meta_reg_MAML_simple.ipynb

{"nbformat":4,"nbformat_minor":0,"metadata":{"accelerator":"GPU","colab":{"name":"04meta_reg_MAML-project.ipynb","provenance":[{"file_id":"179hlUyqebNa-Pt4cnNrDVWf4gklUjb5M","timestamp":1606174186604},{"file_id":"1KYKGP8emNO3DBhA88VZpMbXLvIjRcziy","timestamp":1606171723713},{"file_id":"1dDMz_f4vKShrnkwSY8AARHRcqWDUD5TP","timestamp":1604345329315},{"file_id":"1zbt2A74kM10HvcAEgEy3fGgRSNyHZQKj","timestamp":1602827504745},{"file_id":"1esHcdn0jL8an4SMG23_fMjbK9n-MsWXz","timestamp":1601876029717},{"file_id":"1lnvSzcjmmJknvRG05fk1BMis7fyE0i9C","timestamp":1601844968812},{"file_id":"1yz-_veyTWMbt3pE1SFyWH-ZoA_AqiImd","timestamp":1601781455461},{"file_id":"13qcZDBvQzxXshmMspv0SBJm9-85GRO31","timestamp":1601644081170}],"collapsed_sections":[]},"kernelspec":{"display_name":"Python 3","name":"python3"}},"cells":[{"cell_type":"markdown","metadata":{"id":"JvkoC8rAYBE7"},"source":["\n","##Setup\n","\n","You will need to make a copy of this Colab notebook in your Google Drive before you can edit the homework files. You can do so with **File &rarr; Save a copy in Drive**.\n"]},{"cell_type":"code","metadata":{"id":"RBkP5aBdfFkd","colab":{"base_uri":"https://localhost:8080/"},"executionInfo":{"status":"ok","timestamp":1606184259365,"user_tz":360,"elapsed":719,"user":{"displayName":"Pankaj Rajak","photoUrl":"https://lh3.googleusercontent.com/a-/AOh14GiTXe8B8vdrh4uTyUPqEtb-tRwPltWOZurP8-lxog=s64","userId":"04274727635571437779"}},"outputId":"7d1a640e-0749-41fd-a93a-e898b7805814"},"source":["import os\n","from google.colab import drive\n","drive.mount('/content/gdrive')\n","\n","! ls /content\n","!ls /content/gdrive/My\\ Drive"],"execution_count":null,"outputs":[{"output_type":"stream","text":["Drive already mounted at /content/gdrive; to attempt to forcibly remount, call drive.mount(\"/content/gdrive\", force_remount=True).\n","gdrive\tomniglot_resized  omniglot_resized.zip\tproject_data  sample_data\n"," 001.jpg\n"," 45%H2O-CaDPA_400C.avi\n"," 45%H2OCa-DPA_ROOMTemp.avi\n","'4657091371 (1).py'\n","'4657091371 - Pankaj Rajak (1).py'\n","'4657091371 - Pankaj Rajak.py'\n"," 4657091371.py\n"," acm_mac_2016_word_template.zip\n"," Adsorp_dataset_and_descriptors.xlsx.gsheet\n"," AIHW4_19thnov.zip\n"," AIHW4.-20thNov.zip\n"," AIHW4.zip\n"," AIHW_final3_25thNov.zip\n"," AIHW_final3.zip\n"," Air-force-Report\n"," al2o3cavity.xyz\n"," Al2O3-Facteting-Paper\n","'Al2O3 Fracture'\n"," al2o3sic.cmmf\n"," Al2O3_SIC.pptx\n"," Alumina_Self_Healing.pptx\n"," Alumina_Self_Healing_v5.pptx\n"," ANNMD_ForceModel_GST-v4.docx\n"," assignment1.zip\n"," assignment2.zip\n","'Battlefield Uses of Artifical Intelligence Study - W15QKN19X04YX - Federal Business Opportunities Opportunities.pdf'\n","'Colab Notebooks'\n"," config_gst_600_V0_ex.xyz\n"," CS402-SUMMER-2017\n"," CSCI699-Ultra_Fine_Entity_Typing.docx\n"," Cu_Melting.pptx\n","'Cybermanufacturing  SBIR.gov.pdf'\n"," Deep_Learning_CSCI-599\n"," Defense_Presentation\n"," DFT-seminar.tar.gz\n"," DOE-22May-4.docx.gdoc\n"," DOF.py\n"," DOS_Cv_plugins.zip\n"," dumpfrac.nve.zip\n"," exfoliation-images\n"," fig_damage\n"," Figure-1.png\n"," Figure2.png\n"," Figure3.png\n"," figure_S1.png\n"," figures.zip\n"," flow2000_800.ogv\n"," flow2000_800viewb.ogv\n"," flow2000_800viewc.ogv\n"," fracture-damagedzone.zip\n"," Friction.zip\n"," gan-results\n"," Hard_Material_Synthesis_RL.docx\n"," image_paper.zip\n"," inputdata.zip\n"," Kalia_Q2_CY2016_Jul21.docx\n"," kevlar\n"," kevlar.cmdf\n"," Kirigami_Results\n"," Learning_Python_Fifth_Edition.pdf\n"," Machine_Learning.pdf\n"," MASC_576\n","'MD MoSe2:WSe2.docx'\n"," meta-material.pxm\n"," mira_fracture1.pov.tga\n"," mira_fracture2.pov.tga\n"," ML_MD_Model\n"," ML_Paper_Force_Field\n","'ML_Research_CACS (1).pptx'\n"," ML_Research_CACS.pptx\n"," molecule\n"," MoS2_MoSe2_MD_thermal_conductivity.pdf\n"," MOS2_Thermal_Conductivity\n","'MoSe2-Wse2-Alloy-nat comm submission'\n"," MoSe2-WSe2-Fracture\n","'MoSe2-WSe2 Fracture Simulation.gdoc'\n"," movie.zip\n"," MoWSe2-Fracture\n","'MoWSe2-Fracture Paper .docx'\n"," MOWSe2paper\n"," MoWSe2-VAE-Paper.zip\n"," mrstalk\n","'Nano-letters (1).zip'\n"," Nano-letters.zip\n"," Ni_kagome-movies\n"," Ni_Meta_Material_Paper\n"," NLP_Question_anawering\n"," NN-model-ver1.jpg\n"," NN-model-ver2.jpg\n"," NN-model-ver3.jpg\n"," NN-model-ver4.jpg\n"," obj_3_500.zip\n","'Optimal Band gap prediction using Bayesian Optimizatiion'\n"," OS_Assignment\n"," os-kernel.tar.gz\n"," Pankaj_04thDec_RDX_Friction_JPCL\n"," Pankaj_Argonne_Talk\n"," pankaj_file.mtl\n"," pankaj_file.obj\n","'Pankaj_Machine_Learning-ver1 (1).pptx'\n"," Pankaj_Machine_Learning-ver1.pptx\n","'Pankaj_ML_CVD (1).pptx'\n","'Pankaj_ML_CVD (2).pptx'\n"," Pankaj_ML_CVD.pptx\n","'Pankaj ML Talks'\n"," Pankaj_MoWSe2_Fracture_VAE.pptx\n"," Pankaj_Presentation\n"," Pankaj_Rajak.docx\n"," Pankaj_Rajak_Qualstalk.pptx\n","'Pankaj_self_healing (1).pptx'\n","'Pankaj_self_healing (2).pptx'\n"," Pankaj_self_healing.pptx\n"," Pankaj_self_healing-v2.pptx\n"," Pankaj_self_healing-v3.pptx\n"," Pape1-selfhealing.zip\n"," Paper1-NickelCompression\n"," Pathline_100step\n"," Pathline_refineddata\n"," Patrick_Pankaj_Talk.pptx\n"," PDMS_Papers\n"," peptidoglycan_600K_a.mov\n"," project_data\n"," Proposal2.docx\n","'Quals Talk'\n"," RDXFrictionppt\n"," RDX_Friction.pptx\n"," RDX_Friction_Simulation\n"," RDX_Friction-slide11.pptx\n"," RDX_Friction-v3.pptx\n","'RDXFriction .zip'\n"," RDX_HMX\n"," RL_CVD-MRS.pptx\n"," RL_CVD_v9.docx\n"," RL-Kirigami-Presentation\n"," rxmd.zip\n"," sigma_1-5_epsilon_3.mp4\n"," sigma_3_epsilon_3.mp4\n"," Simulation_Movies\n","'Simulation setup for Mira.docx'\n","'Tax Refund Request.pdf'\n","'The element of Stastitical learning .pdf'\n"," ThermalConductivity.zip\n"," Thermal_conductivty\n"," VAE-Paper\n"," water\n"," water-aanalysis-code.zip\n"," work\n"],"name":"stdout"}]},{"cell_type":"code","metadata":{"colab":{"base_uri":"https://localhost:8080/"},"id":"1sxgn1rWN6qf","executionInfo":{"status":"ok","timestamp":1606184259563,"user_tz":360,"elapsed":910,"user":{"displayName":"Pankaj Rajak","photoUrl":"https://lh3.googleusercontent.com/a-/AOh14GiTXe8B8vdrh4uTyUPqEtb-tRwPltWOZurP8-lxog=s64","userId":"04274727635571437779"}},"outputId":"4aa8dfd6-0303-4d34-f835-83434a34cd74"},"source":["DRIVE_PATH = '/content/gdrive/My\\ Drive/project_data'\n","DRIVE_PYTHON_PATH = DRIVE_PATH.replace('\\\\', '')\n","print(\"DRIVE_PATH:\", DRIVE_PATH)\n","print(\"DRIVE_PYTHON_PATH: \",DRIVE_PYTHON_PATH)\n","if not os.path.exists(DRIVE_PYTHON_PATH):\n","  %mkdir $DRIVE_PATH\n","\n","## the space in `My Drive` causes some issues,\n","## make a symlink to avoid this\n","SYM_PATH = '/content/project_data'\n","if not os.path.exists(SYM_PATH):\n","  !ln -s $DRIVE_PATH $SYM_PATH"],"execution_count":null,"outputs":[{"output_type":"stream","text":["DRIVE_PATH: /content/gdrive/My\\ Drive/project_data\n","DRIVE_PYTHON_PATH:  /content/gdrive/My Drive/project_data\n"],"name":"stdout"}]},{"cell_type":"code","metadata":{"id":"s0uo9NjlLHmC"},"source":["from google_drive_downloader import GoogleDriveDownloader as gdd\n","\n","# Need to download the Omniglot dataset -- DON'T MODIFY THIS CELL\n","#data_path = '{}/data/omniglot_resized.zip'.format(SYM_PATH)\n","#dest_path = '{}/data/omniglot_resized'.format(SYM_PATH)\n","if not os.path.isdir('./omniglot_resized'):\n","    gdd.download_file_from_google_drive(file_id='1iaSFXIYC3AB8q9K_M-oVMa4pmB7yKMtI',\n","                                        dest_path='./omniglot_resized.zip',\n","                                        unzip=True)\n","\n","assert os.path.isdir('./omniglot_resized')"],"execution_count":null,"outputs":[]},{"cell_type":"code","metadata":{"id":"9llNF0vdTNCl"},"source":["\"\"\"Data loading scripts\"\"\"\n","## NOTE: You do not need to modify this block but you will need to use it.\n","import numpy as np\n","import os\n","import random\n","import tensorflow as tf\n","from scipy import misc\n","import imageio\n","\n","def get_images(paths, labels, n_samples=None, shuffle=True):\n","  \"\"\"\n","  Takes a set of character folders and labels and returns paths to image files\n","  paired with labels.\n","  Args:\n","    paths: A list of character folders\n","    labels: List or numpy array of same length as paths\n","    n_samples: Number of images to retrieve per character\n","  Returns:\n","    List of (label, image_path) tuples\n","  \"\"\"\n","  if n_samples is not None:\n","    sampler = lambda x: random.sample(x, n_samples)\n","  else:\n","    sampler = lambda x: x\n","  images_labels = [(i, os.path.join(path, image))\n","           for i, path in zip(labels, paths)\n","           for image in sampler(os.listdir(path))]\n","  if shuffle:\n","    random.shuffle(images_labels)\n","  return images_labels\n","\n","\n","def image_file_to_array(filename, dim_input):\n","  \"\"\"\n","  Takes an image path and returns numpy array\n","  Args:\n","    filename: Image filename\n","    dim_input: Flattened shape of image\n","  Returns:\n","    1 channel image\n","  \"\"\"\n","  image = imageio.imread(filename)\n","  image = image.reshape([dim_input])\n","  image = image.astype(np.float32) / 255.0\n","  image = 1.0 - image\n","  return image\n","\n","\n","class DataGenerator(object):\n","  \"\"\"\n","  Data Generator capable of generating batches of Omniglot data.\n","  A \"class\" is considered a class of omniglot digits.\n","  \"\"\"\n","\n","  def __init__(self, num_classes, num_samples_per_class, num_meta_test_classes, num_meta_test_samples_per_class, \n","               train_type ='memorization', config={}):\n","    \"\"\"\n","    Args:\n","      num_classes: Number of classes for classification (K-way)\n","      num_samples_per_class: num samples to generate per class in one batch\n","      num_meta_test_classes: Number of classes for classification (K-way) at meta-test time\n","      num_meta_test_samples_per_class: num samples to generate per class in one batch at meta-test time\n","      batch_size: size of meta batch size (e.g. number of functions)\n","    \"\"\"\n","    self.num_samples_per_class = num_samples_per_class\n","    self.num_classes = num_classes\n","    self.num_meta_test_samples_per_class = num_meta_test_samples_per_class\n","    self.num_meta_test_classes = num_meta_test_classes\n","    self.train_type = train_type\n","\n","    data_folder = config.get('data_folder', './omniglot_resized')\n","    self.img_size = config.get('img_size', (28, 28))\n","\n","    self.dim_input = np.prod(self.img_size)\n","    self.dim_output = self.num_classes\n","\n","    character_folders = [os.path.join(data_folder, family, character)\n","               for family in os.listdir(data_folder)\n","               if os.path.isdir(os.path.join(data_folder, family))\n","               for character in os.listdir(os.path.join(data_folder, family))\n","               if os.path.isdir(os.path.join(data_folder, family, character))]\n","\n","    random.seed(123)\n","    #random.shuffle(character_folders)\n","    num_val = 100\n","    num_train = 1200\n","    self.metatrain_character_folders = character_folders[: num_train]\n","    self.metaval_character_folders = character_folders[\n","      num_train:num_train + num_val]\n","    self.metatest_character_folders = character_folders[\n","      num_train + num_val:]\n","    \n","    print(\"Number of training examples. : \",len(self.metatrain_character_folders))\n","    print(\"Number of validation examples: \",len(self.metaval_character_folders))\n","    print(\"Number of test examples      : \",len(self.metatest_character_folders))\n","    print(\"Training Type: \",self.train_type)\n","\n","    if self.train_type == 'memorization':\n","      self.metatrain_label = [i%self.num_classes for i in range(len(self.metatrain_character_folders))]\n","      self.metaval_label = [i%self.num_classes for i in range(len(self.metaval_character_folders))]\n","      