_(_training_fn, _holdout_fn, _pred_fn, shared_model, subject_fwrfs, subject_opts, \n",
    "            stim_data, trn_voxel_data, trn_stim_ordering,\n",
    "            num_epochs=num_epochs, batch_size=batch_size, holdout_frac=holdout_frac, randomize=False)\n",
    "    \n",
    "    \n",
    "    ###################################\n",
    "    #val_voxel = {s: val_voxel_data[s] for s in val_voxel_data.keys()}\n",
    "    shared_model.load_state_dict(best_params['enc'])\n",
    "    shared_model.eval() \n",
    "    for s,sd in subject_fwrfs.items():\n",
    "        sd.load_state_dict(best_params['fwrfs'][s])\n",
    "        sd.eval() \n",
    "\n",
    "    subject_val_cc = validation_(_pred_fn, shared_model, subject_fwrfs, stim_data, val_voxel_data, val_stim_ordering, batch_size)\n",
    "    joined_val_cc = np.concatenate(list(subject_val_cc.values()), axis=0)\n",
    "    \n",
    "    print ('***************************************')\n",
    "    print (\"best joint val cc = %.3f\"% np.median(joined_val_cc))\n",
    "    for s,v in subject_val_cc.items():\n",
    "        print (\"subject %s: val cc = %.3f\"%(s, np.median(v)))\n",
    " \n",
    "    print ('***************************************')\n",
    "    ###################################\n",
    "    torch.save({\n",
    "            'group_name': roi_name,\n",
    "            'num_epochs': num_epochs,\n",
    "            'batch_size': batch_size,\n",
    "            'holdout_frac': holdout_frac,   \n",
    "            'best_params': best_params, \n",
    "            'final_params': final_params,\n",
    "            'trn_loss_history': trn_hist,\n",
    "            'hold_loss_history': hold_hist,\n",
    "            'hold_cc_history': hold_cc_hist,\n",
    "            'best_epoch': best_epoch,\n",
    "            'best_joint_cc_score': best_joint_cc_score,\n",
    "            'val_cc': subject_val_cc,\n",
    "            'input_mean': trn_stim_mean,\n",
    "            'voxel_roi': voxel_roi\n",
    "            }, output_subdir+'model_params_%s'%noise_model)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "\n",
    "# Cross model table"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 51,
   "metadata": {},
   "outputs": [],
   "source": [
    "group_names = ['V1', 'V2', 'V3', 'hV4']\n",
    "group = [[1,2], [3,4], [5,6], [7]]\n",
    "\n",
    "crossroi_val_ccs = {}\n",
    "for roi_name in ['V1', 'V2', 'V3', 'hV4']:\n",
    "    checkpoint = torch.load('%s%s/model_params_%s'%(output_dir, roi_name, noise_model) )\n",
    "    crossroi_val_ccs[roi_name] = checkpoint['val_cc']"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 52,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0 - V1\n",
      "1 - V2\n",
      "2 - V3\n",
      "3 - hV4\n"
     ]
    }
   ],
   "source": [
    "subj_val_mat = {s: np.zeros(shape=(len(group_names), len(group_names))) for s in voxel_roi.keys()}\n",
    "all_mat = np.zeros(shape=(len(group_names), len(group_names)))\n",
    "all_count = np.zeros(shape=(len(group_names), len(group_names)))\n",
    "for i,(i_roi_name, val_cc) in enumerate(crossroi_val_ccs.items()):\n",
    "    print ('%d - %s'%(i, i_roi_name))\n",
    "    for s, v in val_cc.items():\n",
    "        for j, (roi_mask, j_roi_name) in enumerate(iterate_roi(group, voxel_roi[s], roi_map)):\n",
    "            #print ('  %d - %s'%(j, j_roi_name))\n",
    "            subj_val_mat[s][i,j] = np.mean(v[roi_mask] -  gnetfr_val_cc[s][roi_mask])\n",
    "            all_mat[i,j] += np.sum(v[roi_mask] -  gnetfr_val_cc[s][roi_mask])\n",
    "            all_count[i,j] += np.sum(roi_mask)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 54,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 864x432 with 8 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "# line = model\n",
    "# column = roi\n",
    "\n",
    "fig=plt.figure(figsize=(12,6))\n",
    "plt.subplots_adjust(left=0.15, bottom=0.1, right=.95, top=.95, wspace=.5, hspace=.5)\n",
    "\n",
    "for k,(s,mat) in enumerate(subj_val_mat.items()):\n",
    "    plt.subplot(2,4,k+1)\n",
    "    plt.imshow(mat, cmap='seismic', vmin=-.05, vmax=.05, origin='lower')\n",
    "    plt.grid(False)\n",
    "    ax = plt.gca()\n",
    "    ax.set_xticks(np.arange(len(group_names)))\n",
    "    ax.set_xticklabels(group_names)\n",
    "    ax.set_yticks(np.arange(len(group_names)))\n",
    "    ax.set_yticklabels(group_names)\n",
    "    if k%4==0:\n",
    "        plt.ylabel('model')\n",
    "    if k>=4:\n",
    "        plt.xlabel('ROI')\n",