self.metatest_label = [i%self.num_classes for i in range(len(self.metatest_character_folders))]\n","\n","  def sample_batch(self, batch_type, batch_size, shuffle=True, swap=False):\n","    \"\"\"\n","    Samples a batch for training, validation, or testing\n","    Args:\n","      batch_type: meta_train/meta_val/meta_test\n","      shuffle: randomly shuffle classes or not\n","      swap: swap number of classes (N) and number of samples per class (K) or not\n","    Returns:\n","      A a tuple of (1) Image batch and (2) Label batch where\n","      image batch has shape [B, N, K, 784] and label batch has shape [B, N, K, N] if swap is False\n","      where B is batch size, K is number of samples per class, N is number of classes\n","    \"\"\"\n","    if batch_type == \"meta_train\":\n","      folders = self.metatrain_character_folders\n","      num_classes = self.num_classes\n","      num_samples_per_class = self.num_samples_per_class\n","      if self.train_type == 'memorization':\n","        class_label = self.metatrain_label\n","    elif batch_type == \"meta_val\":\n","      folders = self.metaval_character_folders\n","      num_classes = self.num_classes\n","      num_samples_per_class = self.num_samples_per_class\n","      if self.train_type == 'memorization':\n","        class_label = self.metaval_label\n","    else:\n","      folders = self.metatest_character_folders\n","      num_classes = self.num_meta_test_classes\n","      num_samples_per_class = self.num_meta_test_samples_per_class\n","      if self.train_type == 'memorization':\n","        class_label = self.metatest_label\n","\n","    all_image_batches, all_label_batches = [], []\n","    total_sample = [val for val in range(len(folders))]\n","    for i in range(batch_size):\n","      data_list = random.sample(total_sample, num_classes)\n","      #print(\"data_list: \",data_list)\n","      sampled_character_folders = [folders[n_val] for n_val in data_list]\n","      if self.train_type == 'memorization':\n","        sampled_label = [class_label[n_val] for n_val in data_list]\n","      else:\n","        sampled_label = [n_val for n_val in range(num_classes)]\n","      #print(\"sampled_label:\",sampled_label)\n","      #print(\"sampled_character_folders: \",sampled_character_folders)\n","      labels_and_images = get_images(sampled_character_folders,sampled_label,n_samples=num_samples_per_class,shuffle=False)\n","      labels = [li[0] for li in labels_and_images]\n","      images = [image_file_to_array(\n","        li[1], self.dim_input) for li in labels_and_images]\n","      images = np.stack(images)\n","      labels = np.array(labels).astype(np.int32)\n","      labels = np.reshape(\n","        labels, (num_classes, num_samples_per_class))\n","      labels = np.eye(num_classes, dtype=np.float32)[labels]\n","      images = np.reshape(\n","        images, (num_classes, num_samples_per_class, -1))\n","\n","      batch = np.concatenate([labels, images], 2)\n","      if shuffle:\n","        for p in range(num_samples_per_class):\n","          np.random.shuffle(batch[:, p])\n","\n","      labels = batch[:, :, :num_classes]\n","      images = batch[:, :, num_classes:]\n","\n","      if swap:\n","        labels = np.swapaxes(labels, 0, 1)\n","        images = np.swapaxes(images, 0, 1)\n","\n","      all_image_batches.append(images)\n","      all_label_batches.append(labels)\n","    all_image_batches = np.stack(all_image_batches)\n","    all_label_batches = np.stack(all_label_batches)\n","    return all_image_batches, all_label_batches"],"execution_count":null,"outputs":[]},{"cell_type":"code","metadata":{"id":"7YHtsAXJZPzl"},"source":["\"\"\" Utility functions. \"\"\"\n","## NOTE: You do not need to modify this block but you will need to use it.\n","import numpy as np\n","import os\n","import random\n","import tensorflow as tf\n","\n","## Loss utilities\n","def cross_entropy_loss(pred, label, k_shot):\n","    return tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(logits=pred, labels=tf.stop_gradient(label)) / k_shot)\n","\n","def accuracy(labels, predictions):\n","  return tf.reduce_mean(tf.cast(tf.equal(labels, predictions), dtype=tf.float32))"],"execution_count":null,"outputs":[]},{"cell_type":"code","metadata":{"id":"hZyrHLzAa6sI"},"source":["\"\"\"Convolutional layers used by MAML model.\"\"\"\n","## NOTE: You do not need to modify this block but you will need to use it.\n","seed = 123\n","def conv_block(inp, cweight, bweight, bn, activation=tf.nn.relu, stride=1):\n","    \"\"\" Perform, conv, batch norm, nonlinearity, and max pool \"\"\"\n","    strides = [1,stride,stride,1]\n","\n","    conv_output = tf.nn.conv2d(input=inp, filters=cweight, strides=strides, padding='SAME') + bweight\n","    normed = bn(conv_output)\n","    normed = activation(normed)\n","    return normed\n","  \n","class ConvLayers(tf.keras.layers.Layer):\n","    def __init__(self, channels, dim_hidden, dim_output, img_size):\n","        super(ConvLayers, self).__init__()\n","        self.channels = channels\n","        self.dim_hidden = dim_hidden\n","        self.dim_output = dim_output\n","        self.img_size = img_size\n","        weights = {}\n","        dtype = tf.float32\n","        weight_initializer =  tf.keras.initializers.GlorotUniform()\n","        k = 3\n","        weights['conv1'] = tf.Variable(weight_initializer(shape=[k, k, self.channels, self.dim_hidden]), name='conv1', dtype=dtype)\n","        weights['b1'] = tf.Variable(tf.zeros([self.dim_hidden]), name='b1')\n","        self.bn1 = tf.keras.layers.BatchNormalization(name='bn1')\n","        weights['conv2'] = tf.Variable(weight_initializer(shape=[k, k, self.dim_hidden, 2*self.dim_hidden]), name='conv2', dtype=dtype)\n","        weights['b2'] = tf.Variable(tf.zeros([2*self.dim_hidden]), name='b2')\n","        self.bn2 = tf.keras.layers.BatchNormalization(name='bn2')\n","        weights['conv3'] = tf.Variable(weight_initializer(shape=[k, k, 2*self.dim_hidden, 4*self.dim_hidden]), name='conv3', dtype=dtype)\n","        weights['b3'] = tf.Variable(tf.zeros([4*self.dim_hidden]), name='b3')\n","        self.bn3 = tf.keras.layers.BatchNormalization(name='bn3')\n","        weights['conv4'] = tf.Variable(weight_initializer([k, k, 4*self.dim_hidden, 4*self.dim_hidden]), name='conv4', dtype=dtype)\n","        weights['b4'] = tf.Variable(tf.zeros([4*self.dim_hidden]), name='b4')\n","        self.bn4 = tf.keras.layers.BatchNormalization(name='bn4')\n","        weights['w5'] = tf.Variable(weight_initializer(shape=[4*self.dim_hidden, self.dim_output]), name='w5', dtype=dtype)\n","        weights['b5'] = tf.Variable(tf.zeros([self.dim_output]), name='b5')\n","        self.conv_weights = weights\n","    def call(self, inp, weights):\n","        channels = self.channels\n","        inp = tf.reshape(inp, [-1, self.img_size, self.img_size, channels])\n","        hidden1 = conv_block(inp, weights['conv1'], weights['b1'], self.bn1, stride=1)\n","        hidden2 = conv_block(hidden1, weights['conv2'], weights['b2'], self.bn2, stride=2)\n","        hidden3 = conv_block(hidden2, weights['conv3'], weights['b3'], self.bn3, stride=2)\n","        hidden4 = conv_block(hidden3, weights['conv4'], weights['b4'], self.bn4, stride=1)\n","        hidden4 = tf.reduce_mean(input_tensor=hidden4, axis=[1, 2])\n","        return tf.matmul(hidden4, weights['w5']) + weights['b5']"],"execution_count":null,"outputs":[]},{"cell_type":"code","metadata":{"id":"-2wPuGX42WNF"},"source":["\"\"\"MAML model code\"\"\"\n","import numpy as np\n","import sys\n","import tensorflow as tf\n","from functools import partial\n","\n","\n","class MAML(tf.keras.Model):\n","  def __init__(self, dim_input=1, dim_output=1,\n","               num_inner_updates=1,\n","               inner_update_lr=0.4, num_filters=32, k_shot=5, learn_inner_update_lr=False):\n","    super(MAML, self).__init__()\n","    self.dim_input = dim_input\n","    self.dim_output = dim_output\n","    self.inner_update_lr = inner_update_lr\n","    #self.loss_func = partial(cross_entropy_loss, k_shot=k_shot) \n","    self.loss_func = partial(cross_entropy_loss, k_shot=1)\n","    self.dim_hidden = num_filters\n","    self.channels = 1\n","    self.img_size = int(np.sqrt(self.dim_input/self.channels))\n","\n","    # outputs_ts[i] and losses_ts_post[i] are the output and loss after i+1 inner gradient updates\n","    losses_tr_pre, outputs_tr, losses_ts_post, outputs_ts = [], [], [], []\n","    accuracies_tr_pre, accuracies_ts = [], []\n","\n","    # for each loop in the inner training loop\n","    outputs_ts = [[]]*num_inner_updates\n","    losses_ts_post = [[]]*num_inner_updates\n","    accuracies_ts = [[]]*num_inner_updates\n","\n","    # Define the weights - these should NOT be directly modified by the\n","    # inner training loop\n","    tf.random.set_seed(seed)\n","    self.conv_layers = ConvLayers(self.channels, self.dim_hidden, self.dim_output, self.img_size)\n","\n","    self.learn_inner_update_lr = learn_inner_update_lr\n","    if self.learn_inner_update_lr:\n","      self.inner_update_lr_dict = {}\n","      for key in self.conv_layers.conv_weights.keys():\n","        self.inner_update_lr_dict[key] = [tf.Variable(self.inner_update_lr, name='inner_update_lr_%s_%d' % (key, j)) for j in range(num_inner_updates)]\n","  \n","  @tf.function\n","  def call(self, inp, meta_batch_size=25, num_inner_updates=1):\n","    def task_inner_loop(inp, reuse=True,\n","                      meta_batch_size=25, num_inner_updates=1):\n","      \"\"\"\n","        Perform gradient descent for one task in the meta-batch (i.e. inner-loop).\n","        Args:\n","          inp: a tuple (input_tr, input_ts, label_tr, label_ts), where input_tr and label_tr are the inputs and\n","            labels used for calculating inner loop gradients and input_ts and label_ts are the inputs and\n","            labels used for evaluating the model after inner updates.\n","            Should be shapes:\n","              input_tr: [N*K, 784]\n","              input_ts: [N*K, 784]\n","              label_tr: [N*K, N]\n","              label_ts: [N*K, N]\n","        Returns:\n","          task_output: a list of outputs, losses and accuracies at each inner update\n","      \"\"\"\n","      # the inner and outer loop data\n","      input_tr, input_ts, label_tr, label_ts = inp\n","\n","      # weights corresponds to the initial weights in MAML (i.e. the meta-parameters)\n","      weights = self.conv_layers.conv_weights\n","\n","      # the predicted outputs, loss values, and accuracy for the pre-update model (with the initial weights)\n","      # evaluated on the inner loop training data\n","      task_output_tr_pre, task_loss_tr_pre, task_accuracy_tr_pre = None, None, None\n","\n","      # lists to keep track of outputs, losses, and accuracies of test data for each inner_update\n","      # where task_outputs_ts[i], task_losses_ts[i], task_accuracies_ts[i] are the output, loss, and accuracy\n","      # after i+1 inner gradient updates\n","      task_outputs_ts, task_losses_ts, task_accuracies_ts = [], [], []\n","  \n","      #############################\n","      #### YOUR CODE GOES HERE ####\n","      # perform num_inner_updates to get modified weights\n","      # modified weights should be used to evaluate performance\n","      # Note that at each inner update, always use input_tr and label_tr for calculating gradients\n","      # and use input_ts and labels for evaluating performance\n","\n","      # HINTS: You will need to use tf.GradientTape().\n","      # Read through the tf.GradientTape() documentation to see how 'persistent' should be set.\n","      # Here is some documentation that may be useful: \n","      # https://www.tensorflow.org/guide/advanced_autodiff#higher-order_gradients\n","      # https://www.tensorflow.org/api_docs/python/tf/GradientTape\n","      weights_old = weights\n","      weights_new = []\n","      with tf.GradientTape(persistent=True) as inner_tape:\n","        # takes i inner gradient steps\n","        for i in range(num_inner_updates):\n","          task_output_tr_pre = self.conv_layers(input_tr,weights_old)\n","          task_loss_tr_pre = self.loss_func(task_output_tr_pre,label_tr)\n","          grad_val = inner_tape.gradient(task_loss_tr_pre,weights_old)\n","          #update the weight\n","          w_dict = {}\n","          if self.learn_inner_update_lr:\n","            for key in weights_old.keys():\n","              w_dict[key] = weights_old[key] - self.inner_update_lr_dict[key] * grad_val[key]\n","          else:\n","            for key in weights_old.keys():\n","              w_dict[key] = weights_old[key] - self.inner_update_lr * grad_val[key]\n","          weights_new.append(w_dict)\n","          #compute outer loss\n","          task_outputs_ts.append(self.conv_layers(input_ts,weights_new[i]))\n","          task_losses_ts.append(self.loss_func(task_outputs_ts[i],label_ts))\n","          weights_old = weights_new[i]\n","      weights_final = weights_new[-1]\n","      w_list = []\n","      for key in weights_final.keys():\n","        w_list.extend([tf.reshape(weights_final[key],[-1])])\n","        #####print(key,tf.reshape(weights_final[key],[-1]))\n","      #####print(\"w_list length\",len(w_list))\n","      #print('out:',input_tr.shape,task_output_tr_pre.shape,type(task_output_tr_pre),task_loss_tr_pre)\n","      #for key in weights.keys():\n","      #  print(\"weight key: \",key,grad_val[key].shape,weights[key].shape,weights_new[-1][key].shape,type(weights[key]),type(weights_new[-1][key]))\n","      #############################\n","\n","      # Compute accuracies from output predictions\n","      task_accuracy_tr_pre = accuracy(tf.argmax(input=label_tr, axis=1), tf.argmax(input=tf.nn.softmax(task_output_tr_pre), axis=1))\n","\n","      for j in range(num_inner_updates):\n","        task_accuracies_ts.append(accuracy(tf.argmax(input=label_ts, axis=1), tf.argmax(input=tf.nn.softmax(task_outputs_ts[j]), axis=1)))\n","\n","      task_output = [task_output_tr_pre, task_outputs_ts, task_loss_tr_pre, task_losses_ts, task_accuracy_tr_pre, task_accuracies_ts, w_list]\n","\n","      #print(\"iteration i finished\")\n","\n","      return task_output\n","\n","    input_tr, input_ts, label_tr, label_ts = inp\n","    # to initialize the batch norm vars, might want to combine this, and not run idx 0 twice.\n","    unused = task_inner_loop((input_tr[0], input_ts[0], label_tr[0], label_ts[0]),\n","                          False,\n","                          meta_batch_size,\n","                          num_inner_updates)\n","    out_dtype = [tf.float32, [tf.float32]*num_inner_updates, tf.float32, [tf.float32]*num_inner_updates]\n","    out_dtype.extend([tf.float32, [tf.float32]*num_inner_updates])\n","    #add inof related to model paramters \n","    model_param = []\n","    for key in self.conv_layers.conv_weights.keys():\n","      model_param.extend([tf.reshape(self.conv_layers.conv_weights[key], [-1])])\n","      ####print(\"model dict actual: \",key,tf.reshape(self.conv_layers.conv_weights[key], [-1]))\n","    ####print(\"model param list: \",len(model_param))\n","    ######for i in range(len(model_param)):\n","    ######  print(\"model param info: \",len(model_param[i]))\n","    out_dtype.extend([[tf.float32]*len(model_param)])\n","\n","    task_inner_loop_partial = partial(task_inner_loop, meta_batch_size=meta_batch_size, num_inner_updates=num_inner_updates)\n","    \n","    result = tf.map_fn(task_inner_loop_partial,\n","                    elems=(input_tr, input_ts, label_tr, label_ts),\n","                    dtype=out_dtype,\n","                    parallel_iterations=meta_batch_size)\n","    return result"],"execution_count":null,"outputs":[]},{"cell_type":"code","metadata":{"id":"Tj55dJTHAesA"},"source":["\"\"\"Model training code\"\"\"\n","\"\"\"\n","Usage Instructions:\n","  5-way, 1-shot omniglot:\n","    python main.py --meta_train_iterations=15000 --meta_batch_size=25 --k_shot=1 --inner_update_lr=0.4 --num_inner_updates=1 --logdir=logs/omniglot5way/\n","  20-way, 1-shot omniglot:\n","    python main.py --meta_train_iterations=15000 --meta_batch_size=16 --k_shot=1 --n_way=20 --inner_update_lr=0.1 --num_inner_updates=5 --logdir=logs/omniglot20way/\n","  To run evaluation, use the '--meta_train=False' flag and the '--meta_test_set=True' flag to use the meta-test set.\n","\"\"\"\n","import csv\n","import numpy as np\n","import pickle\n","import random\n","import tensorflow as tf\n","\n","@tf.function\n","def outer_train_step(inp, model, optim, meta_batch_size=25, num_inner_updates=1):\n","  with tf.GradientTape(persistent=False) as outer_tape:\n","    result = model(inp, meta_batch_size=meta_batch_size, num_inner_updates=num_inner_updates)\n","\n","    outputs_tr, outputs_ts, losses_tr_pre, losses_ts, accuracies_tr_pre, accuracies_ts, w_list = result\n","    #print(\"====== info related to w_list=====\",len(w_list))\n","    #for i in range(len(w_list)):\n","    #  print(\"===wi\",len(w_list[i]),'val',w_list[i])\n","\n","    w_list =tf.concat(w_list, axis=-1)\n","    #print(\"w_list new length: \",len(w_list),len(w_list[0]),len(w_list[1]))\n","    #======= compute meta-regularization =======\n","    count = 0\n","    input_tr, input_ts, label_tr, label_ts = inp\n","    for i in range(0,meta_batch_size-1):\n","      for j in range(i+1,meta_batch_size):\n","        count+=1\n","        # get norm of the parameters \n","        h_val = tf.norm(w_list[i] - w_list[j],ord='euclidean')\n","        # get the norm of the input data\n","        zz =  tf.stop_gradient(tf.norm(input_tr[i] - input_tr[j],ord='euclidean'))\n","        l2_norm = tf.abs(4.0 - (h_val/zz))  #h_val/zz\n","        #compute the regularization for given (i,j) pair\n","        if i == 0 and j ==1:\n","          meta_reg = l2_norm\n","          z_dist = zz\n","          h_dist = h_val\n","        else:\n","          meta_reg = meta_reg + l2_norm\n","          z_dist = z_dist + zz\n","          h_dist = h_dist + h_val\n","    meta_reg = meta_reg / count\n","    z_dist = z_dist / count\n","    h_dist = h_dist / count\n","\n","    total_losses_ts = [tf.reduce_mean(loss_ts) for loss_ts in losses_ts]\n","\n","    loss = total_losses_ts[-1] + meta_reg\n","\n","  gradients = outer_tape.gradient(loss, model.trainable_variables)\n","  optim.apply_gradients(zip(gradients, model.trainable_variables))\n","\n","  total_loss_tr_pre = tf.reduce_mean(losses_tr_pre)\n","  total_accuracy_tr_pre = tf.reduce_mean(accuracies_tr_pre)\n","  total_accuracies_ts = [tf.reduce_mean(accuracy_ts) for accuracy_ts in accuracies_ts]\n","\n","  return [outputs_tr, outputs_ts, total_loss_tr_pre, total_losses_ts, total_accuracy_tr_pre, total_accuracies_ts], meta_reg\n","\n","@tf.function\n","def outer_eval_step(inp, model, meta_batch_size=25, num_inner_updates=1):\n","  result = model(inp, meta_batch_size=meta_batch_size, num_inner_updates=num_inner_updates)\n","\n","  outputs_tr, outputs_ts, losses_tr_pre, losses_ts, accuracies_tr_pre, accuracies_ts, w_list = result\n","\n","  total_loss_tr_pre = tf.reduce_mean(losses_tr_pre)\n","  total_losses_ts = [tf.reduce_mean(loss_ts) for loss_ts in losses_ts]\n","\n","  total_accuracy_tr_pre = tf.reduce_mean(accuracies_tr_pre)\n","  total_accuracies_ts = [tf.reduce_mean(accuracy_ts) for accuracy_ts in accuracies_ts]\n","\n","  return outputs_tr, outputs_ts, total_loss_tr_pre, total_losses_ts, total_accuracy_tr_pre, total_accuracies_ts  \n","\n","\n","def meta_train_fn(model, exp_string, data_generator,\n","               n_way=5, meta_train_iterations=15000, meta_batch_size=25,\n","               log=True, logdir='/tmp/data', k_shot=1, num_inner_updates=1, meta_lr=0.001):\n","  SUMMARY_INTERVAL = 10\n","  SAVE_INTERVAL = 100\n","  PRINT_INTERVAL = 10  \n","  TEST_PRINT_INTERVAL = PRINT_INTERVAL*5\n","\n","  pre_accuracies, post_accuracies = [], []\n","\n","  num_classes = data_generator.num_classes\n","\n","  optimizer = tf.keras.optimizers.Adam(learning_rate=meta_lr)\n","\n","  pre_val_accuracy_trj = []\n","  post_val_accuracy_trj = []\n","  pre_train_accuracy_trj = []\n","  post_train_accuracy_trj = []\n","  step_train_trj = []\n","  step_num_trj = []\n","\n","  for itr in range(meta_train_iterations):\n","    #############################\n","    #### YOUR CODE GOES HERE ####\n","\n","    # sample a batch of training data and partition into\n","    # the support/training set (input_tr, label_tr) and the query/test set (input_ts, label_ts)\n","    # NOTE: The code assumes that the support and query sets have the same number of examples.\n","    image_data, image_label = data_generator.sample_batch('meta_train', meta_batch_size, shuffle=True, swap=False)\n","    input_tr = image_data[:,:,:k_shot,:].reshape(meta_batch_size,n_way*k_shot,-1)\n","    label_tr = image_label[:,:,:k_shot,:].reshape(meta_batch_size,n_way*k_shot,-1)\n","    input_ts = image_data[:,:,k_shot:,:].reshape(meta_batch_size,n_way*k_shot,-1)\n","    label_ts = image_label[:,:,k_shot:,:].reshape(meta_batch_size,n_way*k_shot,-1)\n","    #############################\n","\n","    inp = (input_tr, input_ts, label_tr, label_ts)\n","    \n","    result, meta_reg = outer_train_step(inp, model, optimizer, meta_batch_size=meta_batch_size, num_inner_updates=num_inner_updates)\n","\n","    if itr % SUMMARY_INTERVAL == 0:\n","      pre_accuracies.append(result[-2])\n","      post_accuracies.append(result[-1][-1])\n","\n","    if (itr!=0) and itr % PRINT_INTERVAL == 0:\n","      print_str = 'Iteration %d: pre-inner-loop train accuracy: %.5f, post-inner-loop test accuracy: %.5f, meta_reg : %5f' % (itr, np.mean(pre_accuracies), np.mean(post_accuracies),meta_reg)\n","      print(print_str)\n","      pre_train_accuracy_trj.append(np.mean(pre_accuracies))\n","      post_train_accuracy_trj.append( np.mean(post_accuracies))\n","      step_train_trj.append(itr)\n","      pre_accuracies, post_accuracies = [], []\n","\n","    if (itr!=0) and itr % TEST_PRINT_INTERVAL == 0:\n","      #############################\n","      #### YOUR CODE GOES HERE ####\n","\n","      # sample a batch of validation data and partition it into\n","      # the support/training set (input_tr, label_tr) and the query/test set (input_ts, label_ts)\n","      # NOTE: The code assumes that the support and query sets have the same number of examples.\n","      image_data, image_label = data_generator.sample_batch('meta_val', meta_batch_size, shuffle=True, swap=False)\n","      input_tr = image_data[:,:,:k_shot,:].reshape(meta_batch_size,n_way*k_shot,-1)\n","      label_tr = image_label[:,:,:k_shot,:].reshape(meta_batch_size,n_way*k_shot,-1)\n","      input_ts = image_data[:,:,k_shot:,:].reshape(meta_batch_size,n_way*k_shot,-1)\n","      label_ts = image_label[:,:,k_shot:,:].reshape(meta_batch_size,n_way*k_shot,-1)\n","      #############################\n","\n","      inp = (input_tr, input_ts, label_tr, label_ts)\n","      result = outer_eval_step(inp, model, meta_batch_size=meta_batch_size, num_inner_updates=num_inner_updates)\n","\n","      print('Meta-validation pre-inner-loop train accuracy: %.5f, meta-validation post-inner-loop test accuracy: %.5f' % (result[-2], result[-1][-1]))\n","      pre_val_accuracy_trj.append(result[-2])\n","      post_val_accuracy_trj.append(result[-1][-1])\n","      step_num_trj.append(itr)\n","\n","  model_file = logdir + '/' + exp_string +  '/model' + str(itr)\n","  print(\"Saving to \", model_file)\n","  model.save_weights(model_file)\n","  return step_num_trj,pre_val_accuracy_trj,post_val_accuracy_trj,step_train_trj,pre_train_accuracy_trj,post_train_accuracy_trj\n","\n","# calculated for omniglot\n","NUM_META_TEST_POINTS = 600\n","\n","def meta_test_fn(model, data_generator, n_way=5, meta_batch_size=25, k_shot=1,\n","              num_inner_updates=1):\n","  \n","  num_classes = data_generator.num_classes\n","\n","  np.random.seed(1)\n","  random.seed(1)\n","\n","  meta_test_accuracies = []\n","  pre_accuracies = []\n","  post_accuracies = []\n","\n","  for _ in range(NUM_META_TEST_POINTS):\n","    #############################\n","    #### YOUR CODE GOES HERE ####\n","\n","    # sample a batch of test data and partition it into\n","    # the support/training set (input_tr, label_tr) and the query/test set (input_ts, label_ts)\n","    # NOTE: The code assumes that the support and query sets have the same number of examples.