    "    plt.title('S%s'%s)\n",
    "    #if (k+1)%4==0:\n",
    "    #    plt.colorbar()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 55,
   "metadata": {},
   "outputs": [],
   "source": [
    "filename = output_dir + \"remodel_table_fake_%s%s\" % (noise_model, saveext)\n",
    "fig.patch.set_alpha(0.)\n",
    "fig.savefig(filename, **savearg)\n",
    "plt.close()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 56,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "Text(0.5, 0, 'ROI')"
      ]
     },
     "execution_count": 56,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x432 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "fig=plt.figure(figsize=(6,6))\n",
    "plt.subplots_adjust(left=0.15, bottom=0.1, right=.95, top=.95, wspace=.5, hspace=.5)\n",
    "\n",
    "plt.imshow(all_mat/all_count, cmap='seismic', vmin=-.05, vmax=.05, origin='lower')\n",
    "plt.grid(False)\n",
    "ax = plt.gca()\n",
    "ax.set_xticks(np.arange(len(group_names)))\n",
    "ax.set_xticklabels(group_names)\n",
    "ax.set_yticks(np.arange(len(group_names)))\n",
    "ax.set_yticklabels(group_names)\n",
    "plt.ylabel('model')\n",
    "plt.xlabel('ROI')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 57,
   "metadata": {},
   "outputs": [],
   "source": [
    "filename = output_dir + 'gnet8r_xmodel_table_fake_all_%s%s'  % (noise_model, saveext)\n",
    "fig.patch.set_alpha(0.)\n",
    "fig.savefig(filename, **savearg)\n",
    "plt.close()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 58,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "1 : 0.15940161050338802 +/- 0.06399297448596034\n",
      "2 : 0.18718579351377174 +/- 0.05880494487246381\n",
      "3 : 0.3588682399970319 +/- 0.08477148951696847\n",
      "4 : 0.4005326727838038 +/- 0.09217241803150215\n",
      "5 : 0.4641417134759827 +/- 0.11357713771548505\n",
      "6 : 0.3232343438072127 +/- 0.0750464288772547\n",
      "7 : 0.3083494005914741 +/- 0.07912644502187012\n",
      "8 : 0.1477069582487938 +/- 0.046071053660319716\n"
     ]
    }
   ],
   "source": [
    "D_mat = np.array([[0,1,2,3], [1,0,1,2], [2,1,0,1],  [3,2,1,0]])\n",
    "D_up_tri_mat = np.array([[0,0,0,0], [1,0,0,0], [2,1,0,0], [3,2,1,0]])\n",
    "D_dn_tri_mat = np.array([[0,1,2,3], [0,0,1,2], [0,0,0,1], [0,0,0,0]])\n",
    "\n",
    "from scipy.optimize import curve_fit\n",
    "def fit_fn(x, a):\n",
    "    return a*x\n",
    "\n",
    "alpha, dalpha = {}, {}\n",
    "for s,m in subj_val_mat.items():\n",
    "\n",
    "    popt, pcov = curve_fit(fit_fn, D_up_tri_mat.flatten()[D_up_tri_mat.flatten()>0], m.flatten()[D_up_tri_mat.flatten()>0])\n",
    "    perr = np.sqrt(np.diag(pcov))\n",
    "    a_up, da_up = popt[0], perr[0]\n",
    "\n",
    "    popt, pcov = curve_fit(fit_fn, D_dn_tri_mat.flatten()[D_dn_tri_mat.flatten()>0], m.flatten()[D_dn_tri_mat.flatten()>0])\n",
    "    perr = np.sqrt(np.diag(pcov))\n",
    "    a_dn, da_dn = popt[0], perr[0]\n",
    "    \n",
    "    alpha[s]  = (a_up - a_dn) / ( np.abs(a_up) + np.abs(a_dn) )\n",
    "    dalpha[s] = np.abs(alpha[s]) * np.sqrt( (da_up**2 + da_dn**2) / (a_up - a_dn)**2 + (da_up**2 + da_dn**2) / (np.abs(a_up) + np.abs(a_dn))**2  )\n",
    "    print (s, ':', alpha[s], '+/-', dalpha[s])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 59,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "all : 0.29716742842586547 +/- 0.07427850953640995\n"
     ]
    }
   ],
   "source": [
    "for s,m in {'all': all_mat/all_count}.items():\n",
    "\n",
    "    popt, pcov = curve_fit(fit_fn, D_up_tri_mat.flatten()[D_up_tri_mat.flatten()>0], m.flatten()[D_up_tri_mat.flatten()>0])\n",
    "    perr = np.sqrt(np.diag(pcov))\n",
    "    a_up, da_up = popt[0], perr[0]\n",
    "\n",
    "    popt, pcov = curve_fit(fit_fn, D_dn_tri_mat.flatten()[D_dn_tri_mat.flatten()>0], m.flatten()[D_dn_tri_mat.flatten()>0])\n",
    "    perr = np.sqrt(np.diag(pcov))\n",
    "    a_dn, da_dn = popt[0], perr[0]\n",
    "    \n",
    "    alpha[s]  = (a_up - a_dn) / ( np.abs(a_up) + np.abs(a_dn) )\n",
    "    dalpha[s] = np.abs(alpha[s]) * np.sqrt( (da_up**2 + da_dn**2) / (a_up - a_dn)**2 + (da_up**2 + da_dn**2) / (np.abs(a_up) + np.abs(a_dn))**2  )\n",
    "    print (s, ':', alpha[s], '+/-', dalpha[s])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 60,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "no_noise\n"
     ]
    }
   ],
   "source": [
    "print (noise_model)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.6.8"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 1
}