\n","    image_data, image_label = data_generator.sample_batch('meta_test', meta_batch_size, shuffle=True, swap=False)\n","    input_tr = image_data[:,:,:k_shot,:].reshape(meta_batch_size,n_way*k_shot,-1)\n","    label_tr = image_label[:,:,:k_shot,:].reshape(meta_batch_size,n_way*k_shot,-1)\n","    input_ts = image_data[:,:,k_shot:,:].reshape(meta_batch_size,n_way*k_shot,-1)\n","    label_ts = image_label[:,:,k_shot:,:].reshape(meta_batch_size,n_way*k_shot,-1)    \n","    #############################\n","    inp = (input_tr, input_ts, label_tr, label_ts)\n","    result = outer_eval_step(inp, model, meta_batch_size=meta_batch_size, num_inner_updates=num_inner_updates)\n","\n","    pre_accuracies.append(result[-2])\n","    post_accuracies.append(result[-1][-1])\n","    meta_test_accuracies.append(result[-1][-1])\n","\n","  meta_test_accuracies = np.array(meta_test_accuracies)\n","  means = np.mean(meta_test_accuracies)\n","  stds = np.std(meta_test_accuracies)\n","  ci95 = 1.96*stds/np.sqrt(NUM_META_TEST_POINTS)\n","\n","  print('Mean meta-test accuracy/loss, stddev, and confidence intervals')\n","  print((means, stds, ci95))\n","  print('pre accuracy stats')\n","  print('mean',np.mean(pre_accuracies),'std',np.std(pre_accuracies))\n","  print('post accuracy stats')\n","  print('mean',np.mean(post_accuracies),'std',np.std(post_accuracies))\n","  return means,stds,ci95\n","\n","\n","def run_maml(n_way=5, k_shot=1, meta_batch_size=25, meta_lr=0.001,\n","             inner_update_lr=0.4, num_filters=32, num_inner_updates=1,\n","             learn_inner_update_lr=False,\n","             resume=False, resume_itr=0, log=True, logdir='/tmp/data',\n","             data_path='./omniglot_resized',meta_train=True,\n","             meta_train_iterations=15000, meta_train_k_shot=-1,\n","             meta_train_inner_update_lr=-1,train_type='memorization'):\n","\n","\n","  # call data_generator and get data with k_shot*2 samples per class\n","  data_generator = DataGenerator(n_way, k_shot*2, n_way, k_shot*2,train_type =train_type, config={'data_folder': data_path})\n","\n","  # set up MAML model\n","  dim_output = data_generator.dim_output\n","  dim_input = data_generator.dim_input\n","  model = MAML(dim_input,\n","              dim_output,\n","              num_inner_updates=num_inner_updates,\n","              inner_update_lr=inner_update_lr,\n","              k_shot=k_shot,\n","              num_filters=num_filters,\n","              learn_inner_update_lr=learn_inner_update_lr)\n","\n","  if meta_train_k_shot == -1:\n","    meta_train_k_shot = k_shot\n","  if meta_train_inner_update_lr == -1:\n","    meta_train_inner_update_lr = inner_update_lr\n","\n","  exp_string = 'cls_'+str(n_way)+'.mbs_'+str(meta_batch_size) + '.k_shot_' + str(meta_train_k_shot) + '.inner_numstep_' + str(num_inner_updates) + '.inner_updatelr_' + str(meta_train_inner_update_lr) + '.learn_inner_update_lr_' + str(learn_inner_update_lr) + '.train_type' + str(train_type)\n","\n","\n","  if meta_train:\n","    step_num_trj,pre_val_accuracy_trj,post_val_accuracy_trj,step_train_trj,pre_train_accuracy_trj,post_train_accuracy_trj = meta_train_fn(model, exp_string, data_generator,\n","                  n_way, meta_train_iterations, meta_batch_size, log, logdir,\n","                  k_shot, num_inner_updates, meta_lr)\n","    #do evalvulation on the Meta-test data\n","    meta_batch_size = 1\n","    means,stds,ci95=meta_test_fn(model, data_generator, n_way, meta_batch_size, k_shot, num_inner_updates)\n","    return step_num_trj,pre_val_accuracy_trj,post_val_accuracy_trj,step_train_trj,pre_train_accuracy_trj,post_train_accuracy_trj\n","  else:\n","    meta_batch_size = 1\n","\n","    model_file = tf.train.latest_checkpoint(logdir + '/' + exp_string)\n","    print(\"Restoring model weights from \", model_file)\n","    model.load_weights(model_file)\n","    means,stds,ci95=meta_test_fn(model, data_generator, n_way, meta_batch_size, k_shot, num_inner_updates)\n","    return means,stds,ci95"],"execution_count":null,"outputs":[]},{"cell_type":"code","metadata":{"colab":{"base_uri":"https://localhost:8080/"},"id":"67__oMbZB7FW","executionInfo":{"status":"ok","timestamp":1606184260843,"user_tz":360,"elapsed":2161,"user":{"displayName":"Pankaj Rajak","photoUrl":"https://lh3.googleusercontent.com/a-/AOh14GiTXe8B8vdrh4uTyUPqEtb-tRwPltWOZurP8-lxog=s64","userId":"04274727635571437779"}},"outputId":"404eb143-609b-4ab0-cd4b-f1a86c0f3fea"},"source":["log_path = '{}/data/checkpoint'.format(SYM_PATH)\n","if not os.path.exists(log_path):\n","  %mkdir $log_path\n","print(log_path)"],"execution_count":null,"outputs":[{"output_type":"stream","text":["/content/project_data/data/checkpoint\n"],"name":"stdout"}]},{"cell_type":"code","metadata":{"colab":{"base_uri":"https://localhost:8080/"},"id":"p7lqTgvXNwxW","executionInfo":{"status":"ok","timestamp":1606186419279,"user_tz":360,"elapsed":2160592,"user":{"displayName":"Pankaj Rajak","photoUrl":"https://lh3.googleusercontent.com/a-/AOh14GiTXe8B8vdrh4uTyUPqEtb-tRwPltWOZurP8-lxog=s64","userId":"04274727635571437779"}},"outputId":"b4e3a3ba-e4d2-4173-cf66-6f09670ad4da"},"source":["step_num_trj,pre_val_accuracy_trj,post_val_accuracy_trj,step_train_trj,pre_train_accuracy_trj,post_train_accuracy_trj = run_maml(n_way=20, k_shot=1, \n","                                              inner_update_lr=0.04, num_inner_updates=1,logdir=log_path,meta_train_iterations=5000,\n","                                              train_type='memorization',meta_lr=0.0025,meta_batch_size = 16) #data_path=dest_path"],"execution_count":null,"outputs":[{"output_type":"stream","text":["Number of training examples. :  1200\n","Number of validation examples:  100\n","Number of test examples      :  323\n","Training Type:  memorization\n","Iteration 10: pre-inner-loop train accuracy: 0.04375, post-inner-loop test accuracy: 0.06562, meta_reg : 3.999618\n","Iteration 20: pre-inner-loop train accuracy: 0.05938, post-inner-loop test accuracy: 0.05938, meta_reg : 3.999550\n","Iteration 30: pre-inner-loop train accuracy: 0.04688, post-inner-loop test accuracy: 0.05000, meta_reg : 3.998782\n","Iteration 40: pre-inner-loop train accuracy: 0.07812, post-inner-loop test accuracy: 0.07812, meta_reg : 3.998250\n","Iteration 50: pre-inner-loop train accuracy: 0.05937, post-inner-loop test accuracy: 0.05938, meta_reg : 3.998515\n","Meta-validation pre-inner-loop train accuracy: 0.05000, meta-validation post-inner-loop test accuracy: 0.05937\n","Iteration 60: pre-inner-loop train accuracy: 0.05625, post-inner-loop test accuracy: 0.06562, meta_reg : 3.998136\n","Iteration 70: pre-inner-loop train accuracy: 0.05313, post-inner-loop test accuracy: 0.07188, meta_reg : 3.997549\n","Iteration 80: pre-inner-loop train accuracy: 0.02188, post-inner-loop test accuracy: 0.16875, meta_reg : 3.990200\n","Iteration 90: pre-inner-loop train accuracy: 0.07500, post-inner-loop test accuracy: 0.15625, meta_reg : 3.994823\n","Iteration 100: pre-inner-loop train accuracy: 0.03750, post-inner-loop test accuracy: 0.16875, meta_reg : 3.994310\n","Meta-validation pre-inner-loop train accuracy: 0.04063, meta-validation post-inner-loop test accuracy: 0.14062\n","Iteration 110: pre-inner-loop train accuracy: 0.05938, post-inner-loop test accuracy: 0.15938, meta_reg : 3.994613\n","Iteration 120: pre-inner-loop train accuracy: 0.04688, post-inner-loop test accuracy: 0.17188, meta_reg : 3.994486\n","Iteration 130: pre-inner-loop train accuracy: 0.05938, post-inner-loop test accuracy: 0.16250, meta_reg : 3.994444\n","Iteration 140: pre-inner-loop train accuracy: 0.03438, post-inner-loop test accuracy: 0.14375, meta_reg : 3.994799\n","Iteration 150: pre-inner-loop train accuracy: 0.05000, post-inner-loop test accuracy: 0.15313, meta_reg : 3.994792\n","Meta-validation pre-inner-loop train accuracy: 0.05938, meta-validation post-inner-loop test accuracy: 0.13750\n","Iteration 160: pre-inner-loop train accuracy: 0.05313, post-inner-loop test accuracy: 0.15938, meta_reg : 3.994508\n","Iteration 170: pre-inner-loop train accuracy: 0.04688, post-inner-loop test accuracy: 0.16250, meta_reg : 3.994450\n","Iteration 180: pre-inner-loop train accuracy: 0.05313, post-inner-loop test accuracy: 0.16875, meta_reg : 3.994490\n","Iteration 190: pre-inner-loop train accuracy: 0.05313, post-inner-loop test accuracy: 0.15938, meta_reg : 3.994938\n","Iteration 200: pre-inner-loop train accuracy: 0.05938, post-inner-loop test accuracy: 0.15938, meta_reg : 3.995168\n","Meta-validation pre-inner-loop train accuracy: 0.04688, meta-validation post-inner-loop test accuracy: 0.15313\n","Iteration 210: pre-inner-loop train accuracy: 0.06562, post-inner-loop test accuracy: 0.15000, meta_reg : 3.995060\n","Iteration 220: pre-inner-loop train accuracy: 0.04063, post-inner-loop test accuracy: 0.15938, meta_reg : 3.994516\n","Iteration 230: pre-inner-loop train accuracy: 0.04688, post-inner-loop test accuracy: 0.16875, meta_reg : 3.994501\n","Iteration 240: pre-inner-loop train accuracy: 0.07187, post-inner-loop test accuracy: 0.15313, meta_reg : 3.994909\n","Iteration 250: pre-inner-loop train accuracy: 0.05313, post-inner-loop test accuracy: 0.15625, meta_reg : 3.994896\n","Meta-validation pre-inner-loop train accuracy: 0.03438, meta-validation post-inner-loop test accuracy: 0.13438\n","Iteration 260: pre-inner-loop train accuracy: 0.04688, post-inner-loop test accuracy: 0.16563, meta_reg : 3.994863\n","Iteration 270: pre-inner-loop train accuracy: 0.07500, post-inner-loop test accuracy: 0.15625, meta_reg : 3.995082\n","Iteration 280: pre-inner-loop train accuracy: 0.05313, post-inner-loop test accuracy: 0.15937, meta_reg : 3.994878\n","Iteration 290: pre-inner-loop train accuracy: 0.05313, post-inner-loop test accuracy: 0.15625, meta_reg : 3.994714\n","Iteration 300: pre-inner-loop train accuracy: 0.05000, post-inner-loop test accuracy: 0.15938, meta_reg : 3.994611\n","Meta-validation pre-inner-loop train accuracy: 0.06562, meta-validation post-inner-loop test accuracy: 0.15625\n","Iteration 310: pre-inner-loop train accuracy: 0.05313, post-inner-loop test accuracy: 0.17188, meta_reg : 3.994231\n","Iteration 320: pre-inner-loop train accuracy: 0.05625, post-inner-loop test accuracy: 0.16875, meta_reg : 3.994356\n","Iteration 330: pre-inner-loop train accuracy: 0.05937, post-inner-loop test accuracy: 0.15938, meta_reg : 3.994755\n","Iteration 340: pre-inner-loop train accuracy: 0.06875, post-inner-loop test accuracy: 0.16563, meta_reg : 3.994648\n","Iteration 350: pre-inner-loop train accuracy: 0.05938, post-inner-loop test accuracy: 0.16250, meta_reg : 3.995195\n","Meta-validation pre-inner-loop train accuracy: 0.05625, meta-validation post-inner-loop test accuracy: 0.13750\n","Iteration 360: pre-inner-loop train accuracy: 0.05000, post-inner-loop test accuracy: 0.15938, meta_reg : 3.994909\n","Iteration 370: pre-inner-loop train accuracy: 0.05313, post-inner-loop test accuracy: 0.17188, meta_reg : 3.994651\n","Iteration 380: pre-inner-loop train accuracy: 0.08125, post-inner-loop test accuracy: 0.17813, meta_reg : 3.994635\n","Iteration 390: pre-inner-loop train accuracy: 0.05938, post-inner-loop test accuracy: 0.19063, meta_reg : 3.994493\n","Iteration 400: pre-inner-loop train accuracy: 0.03438, post-inner-loop test accuracy: 0.16875, meta_reg : 3.994447\n","Meta-validation pre-inner-loop train accuracy: 0.04688, meta-validation post-inner-loop test accuracy: 0.14375\n","Iteration 410: pre-inner-loop train accuracy: 0.05313, post-inner-loop test accuracy: 0.15625, meta_reg : 3.994922\n","Iteration 420: pre-inner-loop train accuracy: 0.05313, post-inner-loop test accuracy: 0.16563, meta_reg : 3.994867\n","Iteration 430: pre-inner-loop train accuracy: 0.08125, post-inner-loop test accuracy: 0.16250, meta_reg : 3.994440\n","Iteration 440: pre-inner-loop train accuracy: 0.05000, post-inner-loop test accuracy: 0.14687, meta_reg : 3.994642\n","Iteration 450: pre-inner-loop train accuracy: 0.03750, post-inner-loop test accuracy: 0.15000, meta_reg : 3.994352\n","Meta-validation pre-inner-loop train accuracy: 0.03438, meta-validation post-inner-loop test accuracy: 0.14062\n","Iteration 460: pre-inner-loop train accuracy: 0.06562, post-inner-loop test accuracy: 0.17500, meta_reg : 3.994525\n","Iteration 470: pre-inner-loop train accuracy: 0.04688, post-inner-loop test accuracy: 0.15000, meta_reg : 3.995103\n","Iteration 480: pre-inner-loop train accuracy: 0.05625, post-inner-loop test accuracy: 0.15313, meta_reg : 3.994528\n","Iteration 490: pre-inner-loop train accuracy: 0.03750, post-inner-loop test accuracy: 0.17813, meta_reg : 3.994617\n","Iteration 500: pre-inner-loop train accuracy: 0.03750, post-inner-loop test accuracy: 0.16563, meta_reg : 3.994886\n","Meta-validation pre-inner-loop train accuracy: 0.04688, meta-validation post-inner-loop test accuracy: 0.13438\n","Iteration 510: pre-inner-loop train accuracy: 0.03125, post-inner-loop test accuracy: 0.16563, meta_reg : 3.994289\n","Iteration 520: pre-inner-loop train accuracy: 0.03125, post-inner-loop test accuracy: 0.15625, meta_reg : 3.994263\n","Iteration 530: pre-inner-loop train accuracy: 0.06563, post-inner-loop test accuracy: 0.16563, meta_reg : 3.994530\n","Iteration 540: pre-inner-loop train accuracy: 0.05625, post-inner-loop test accuracy: 0.16250, meta_reg : 3.995099\n","Iteration 550: pre-inner-loop train accuracy: 0.04063, post-inner-loop test accuracy: 0.16563, meta_reg : 3.994856\n","Meta-validation pre-inner-loop train accuracy: 0.05000, meta-validation post-inner-loop test accuracy: 0.14375\n","Iteration 560: pre-inner-loop train accuracy: 0.05313, post-inner-loop test accuracy: 0.15938, meta_reg : 3.994951\n","Iteration 570: pre-inner-loop train accuracy: 0.06875, post-inner-loop test accuracy: 0.14688, meta_reg : 3.994975\n","Iteration 580: pre-inner-loop train accuracy: 0.05938, post-inner-loop test accuracy: 0.17500, meta_reg : 3.994362\n","Iteration 590: pre-inner-loop train accuracy: 0.04062, post-inner-loop test accuracy: 0.16250, meta_reg : 3.995029\n","Iteration 600: pre-inner-loop train accuracy: 0.05625, post-inner-loop test accuracy: 0.16250, meta_reg : 3.994500\n","Meta-validation pre-inner-loop train accuracy: 0.05000, meta-validation post-inner-loop test accuracy: 0.12188\n","Iteration 610: pre-inner-loop train accuracy: 0.05313, post-inner-loop test accuracy: 0.16875, meta_reg : 3.995133\n","Iteration 620: pre-inner-loop train accuracy: 0.07500, post-inner-loop test accuracy: 0.15000, meta_reg : 3.994782\n","Iteration 630: pre-inner-loop train accuracy: 0.05313, post-inner-loop test accuracy: 0.15000, meta_reg : 3.995425\n","Iteration 640: pre-inner-loop train accuracy: 0.05313, post-inner-loop test accuracy: 0.16563, meta_reg : 3.995268\n","Iteration 650: pre-inner-loop train accuracy: 0.05625, post-inner-loop test accuracy: 0.17188, meta_reg : 3.995072\n","Meta-validation pre-inner-loop train accuracy: 0.04688, meta-validation post-inner-loop test accuracy: 0.14375\n","Iteration 660: pre-inner-loop train accuracy: 0.05000, post-inner-loop test accuracy: 0.18125, meta_reg : 3.994359\n","Iteration 670: pre-inner-loop train accuracy: 0.06250, post-inner-loop test accuracy: 0.16250, meta_reg : 3.994689\n","Iteration 680: pre-inner-loop train accuracy: 0.06250, post-inner-loop test accuracy: 0.15312, meta_reg : 3.994947\n","Iteration 690: pre-inner-loop train accuracy: 0.04688, post-inner-loop test accuracy: 0.15938, meta_reg : 3.994399\n","Iteration 700: pre-inner-loop train accuracy: 0.04688, post-inner-loop test accuracy: 0.15625, meta_reg : 3.994509\n","Meta-validation pre-inner-loop train accuracy: 0.04688, meta-validation post-inner-loop test accuracy: 0.14375\n","Iteration 710: pre-inner-loop train accuracy: 0.04688, post-inner-loop test accuracy: 0.15938, meta_reg : 3.995080\n","Iteration 720: pre-inner-loop train accuracy: 0.06562, post-inner-loop test accuracy: 0.15000, meta_reg : 3.995256\n","Iteration 730: pre-inner-loop train accuracy: 0.04688, post-inner-loop test accuracy: 0.16250, meta_reg : 3.994961\n","Iteration 740: pre-inner-loop train accuracy: 0.05625, post-inner-loop test accuracy: 0.18438, meta_reg : 3.994613\n","Iteration 750: pre-inner-loop train accuracy: 0.05000, post-inner-loop test accuracy: 0.15625, meta_reg : 3.995053\n","Meta-validation pre-inner-loop train accuracy: 0.04062, meta-validation post-inner-loop test accuracy: 0.16563\n","Iteration 760: pre-inner-loop train accuracy: 0.04375, post-inner-loop test accuracy: 0.14375, meta_reg : 3.994750\n","Iteration 770: pre-inner-loop train accuracy: 0.05000, post-inner-loop test accuracy: 0.15938, meta_reg : 3.994623\n","Iteration 780: pre-inner-loop train accuracy: 0.05625, post-inner-loop test accuracy: 0.20625, meta_reg : 3.993829\n","Iteration 790: pre-inner-loop train accuracy: 0.07500, post-inner-loop test accuracy: 0.18125, meta_reg : 3.994812\n","Iteration 800: pre-inner-loop train accuracy: 0.03438, post-inner-loop test accuracy: 0.15938, meta_reg : 3.994691\n","Meta-validation pre-inner-loop train accuracy: 0.06250, meta-validation post-inner-loop test accuracy: 0.12812\n","Iteration 810: pre-inner-loop train accuracy: 0.02188, post-inner-loop test accuracy: 0.17188, meta_reg : 3.994927\n","Iteration 820: pre-inner-loop train accuracy: 0.05625, post-inner-loop test accuracy: 0.14375, meta_reg : 3.994909\n","Iteration 830: pre-inner-loop train accuracy: 0.05000, post-inner-loop test accuracy: 0.13750, meta_reg : 3.994993\n","Iteration 840: pre-inner-loop train accuracy: 0.05313, post-inner-loop test accuracy: 0.16563, meta_reg : 3.994453\n","Iteration 850: pre-inner-loop train accuracy: 0.06875, post-inner-loop test accuracy: 0.16875, meta_reg : 3.994730\n","Meta-validation pre-inner-loop train accuracy: 0.05000, meta-validation post-inner-loop test accuracy: 0.12500\n","Iteration 860: pre-inner-loop train accuracy: 0.05313, post-inner-loop test accuracy: 0.16563, meta_reg : 3.994869\n","Iteration 870: pre-inner-loop train accuracy: 0.05000, post-inner-loop test accuracy: 0.15625, meta_reg : 3.994617\n","Iteration 880: pre-inner-loop train accuracy: 0.07187, post-inner-loop test accuracy: 0.17188, meta_reg : 3.994493\n","Iteration 890: pre-inner-loop train accuracy: 0.06875, post-inner-loop test accuracy: 0.15313, meta_reg : 3.994745\n","Iteration 900: pre-inner-loop train accuracy: 0.04375, post-inner-loop test accuracy: 0.16563, meta_reg : 3.994832\n","Meta-validation pre-inner-loop train accuracy: 0.04375, meta-validation post-inner-loop test accuracy: 0.13438\n","Iteration 910: pre-inner-loop train accuracy: 0.06875, post-inner-loop test accuracy: 0.15625, meta_reg : 3.995002\n","Iteration 920: pre-inner-loop train accuracy: 0.05000, post-inner-loop test accuracy: 0.15937, meta_reg : 3.995034\n","Iteration 930: pre-inner-loop train accuracy: 0.06875, post-inner-loop test accuracy: 0.20000, meta_reg : 3.994810\n","Iteration 940: pre-inner-loop train accuracy: 0.06875, post-inner-loop test accuracy: 0.18750, meta_reg : 3.994496\n","Iteration 950: pre-inner-loop train accuracy: 0.05000, post-inner-loop test accuracy: 0.18125, meta_reg : 3.994296\n","Meta-validation pre-inner-loop train accuracy: 0.04375, meta-validation post-inner-loop test accuracy: 0.13438\n","Iteration 960: pre-inner-loop train accuracy: 0.04375, post-inner-loop test accuracy: 0.19062, meta_reg : 3.993491\n","Iteration 970: pre-inner-loop train accuracy: 0.07500, post-inner-loop test accuracy: 0.20625, meta_reg : 3.993600\n","Iteration 980: pre-inner-loop train accuracy: 0.03750, post-inner-loop test accuracy: 0.19375, meta_reg : 3.994171\n","Iteration 990: pre-inner-loop train accuracy: 0.04688, post-inner-loop test accuracy: 0.20937, meta_reg : 3.994005\n","Iteration 1000: pre-inner-loop train accuracy: 0.05937, post-inner-loop test accuracy: 0.19688, meta_reg : 3.993773\n","Meta-validation pre-inner-loop train accuracy: 0.03438, meta-validation post-inner-loop test accuracy: 0.14375\n","Iteration 1010: pre-inner-loop train accuracy: 0.05625, post-inner-loop test accuracy: 0.20312, meta_reg : 3.993809\n","Iteration 1020: pre-inner-loop train accuracy: 0.06563, post-inner-loop test accuracy: 0.21563, meta_reg : 3.993877\n","Iteration 1030: pre-inner-loop train accuracy: 0.05625, post-inner-loop test accuracy: 0.21250, meta_reg : 3.993951\n","Iteration 1040: pre-inner-loop train accuracy: 0.05625, post-inner-loop test accuracy: 0.20625, meta_reg : 3.994544\n","Iteration 1050: pre-inner-loop train accuracy: 0.07188, post-inner-loop test accuracy: 0.20625, meta_reg : 3.993893\n","Meta-validation pre-inner-loop train accuracy: 0.04063, meta-validation post-inner-loop test accuracy: 0.16563\n","Iteration 1060: pre-inner-loop train accuracy: 0.06875, post-inner-loop test accuracy: 0.22500, meta_reg : 3.994165\n","Iteration 1070: pre-inner-loop train accuracy: 0.08125, post-inner-loop test accuracy: 0.22187, meta_reg : 3.992857\n","Iteration 1080: pre-inner-loop train accuracy: 0.10625, post-inner-loop test accuracy: 0.26875, meta_reg : 3.993347\n","Iteration 1090: pre-inner-loop train accuracy: 0.05313, post-inner-loop test accuracy: 0.23125, meta_reg : 3.993817\n","Iteration 1100: pre-inner-loop train accuracy: 0.07500, post-inner-loop test accuracy: 0.24062, meta_reg : 3.993061\n","Meta-validation pre-inner-loop train accuracy: 0.04688, meta-validation post-inner-loop test accuracy: 0.18438\n","Iteration 1110: pre-inner-loop train accuracy: 0.06562, post-inner-loop test accuracy: 0.23438, meta_reg : 3.994051\n","Iteration 1120: pre-inner-loop train accuracy: 0.08750, post-inner-loop test accuracy: 0.23125, meta_reg : 3.992893\n","Iteration 1130: pre-inner-loop train accuracy: 0.07188, post-inner-loop test accuracy: 0.23438, meta_reg : 3.992843\n","Iteration 1140: pre-inner-loop train accuracy: 0.08437, post-inner-loop test accuracy: 0.28438, meta_reg : 3.992654\n","Iteration 1150: pre-inner-loop train accuracy: 0.07187, post-inner-loop test accuracy: 0.25313, meta_reg : 3.992885\n","Meta-validation pre-inner-loop train accuracy: 0.05313, meta-validation post-inner-loop test accuracy: 0.15938\n","Iteration 1160: pre-inner-loop train accuracy: 0.07500, post-inner-loop test accuracy: 0.20937, meta_reg : 3.993083\n","Iteration 1170: pre-inner-loop train accuracy: 0.10625, post-inner-loop test accuracy: 0.24063, meta_reg : 3.993381\n","Iteration 1180: pre-inner-loop train accuracy: 0.08437, post-inner-loop test accuracy: 0.26250, meta_reg : 3.992689\n","Iteration 1190: pre-inner-loop train accuracy: 0.05625, post-inner-loop test accuracy: 0.24063, meta_reg : 3.992600\n","Iteration 1200: pre-inner-loop train accuracy: 0.07500, post-inner-loop test accuracy: 0.25625, meta_reg : 3.993070\n","Meta-validation pre-inner-loop train accuracy: 0.04688, meta-validation post-inner-loop test accuracy: 0.17812\n","Iteration 1210: pre-inner-loop train accuracy: 0.10000, post-inner-loop test accuracy: 0.27187, meta_reg : 3.992167\n","Iteration 1220: pre-inner-loop train accuracy: 0.06563, post-inner-loop test accuracy: 0.26250, meta_reg : 3.992998\n","Iteration 1230: pre-inner-loop train accuracy: 0.06875, post-inner-loop test accuracy: 0.24062, meta_reg : 3.992927\n","Iteration 1240: pre-inner-loop train accuracy: 0.08750, post-inner-loop test accuracy: 0.26562, meta_reg : 3.992534\n","Iteration 1250: pre-inner-loop train accuracy: 0.10625, post-inner-loop test accuracy: 0.26250, meta_reg : 3.993426\n","Meta-validation pre-inner-loop train accuracy: 0.03750, meta-validation post-inner-loop test accuracy: 0.17500\n","Iteration 1260: pre-inner-loop train accuracy: 0.09063, post-inner-loop test accuracy: 0.25000, meta_reg : 3.993242\n","Iteration 1270: pre-inner-loop train accuracy: 0.09687, post-inner-loop test accuracy: 0.29062, meta_reg : 3.992597\n","Iteration 1280: pre-inner-loop train accuracy: 0.12188, post-inner-loop test accuracy: 0.26250, meta_reg : 3.993305\n","Iteration 1290: pre-inner-loop train accuracy: 0.09063, post-inner-loop test accuracy: 0.24688, meta_reg : 3.993148\n","Iteration 1300: pre-inner-loop train accuracy: 0.11250, post-inner-loop test accuracy: 0.25000, meta_reg : 3.992660\n","Meta-validation pre-inner-loop train accuracy: 0.05938, meta-validation post-inner-loop test accuracy: 0.16875\n","Iteration 1310: pre-inner-loop train accuracy: 0.08750, post-inner-loop test accuracy: 0.27187, meta_reg : 3.992533\n","Iteration 1320: pre-inner-loop train accuracy: 0.11250, post-inner-loop test accuracy: 0.24375, meta_reg : 3.993093\n","Iteration 1330: pre-inner-loop train accuracy: 0.10625, post-inner-loop test accuracy: 0.29688, meta_reg : 3.992692\n","Iteration 1340: pre-inner-loop train accuracy: 0.10625, post-inner-loop test accuracy: 0.28438, meta_reg : 3.992419\n","Iteration 1350: pre-inner-loop train accuracy: 0.15313, post-inner-loop test accuracy: 0.25625, meta_reg : 3.992643\n","Meta-validation pre-inner-loop train accuracy: 0.05625, meta-validation post-inner-loop test accuracy: 0.15000\n","Iteration 1360: pre-inner-loop train accuracy: 0.10000, post-inner-loop test accuracy: 0.25000, meta_reg : 3.992143\n","Iteration 1370: pre-inner-loop train accuracy: 0.10313, post-inner-loop test accuracy: 0.26562, meta_reg : 3.993314\n","Iteration 1380: pre-inner-loop train accuracy: 0.10313, post-inner-loop test accuracy: 0.28125, meta_reg : 3.992353\n","Iteration 1390: pre-inner-loop train accuracy: 0.11250, post-inner-loop test accuracy: 0.27188, meta_reg : 3.992887\n","Iteration 1400: pre-inner-loop train accuracy: 0.12500, post-inner-loop test accuracy: 0.28125, meta_reg : 3.992167\n","Meta-validation pre-inner-loop train accuracy: 0.07500, meta-validation post-inner-loop test accuracy: 0.17813\n","Iteration 1410: pre-inner-loop train accuracy: 0.11563, post-inner-loop test accuracy: 0.26562, meta_reg : 3.992936\n","Iteration 1420: pre-inner-loop train accuracy: 0.10938, post-inner-loop test accuracy: 0.28125, meta_reg : 3.992914\n","Iteration 1430: pre-inner-loop train accuracy: 0.11875, post-inner-loop test accuracy: 0.30000, meta_reg : 3.993257\n","Iteration 1440: pre-inner-loop train accuracy: 0.10000, post-inner-loop test accuracy: 0.31250, meta_reg : 3.992109\n","Iteration 1450: pre-inner-loop train accuracy: 0.08125, post-inner-loop test accuracy: 0.29688, meta_reg : 3.992842\n","Meta-validation pre-inner-loop train accuracy: 0.03125, meta-validation post-inner-loop test accuracy: 0.17500\n","Iteration 1460: pre-inner-loop train accuracy: 0.13125, post-inner-loop test accuracy: 0.22813, meta_reg : 3.992862\n","Iteration 1470: pre-inner-loop train accuracy: 0.11563, post-inner-loop test accuracy: 0.27813, meta_reg : 3.992020\n","Iteration 1480: pre-inner-loop train accuracy: 0.14062, post-inner-loop test accuracy: 0.31875, meta_reg : 3.992396\n","Iteration 1490: pre-inner-loop train accuracy: 0.11250, post-inner-loop test accuracy: 0.27500, meta_reg : 3.992719\n","Iteration 1500: pre-inner-loop train accuracy: 0.12500, post-inner-loop test accuracy: 0.31563, meta_reg : 3.992377\n","Meta-validation pre-inner-loop train accuracy: 0.05313, meta-validation post-inner-loop test accuracy: 0.19375\n","Iteration 1510: pre-inner-loop train accuracy: 0.10625, post-inner-loop test accuracy: 0.33750, meta_reg : 3.992166\n","Iteration 1520: pre-inner-loop train accuracy: 0.09063, post-inner-loop test accuracy: 0.29375, meta_reg : 3.991752\n","Iteration 1530: pre-inner-loop train accuracy: 0.11563, post-inner-loop test accuracy: 0.32187, meta_reg : 3.992558\n","Iteration 1540: pre-inner-loop train accuracy: 0.12500, post-inner-loop test accuracy: 0.29688, meta_reg : 3.992542\n","Iteration 1550: pre-inner-loop train accuracy: 0.11562, post-inner-loop test accuracy: 0.29688, meta_reg : 3.992237\n","Meta-validation pre-inner-loop train accuracy: 0.05313, meta-validation post-inner-loop test accuracy: 0.17188\n","Iteration 1560: pre-inner-loop train accuracy: 0.12813, post-inner-loop test accuracy: 0.35313, meta_reg : 3.992213\n","Iteration 1570: pre-inner-loop train accuracy: 0.15313, post-inner-loop test accuracy: 0.33125, meta_reg : 3.993105\n","Iteration 1580: pre-inner-loop train accuracy: 0.10938, post-inner-loop test accuracy: 0.30937, meta_reg : 3.992175\n","Iteration 1590: pre-inner-loop train accuracy: 0.12812, post-inner-loop test accuracy: 0.30625, meta_reg : 3.992570\n","Iteration 1600: pre-inner-loop train accuracy: 0.10313, post-inner-loop test accuracy: 0.33437, meta_reg : 3.992491\n","Meta-validation pre-inner-loop train accuracy: 0.03438, meta-validation post-inner-loop test accuracy: 0.18750\n","Iteration 1610: pre-inner-loop train accuracy: 0.15000, post-inner-loop test accuracy: 0.30312, meta_reg : 3.992129\n","Iteration 1620: pre-inner-loop train accuracy: 0.11563, post-inner-loop test accuracy: 0.31875, meta_reg : 3.992266\n","Iteration 1630: pre-inner-loop train accuracy: 0.14062, post-inner-loop test accuracy: 0.33437, meta_reg : 3.992280\n","Iteration 1640: pre-inner-loop train accuracy: 0.14375, post-inner-loop test accuracy: 0.30312, meta_reg : 3.992181\n","Iteration 1650: pre-inner-loop train accuracy: 0.14375, post-inner-loop test accuracy: 0.30938, meta_reg : 3.992471\n","Meta-validation pre-inner-loop train accuracy: 0.05313, meta-validation post-inner-loop test accuracy: 0.19375\n","Iteration 1660: pre-inner-loop train accuracy: 0.11562, post-inner-loop test accuracy: 0.34375, meta_reg : 3.992252\n","Iteration 1670: pre-inner-loop train accuracy: 0.11250, post-inner-loop test accuracy: 0.33750, meta_reg : 3.991948\n","Iteration 1680: pre-inner-loop train accuracy: 0.16250, post-inner-loop test accuracy: 0.34687, meta_reg : 3.991421\n","Iteration 1690: pre-inner-loop train accuracy: 0.12500, post-inner-loop test accuracy: 0.31875, meta_reg : 3.992317\n","Iteration 1700: pre-inner-loop train accuracy: 0.15625, post-inner-loop test accuracy: 0.34688, meta_reg : 3.992298\n","Meta-validation pre-inner-loop train accuracy: 0.05625, meta-validation post-inner-loop test accuracy: 0.15000\n","Iteration 1710: pre-inner-loop train accuracy: 0.12812, post-inner-loop test accuracy: 0.35938, meta_reg : 3.991637\n","Iteration 1720: pre-inner-loop train accuracy: 0.16875, post-inner-loop test accuracy: 0.35313, meta_reg : 3.991753\n","Iteration 1730: pre-inner-loop train accuracy: 0.19062, post-inner-loop test accuracy: 0.35938, meta_reg : 3.992084\n","Iteration 1740: pre-inner-loop train accuracy: 0.11250, post-inner-loop test accuracy: 0.35625, meta_reg : 3.992126\n","Iteration 1750: pre-inner-loop train accuracy: 0.13438, post-inner-loop test accuracy: 0.35625, meta_reg : 3.991637\n","Meta-validation pre-inner-loop train accuracy: 0.03438, meta-validation post-inner-loop test accuracy: 0.18437\n","Iteration 1760: pre-inner-loop train accuracy: 0.15312, post-inner-loop test accuracy: 0.33125, meta_reg : 3.991901\n","Iteration 1770: pre-inner-loop train accuracy: 0.16563, post-inner-loop test accuracy: 0.36562, meta_reg : 3.992085\n","Iteration 1780: pre-inner-loop train accuracy: 0.15938, post-inner-loop test accuracy: 0.39375, meta_reg : 3.991844\n","Iteration 1790: pre-inner-loop train accuracy: 0.15937, post-inner-loop test accuracy: 0.35313, meta_reg : 3.992254\n","Iteration 1800: pre-inner-loop train accuracy: 0.16250, post-inner-loop test accuracy: 0.37500, meta_reg : 3.991965\n","Meta-validation pre-inner-loop train accuracy: 0.03438, meta-validation post-inner-loop test accuracy: 0.16563\n","Iteration 1810: pre-inner-loop train accuracy: 0.17188, post-inner-loop test accuracy: 0.35938, meta_reg : 3.991399\n","Iteration 1820: pre-inner-loop train accuracy: 0.18438, post-inner-loop test accuracy: 0.37187, meta_reg : 3.992024\n","Iteration 1830: pre-inner-loop train accuracy: 0.18750, post-inner-loop test accuracy: 0.36563, meta_reg : 3.992118\n","Iteration 1840: pre-inner-loop train accuracy: 0.16875, post-inner-loop test accuracy: 0.35938, meta_reg : 3.991842\n","Iteration 1850: pre-inner-loop train accuracy: 0.21250, post-inner-loop test accuracy: 0.42813, meta_reg : 3.991818\n","Meta-validation pre-inner-loop train accuracy: 0.05938, meta-validation post-inner-loop test accuracy: 0.20000\n","Iteration 1860: pre-inner-loop train accuracy: 0.19375, post-inner-loop test accuracy: 0.33750, meta_reg : 3.991881\n","Iteration 1870: pre-inner-loop train accuracy: 0.14375, post-inner-loop test accuracy: 0.36562, meta_reg : 3.991451\n","Iteration 1880: pre-inner-loop train accuracy: 0.14375, post-inner-loop test accuracy: 0.33750, meta_reg : 3.991689\n","Iteration 1890: pre-inner-loop train accuracy: 0.19375, post-inner-loop test accuracy: 0.40938, meta_reg : 3.992026\n","Iteration 1900: pre-inner-loop train accuracy: 0.18437, post-inner-loop test accuracy: 0.40312, meta_reg : 3.991481\n","Meta-validation pre-inner-loop train accuracy: 0.04375, meta-validation post-inner-loop test accuracy: 0.23438\n","Iteration 1910: pre-inner-loop train accuracy: 0.17188, post-inner-loop test accuracy: 0.37500, meta_reg : 3.992503\n","Iteration 1920: pre-inner-loop train accuracy: 0.16250, post-inner-loop test accuracy: 0.37187, meta_reg : 3.991875\n","Iteration 1930: pre-inner-loop train accuracy: 0.14687, post-inner-loop test accuracy: 0.37813, meta_reg : 3.991348\n","Iteration 1940: pre-inner-loop train accuracy: 0.15938, post-inner-loop test accuracy: 0.35000, meta_reg : 3.991171\n","Iteration 1950: pre-inner-loop train accuracy: 0.16563, post-inner-loop test accuracy: 0.41562, meta_reg : 3.991809\n","Meta-validation pre-inner-loop train accuracy: 0.04688, meta-validation post-inner-loop test accuracy: 0.20312\n","Iteration 1960: pre-inner-loop train accuracy: 0.19062, post-inner-loop test accuracy: 0.39062, meta_reg : 3.990970\n","Iteration 1970: pre-inner-loop train accuracy: 0.18125, post-inner-loop test accuracy: 0.42500, meta_reg : 3.991421\n","Iteration 1980: pre-inner-loop train accuracy: 0.17500, post-inner-loop test accuracy: 0.36875, meta_reg : 3.991283\n","Iteration 1990: pre-inner-loop train accuracy: 0.17812, post-inner-loop test accuracy: 0.37500, meta_reg : 3.991578\n","Iteration 2000: pre-inner-loop train accuracy: 0.19688, post-inner-loop test accuracy: 0.40000, meta_reg : 3.992091\n","Meta-validation pre-inner-loop train accuracy: 0.07187, meta-validation post-inner-loop test accuracy: 0.23438\n","Iteration 2010: pre-inner-loop train accuracy: 0.16250, post-inner-loop test accuracy: 0.42812, meta_reg : 3.991635\n","Iteration 2020: pre-inner-loop train accuracy: 0.15938, post-inner-loop test accuracy: 0.40000, meta_reg : 3.991663\n","Iteration 2030: pre-inner-loop train accuracy: 0.18438, post-inner-loop test accuracy: 0.40000, meta_reg : 3.991661\n","Iteration 2040: pre-inner-loop train accuracy: 0.17500, post-inner-loop test accuracy: 0.42188, meta_reg : 3.991830\n","Iteration 2050: pre-inner-loop train accuracy: 0.21563, post-inner-loop test accuracy: 0.38125, meta_reg : 3.990966\n","Meta-validation pre-inner-loop train accuracy: 0.05000, meta-validation post-inner-loop test accuracy: 0.22188\n","Iteration 2060: pre-inner-loop train accuracy: 0.16563, post-inner-loop test accuracy: 0.45000, meta_reg : 3.990483\n","Iteration 2070: pre-inner-loop train accuracy: 0.22188, post-inner-loop test accuracy: 0.44062, meta_reg : 3.991825\n","Iteration 2080: pre-inner-loop train accuracy: 0.18438, post-inner-loop test accuracy: 0.44688, meta_reg : 3.990803\n","Iteration 2090: pre-inner-loop train accuracy: 0.14062, post-inner-loop test accuracy: 0.43437, meta_reg : 3.991301\n","Iteration 2100: pre-inner-loop train accuracy: 0.17813, post-inner-loop test accuracy: 0.42812, meta_reg : 3.991644\n","Meta-validation pre-inner-loop train accuracy: 0.05625, meta-validation post-inner-loop test accuracy: 0.20312\n","Iteration 2110: pre-inner-loop train accuracy: 0.22813, post-inner-loop test accuracy: 0.45000, meta_reg : 3.991507\n","Iteration 2120: pre-inner-loop train accuracy: 0.20312, post-inner-loop test accuracy: 0.42500, meta_reg : 3.991642\n","Iteration 2130: pre-inner-loop train accuracy: 0.16563, post-inner-loop test accuracy: 0.43750, meta_reg : 3.991542\n","Iteration 2140: pre-inner-loop train accuracy: 0.18750, post-inner-loop test accuracy: 0.40938, meta_reg : 3.990822\n","Iteration 2150: pre-inner-loop train accuracy: 0.23125, post-inner-loop test accuracy: 0.43750, meta_reg : 3.991495\n","Meta-validation pre-inner-loop train accuracy: 0.04063, meta-validation post-inner-loop test accuracy: 0.19375\n","Iteration 2160: pre-inner-loop train accuracy: 0.22813, post-inner-loop test accuracy: 0.45000, meta_reg : 3.991403\n","Iteration 2170: pre-inner-loop train accuracy: 0.17500, post-inner-loop test accuracy: 0.44063, meta_reg : 3.990977\n","Iteration 2180: pre-inner-loop train accuracy: 0.21563, post-inner-loop test accuracy: 0.41875, meta_reg : 3.990879\n","Iteration 2190: pre-inner-loop train accuracy: 0.15000, post-inner-loop test accuracy: 0.45312, meta_reg : 3.990887\n","Iteration 2200: pre-inner-loop train accuracy: 0.16875, post-inner-loop test accuracy: 0.42813, meta_reg : 3.990859\n","Meta-validation pre-inner-loop train accuracy: 0.07500, meta-validation post-inner-loop test accuracy: 0.21563\n","Iteration 2210: pre-inner-loop train accuracy: 0.20312, post-inner-loop test accuracy: 0.49375, meta_reg : 3.991478\n","Iteration 2220: pre-inner-loop train accuracy: 0.20000, post-inner-loop test accuracy: 0.48750, meta_reg : 3.991338\n","Iteration 2230: pre-inner-loop train accuracy: 0.26562, post-inner-loop test accuracy: 0.43750, meta_reg : 3.991754\n","Iteration 2240: pre-inner-loop train accuracy: 0.23750, post-inner-loop test accuracy: 0.46562, meta_reg : 3.991145\n","Iteration 2250: pre-inner-loop train accuracy: 0.18125, post-inner-loop test accuracy: 0.40000, meta_reg : 3.990936\n","Meta-validation pre-inner-loop train accuracy: 0.06250, meta-validation post-inner-loop test accuracy: 0.21875\n","Iteration 2260: pre-inner-loop train accuracy: 0.20625, post-inner-loop test accuracy: 0.45312, meta_reg : 3.990425\n","Iteration 2270: pre-inner-loop train accuracy: 0.24062, post-inner-loop test accuracy: 0.47500, meta_reg : 3.991048\n","Iteration 2280: pre-inner-loop train accuracy: 0.19375, post-inner-loop test accuracy: 0.45938, meta_reg : 3.990731\n","Iteration 2290: pre-inner-loop train accuracy: 0.24375, post-inner-loop test accuracy: 0.43437, meta_reg : 3.991323\n","Iteration 2300: pre-inner-loop train accuracy: 0.18437, post-inner-loop test accuracy: 0.41250, meta_reg : 3.990903\n","Meta-validation pre-inner-loop train accuracy: 0.06562, meta-validation post-inner-loop test accuracy: 0.21250\n","Iteration 2310: pre-inner-loop train accuracy: 0.22813, post-inner-loop test accuracy: 0.46875, meta_reg : 3.990664\n","Iteration 2320: pre-inner-loop train accuracy: 0.23750, post-inner-loop test accuracy: 0.46250, meta_reg : 3.990743\n","Iteration 2330: pre-inner-loop train accuracy: 0.25625, post-inner-loop test accuracy: 0.45312, meta_reg : 3.991435\n","Iteration 2340: pre-inner-loop train accuracy: 0.20937, post-inner-loop test accuracy: 0.42812, meta_reg : 3.991044\n","Iteration 2350: pre-inner-loop train accuracy: 0.25000, post-inner-loop test accuracy: 0.46875, meta_reg : 3.991370\n","Meta-validation pre-inner-loop train accuracy: 0.07812, meta-validation post-inner-loop test accuracy: 0.20312\n","Iteration 2360: pre-inner-loop train accuracy: 0.20625, post-inner-loop test accuracy: 0.50625, meta_reg : 3.991080\n","Iteration 2370: pre-inner-loop train accuracy: 0.20000, post-inner-loop test accuracy: 0.46250, meta_reg : 3.989813\n","Iteration 2380: pre-inner-loop train accuracy: 0.19063, post-inner-loop test accuracy: 0.51875, meta_reg : 3.991212\n","Iteration 2390: pre-inner-loop train accuracy: 0.22187, post-inner-loop test accuracy: 0.47500, meta_reg : 3.990928\n","Iteration 2400: pre-inner-loop train accuracy: 0.27500, post-inner-loop test accuracy: 0.47813, meta_reg : 3.990513\n","Meta-validation pre-inner-loop train accuracy: 0.05000, meta-validation post-inner-loop test accuracy: 0.21250\n","Iteration 2410: pre-inner-loop train accuracy: 0.21250, post-inner-loop test accuracy: 0.48125, meta_reg : 3.990773\n","Iteration 2420: pre-inner-loop train accuracy: 0.18438, post-inner-loop test accuracy: 0.45625, meta_reg : 3.990177\n","Iteration 2430: pre-inner-loop train accuracy: 0.24062, post-inner-loop test accuracy: 0.45312, meta_reg : 3.990480\n","Iteration 2440: pre-inner-loop train accuracy: 0.21250, post-inner-loop test accuracy: 0.45000, meta_reg : 3.991185\n","Iteration 2450: pre-inner-loop train accuracy: 0.24063, post-inner-loop test accuracy: 0.50937, meta_reg : 3.990233\n","Meta-validation pre-inner-loop train accuracy: 0.07812, meta-validation post-inner-loop test accuracy: 0.19375\n","Iteration 2460: pre-inner-loop train accuracy: 0.25313, post-inner-loop test accuracy: 0.47813, meta_reg : 3.991061\n","Iteration 2470: pre-inner-loop train accuracy: 0.21563, post-inner-loop test accuracy: 0.46562, meta_reg : 3.990790\n","Iteration 2480: pre-inner-loop train accuracy: 0.20312, post-inner-loop test accuracy: 0.46562, meta_reg : 3.990884\n","Iteration 2490: pre-inner-loop train accuracy: 0.22500, post-inner-loop test accuracy: 0.49062, meta_reg : 3.990390\n","Iteration 2500: pre-inner-loop train accuracy: 0.22812, post-inner-loop test accuracy: 0.46562, meta_reg : 3.990702\n","Meta-validation pre-inner-loop train accuracy: 0.06875, meta-validation post-inner-loop test accuracy: 0.18125\n","Iteration 2510: pre-inner-loop train accuracy: 0.25000, post-inner-loop test accuracy: 0.50000, meta_reg : 3.990690\n","Iteration 2520: pre-inner-loop train accuracy: 0.24375, post-inner-loop test accuracy: 0.47500, meta_reg : 3.990156\n","Iteration 2530: pre-inner-loop train accuracy: 0.24688, post-inner-loop test accuracy: 0.45938, meta_reg : 3.990278\n","Iteration 2540: pre-inner-loop train accuracy: 0.23438, post-inner-loop test accuracy: 0.56563, meta_reg : 3.990993\n","Iteration 2550: pre-inner-loop train accuracy: 0.22500, post-inner-loop test accuracy: 0.46562, meta_reg : 3.990028\n","Meta-validation pre-inner-loop train accuracy: 0.04375, meta-validation post-inner-loop test accuracy: 0.21563\n","Iteration 2560: pre-inner-loop train accuracy: 0.25313, post-inner-loop test accuracy: 0.52813, meta_reg : 3.990351\n","Iteration 2570: pre-inner-loop train accuracy: 0.22500, post-inner-loop test accuracy: 0.42500, meta_reg : 3.990760\n","Iteration 2580: pre-inner-loop train accuracy: 0.25625, post-inner-loop test accuracy: 0.50937, meta_reg : 3.990441\n","Iteration 2590: pre-inner-loop train accuracy: 0.21563, post-inner-loop test accuracy: 0.50937, meta_reg : 3.990425\n","Iteration 2600: pre-inner-loop train accuracy: 0.21875, post-inner-loop test accuracy: 0.48750, meta_reg : 3.990164\n","Meta-validation pre-inner-loop train accuracy: 0.07812, meta-validation post-inner-loop test accuracy: 0.20625\n","Iteration 2610: pre-inner-loop train accuracy: 0.25938, post-inner-loop test accuracy: 0.49687, meta_reg : 3.990204\n","Iteration 2620: pre-inner-loop train accuracy: 0.22500, post-inner-loop test accuracy: 0.46562, meta_reg : 3.990376\n","Iteration 2630: pre-inner-loop train accuracy: 0.25625, post-inner-loop test accuracy: 0.52813, meta_reg : 3.990774\n","Iteration 2640: pre-inner-loop train accuracy: 0.24062, post-inner-loop test accuracy: 0.50313, meta_reg : 3.990293\n","Iteration 2650: pre-inner-loop train accuracy: 0.24375, post-inner-loop test accuracy: 0.50000, meta_reg : 3.990555\n","Meta-validation pre-inner-loop train accuracy: 0.09375, meta-validation post-inner-loop test accuracy: 0.20312\n","Iteration 2660: pre-inner-loop train accuracy: 0.22500, post-inner-loop test accuracy: 0.53125, meta_reg : 3.989913\n","Iteration 2670: pre-inner-loop train accuracy: 0.21250, post-inner-loop test accuracy: 0.49063, meta_reg : 3.990294\n","Iteration 2680: pre-inner-loop train accuracy: 0.25625, post-inner-loop test accuracy: 0.51250, meta_reg : 3.990066\n","Iteration 2690: pre-inner-loop train accuracy: 0.18750, post-inner-loop test accuracy: 0.52812, meta_reg : 3.990230\n","Iteration 2700: pre-inner-loop train accuracy: 0.25313, post-inner-loop test accuracy: 0.47500, meta_reg : 3.990525\n","Meta-validation pre-inner-loop train accuracy: 0.07188, meta-validation post-inner-loop test accuracy: 0.19062\n","Iteration 2710: pre-inner-loop train accuracy: 0.24062, post-inner-loop test accuracy: 0.49062, meta_reg : 3.990420\n","Iteration 2720: pre-inner-loop train accuracy: 0.24375, post-inner-loop test accuracy: 0.54375, meta_reg : 3.990475\n","Iteration 2730: pre-inner-loop train accuracy: 0.26250, post-inner-loop test accuracy: 0.49062, meta_reg : 3.990519\n","Iteration 2740: pre-inner-loop train accuracy: 0.28438, post-inner-loop test accuracy: 0.52500, meta_reg : 3.990759\n","Iteration 2750: pre-inner-loop train accuracy: 0.25625, post-inner-loop test accuracy: 0.52812, meta_reg : 3.989809\n","Meta-validation pre-inner-loop train accuracy: 0.07188, meta-validation post-inner-loop test accuracy: 0.20937\n","Iteration 2760: pre-inner-loop train accuracy: 0.24062, post-inner-loop test accuracy: 0.53438, meta_reg : 3.990420\n","Iteration 2770: pre-inner-loop train accuracy: 0.22188, post-inner-loop test accuracy: 0.55313, meta_reg : 3.990394\n","Iteration 2780: pre-inner-loop train accuracy: 0.23125, post-inner-loop test accuracy: 0.51562, meta_reg : 3.989974\n","Iteration 2790: pre-inner-loop train accuracy: 0.29062, post-inner-loop test accuracy: 0.54688, meta_reg : 3.990111\n","Iteration 2800: pre-inner-loop train accuracy: 0.20312, post-inner-loop test accuracy: 0.52188, meta_reg : 3.990176\n","Meta-validation pre-inner-loop train accuracy: 0.07500, meta-validation post-inner-loop test accuracy: 0.23750\n","Iteration 2810: pre-inner-loop train accuracy: 0.28750, post-inner-loop test accuracy: 0.54375, meta_reg : 3.989976\n","Iteration 2820: pre-inner-loop train accuracy: 0.31250, post-inner-loop test accuracy: 0.52812, meta_reg : 3.990329\n","Iteration 2830: pre-inner-loop train accuracy: 0.26562, post-inner-loop test accuracy: 0.54375, meta_reg : 3.989592\n","Iteration 2840: pre-inner-loop train accuracy: 0.22813, post-inner-loop test accuracy: 0.52188, meta_reg : 3.990109\n","Iteration 2850: pre-inner-loop train accuracy: 0.27500, post-inner-loop test accuracy: 0.53125, meta_reg : 3.990464\n","Meta-validation pre-inner-loop train accuracy: 0.05000, meta-validation post-inner-loop test accuracy: 0.21250\n","Iteration 2860: pre-inner-loop train accuracy: 0.27187, post-inner-loop test accuracy: 0.51875, meta_reg : 3.989980\n","Iteration 2870: pre-inner-loop train accuracy: 0.27188, post-inner-loop test accuracy: 0.56250, meta_reg : 3.990195\n","Iteration 2880: pre-inner-loop train accuracy: 0.25000, post-inner-loop test accuracy: 0.53437, meta_reg : 3.989990\n","Iteration 2890: pre-inner-loop train accuracy: 0.25313, post-inner-loop test accuracy: 0.50625, meta_reg : 3.990179\n","Iteration 2900: pre-inner-loop train accuracy: 0.25625, post-inner-loop test accuracy: 0.52188, meta_reg : 3.989645\n","Meta-validation pre-inner-loop train accuracy: 0.06250, meta-validation post-inner-loop test accuracy: 0.25625\n","Iteration 2910: pre-inner-loop train accuracy: 0.22187, post-inner-loop test accuracy: 0.54375, meta_reg : 3.990254\n","Iteration 2920: pre-inner-loop train accuracy: 0.29063, post-inner-loop test accuracy: 0.56875, meta_reg : 3.990370\n","Iteration 2930: pre-inner-loop train accuracy: 0.28438, post-inner-loop test accuracy: 0.51875, meta_reg : 3.990236\n","Iteration 2940: pre-inner-loop train accuracy: 0.21875, post-inner-loop test accuracy: 0.54688, meta_reg : 3.990459\n","Iteration 2950: pre-inner-loop train accuracy: 0.27812, post-inner-loop test accuracy: 0.50312, meta_reg : 3.990099\n","Meta-validation pre-inner-loop train accuracy: 0.04375, meta-validation post-inner-loop test accuracy: 0.27500\n","Iteration 2960: pre-inner-loop train accuracy: 0.29375, post-inner-loop test accuracy: 0.60313, meta_reg : 3.990570\n","Iteration 2970: pre-inner-loop train accuracy: 0.30937, post-inner-loop test accuracy: 0.55625, meta_reg : 3.990500\n","Iteration 2980: pre-inner-loop train accuracy: 0.24062, post-inner-loop test accuracy: 0.58438, meta_reg : 3.990133\n","Iteration 2990: pre-inner-loop train accuracy: 0.23438, post-inner-loop test accuracy: 0.54062, meta_reg : 3.989141\n","Iteration 3000: pre-inner-loop train accuracy: 0.22813, post-inner-loop test accuracy: 0.53438, meta_reg : 3.990287\n","Meta-validation pre-inner-loop train accuracy: 0.05938, meta-validation post-inner-loop test accuracy: 0.21250\n","Iteration 3010: pre-inner-loop train accuracy: 0.26875, post-inner-loop test accuracy: 0.55938, meta_reg : 3.989516\n","Iteration 3020: pre-inner-loop train accuracy: 0.27500, post-inner-loop test accuracy: 0.51562, meta_reg : 3.990224\n","Iteration 3030: pre-inner-loop train accuracy: 0.24375, post-inner-loop test accuracy: 0.54063, meta_reg : 3.989956\n","Iteration 3040: pre-inner-loop train accuracy: 0.27500, post-inner-loop test accuracy: 0.54688, meta_reg : 3.989819\n","Iteration 3050: pre-inner-loop train accuracy: 0.28750, post-inner-loop test accuracy: 0.55313, meta_reg : 3.990485\n","Meta-validation pre-inner-loop train accuracy: 0.06250, meta-validation post-inner-loop test accuracy: 0.21875\n","Iteration 3060: pre-inner-loop train accuracy: 0.24062, post-inner-loop test accuracy: 0.55313, meta_reg : 3.990046\n","Iteration 3070: pre-inner-loop train accuracy: 0.24062, post-inner-loop test accuracy: 0.57187, meta_reg : 3.990247\n","Iteration 3080: pre-inner-loop train accuracy: 0.25938, post-inner-loop test accuracy: 0.55313, meta_reg : 3.989840\n","Iteration 3090: pre-inner-loop train accuracy: 0.29063, post-inner-loop test accuracy: 0.56875, meta_reg : 3.990518\n","Iteration 3100: pre-inner-loop train accuracy: 0.23438, post-inner-loop test accuracy: 0.54688, meta_reg : 3.990066\n","Meta-validation pre-inner-loop train accuracy: 0.06875, meta-validation post-inner-loop test accuracy: 0.24375\n","Iteration 3110: pre-inner-loop train accuracy: 0.27500, post-inner-loop test accuracy: 0.54062, meta_reg : 3.990372\n","Iteration 3120: pre-inner-loop train accuracy: 0.25313, post-inner-loop test accuracy: 0.56562, meta_reg : 3.989994\n","Iteration 3130: pre-inner-loop train accuracy: 0.22500, post-inner-loop test accuracy: 0.57813, meta_reg : 3.990125\n","Iteration 3140: pre-inner-loop train accuracy: 0.30937, post-inner-loop test accuracy: 0.56563, meta_reg : 3.989872\n","Iteration 3150: pre-inner-loop train accuracy: 0.29063, post-inner-loop test accuracy: 0.62187, meta_reg : 3.990063\n","Meta-validation pre-inner-loop train accuracy: 0.03750, meta-validation post-inner-loop test accuracy: 0.21563\n","Iteration 3160: pre-inner-loop train accuracy: 0.30937, post-inner-loop test accuracy: 0.57500, meta_reg : 3.990066\n","Iteration 3170: pre-inner-loop train accuracy: 0.27812, post-inner-loop test accuracy: 0.53438, meta_reg : 3.990464\n","Iteration 3180: pre-inner-loop train accuracy: 0.26250, post-inner-loop test accuracy: 0.58125, meta_reg : 3.989911\n","Iteration 3190: pre-inner-loop train accuracy: 0.34375, post-inner-loop test accuracy: 0.56563, meta_reg : 3.989692\n","Iteration 3200: pre-inner-loop train accuracy: 0.26250, post-inner-loop test accuracy: 0.55625, meta_reg : 3.989401\n","Meta-validation pre-inner-loop train accuracy: 0.05000, meta-validation post-inner-loop test accuracy: 0.20312\n","Iteration 3210: pre-inner-loop train accuracy: 0.32500, post-inner-loop test accuracy: 0.58750, meta_reg : 3.989685\n","Iteration 3220: pre-inner-loop train accuracy: 0.28125, post-inner-loop test accuracy: 0.56563, meta_reg : 3.989805\n","Iteration 3230: pre-inner-loop train accuracy: 0.26250, post-inner-loop test accuracy: 0.57500, meta_reg : 3.989881\n","Iteration 3240: pre-inner-loop train accuracy: 0.25313, post-inner-loop test accuracy: 0.50313, meta_reg : 3.988986\n","Iteration 3250: pre-inner-loop train accuracy: 0.22188, post-inner-loop test accuracy: 0.59375, meta_reg : 3.989996\n","Meta-validation pre-inner-loop train accuracy: 0.05000, meta-validation post-inner-loop test accuracy: 0.24062\n","Iteration 3260: pre-inner-loop train accuracy: 0.23750, post-inner-loop test accuracy: 0.55938, meta_reg : 3.989450\n","Iteration 3270: pre-inner-loop train accuracy: 0.24688, post-inner-loop test accuracy: 0.53750, meta_reg : 3.990612\n","Iteration 3280: pre-inner-loop train accuracy: 0.24688, post-inner-loop test accuracy: 0.53125, meta_reg : 3.989403\n","Iteration 3290: pre-inner-loop train accuracy: 0.25938, post-inner-loop test accuracy: 0.55000, meta_reg : 3.990161\n","Iteration 3300: pre-inner-loop train accuracy: 0.29375, post-inner-loop test accuracy: 0.61875, meta_reg : 3.989661\n","Meta-validation pre-inner-loop train accuracy: 0.02812, meta-validation post-inner-loop test accuracy: 0.22188\n","Iteration 3310: pre-inner-loop train accuracy: 0.30937, post-inner-loop test accuracy: 0.56875, meta_reg : 3.990044\n","Iteration 3320: pre-inner-loop train accuracy: 0.28438, post-inner-loop test accuracy: 0.56563, meta_reg : 3.990139\n","Iteration 3330: pre-inner-loop train accuracy: 0.27500, post-inner-loop test accuracy: 0.58750, meta_reg : 3.988922\n","Iteration 3340: pre-inner-loop train accuracy: 0.31250, post-inner-loop test accuracy: 0.55937, meta_reg : 3.990194\n","Iteration 3350: pre-inner-loop train accuracy: 0.31875, post-inner-loop test accuracy: 0.59375, meta_reg : 3.989915\n","Meta-validation pre-inner-loop train accuracy: 0.05938, meta-validation post-inner-loop test accuracy: 0.23750\n","Iteration 3360: pre-inner-loop train accuracy: 0.27187, post-inner-loop test accuracy: 0.60000, meta_reg : 3.989588\n","Iteration 3370: pre-inner-loop train accuracy: 0.31875, post-inner-loop test accuracy: 0.51250, meta_reg : 3.990313\n","Iteration 3380: pre-inner-loop train accuracy: 0.25625, post-inner-loop test accuracy: 0.58750, meta_reg : 3.989795\n","Iteration 3390: pre-inner-loop train accuracy: 0.33437, post-inner-loop test accuracy: 0.55000, meta_reg : 3.990360\n","Iteration 3400: pre-inner-loop train accuracy: 0.28438, post-inner-loop test accuracy: 0.55313, meta_reg : 3.989629\n","Meta-validation pre-inner-loop train accuracy: 0.06250, meta-validation post-inner-loop test accuracy: 0.24062\n","Iteration 3410: pre-inner-loop train accuracy: 0.30625, post-inner-loop test accuracy: 0.56562, meta_reg : 3.990130\n","Iteration 3420: pre-inner-loop train accuracy: 0.27188, post-inner-loop test accuracy: 0.57812, meta_reg : 3.988880\n","Iteration 3430: pre-inner-loop train accuracy: 0.32187, post-inner-loop test accuracy: 0.56875, meta_reg : 3.989784\n","Iteration 3440: pre-inner-loop train accuracy: 0.26875, post-inner-loop test accuracy: 0.57187, meta_reg : 3.989722\n","Iteration 3450: pre-inner-loop train accuracy: 0.30625, post-inner-loop test accuracy: 0.56250, meta_reg : 3.989444\n","Meta-validation pre-inner-loop train accuracy: 0.05938, meta-validation post-inner-loop test accuracy: 0.23125\n","Iteration 3460: pre-inner-loop train accuracy: 0.30625, post-inner-loop test accuracy: 0.59688, meta_reg : 3.989154\n","Iteration 3470: pre-inner-loop train accuracy: 0.26875, post-inner-loop test accuracy: 0.52812, meta_reg : 3.989571\n","Iteration 3480: pre-inner-loop train accuracy: 0.28125, post-inner-loop test accuracy: 0.60000, meta_reg : 3.989073\n","Iteration 3490: pre-inner-loop train accuracy: 0.30313, post-inner-loop test accuracy: 0.59375, meta_reg : 3.989941\n","Iteration 3500: pre-inner-loop train accuracy: 0.28750, post-inner-loop test accuracy: 0.55938, meta_reg : 3.989895\n","Meta-validation pre-inner-loop train accuracy: 0.09375, meta-validation post-inner-loop test accuracy: 0.26875\n","Iteration 3510: pre-inner-loop train accuracy: 0.30313, post-inner-loop test accuracy: 0.56250, meta_reg : 3.989451\n","Iteration 3520: pre-inner-loop train accuracy: 0.22813, post-inner-loop test accuracy: 0.58438, meta_reg : 3.989872\n","Iteration 3530: pre-inner-loop train accuracy: 0.29062, post-inner-loop test accuracy: 0.58750, meta_reg : 3.990360\n","Iteration 3540: pre-inner-loop train accuracy: 0.29062, post-inner-loop test accuracy: 0.57500, meta_reg : 3.989772\n","Iteration 3550: pre-inner-loop train accuracy: 0.30625, post-inner-loop test accuracy: 0.62812, meta_reg : 3.990276\n","Meta-validation pre-inner-loop train accuracy: 0.05937, meta-validation post-inner-loop test accuracy: 0.24375\n","Iteration 3560: pre-inner-loop train accuracy: 0.29375, post-inner-loop test accuracy: 0.55313, meta_reg : 3.989883\n","Iteration 3570: pre-inner-loop train accuracy: 0.27500, post-inner-loop test accuracy: 0.58125, meta_reg : 3.989805\n","Iteration 3580: pre-inner-loop train accuracy: 0.31250, post-inner-loop test accuracy: 0.59687, meta_reg : 3.989394\n","Iteration 3590: pre-inner-loop train accuracy: 0.31875, post-inner-loop test accuracy: 0.61250, meta_reg : 3.990066\n","Iteration 3600: pre-inner-loop train accuracy: 0.28125, post-inner-loop test accuracy: 0.55625, meta_reg : 3.989261\n","Meta-validation pre-inner-loop train accuracy: 0.05937, meta-validation post-inner-loop test accuracy: 0.24688\n","Iteration 3610: pre-inner-loop train accuracy: 0.32500, post-inner-loop test accuracy: 0.58750, meta_reg : 3.989962\n","Iteration 3620: pre-inner-loop train accuracy: 0.31563, post-inner-loop test accuracy: 0.63437, meta_reg : 3.989176\n","Iteration 3630: pre-inner-loop train accuracy: 0.27500, post-inner-loop test accuracy: 0.57500, meta_reg : 3.989560\n","Iteration 3640: pre-inner-loop train accuracy: 0.27187, post-inner-loop test accuracy: 0.61875, meta_reg : 3.989434\n","Iteration 3650: pre-inner-loop train accuracy: 0.31875, post-inner-loop test accuracy: 0.64062, meta_reg : 3.989614\n","Meta-validation pre-inner-loop train accuracy: 0.06875, meta-validation post-inner-loop test accuracy: 0.20938\n","Iteration 3660: pre-inner-loop train accuracy: 0.30625, post-inner-loop test accuracy: 0.58125, meta_reg : 3.989758\n","Iteration 3670: pre-inner-loop train accuracy: 0.31875, post-inner-loop test accuracy: 0.59062, meta_reg : 3.989060\n","Iteration 3680: pre-inner-loop train accuracy: 0.27500, post-inner-loop test accuracy: 0.58438, meta_reg : 3.989794\n","Iteration 3690: pre-inner-loop train accuracy: 0.31250, post-inner-loop test accuracy: 0.61875, meta_reg : 3.989828\n","Iteration 3700: pre-inner-loop train accuracy: 0.34687, post-inner-loop test accuracy: 0.60312, meta_reg : 3.990061\n","Meta-validation pre-inner-loop train accuracy: 0.05625, meta-validation post-inner-loop test accuracy: 0.25313\n","Iteration 3710: pre-inner-loop train accuracy: 0.29688, post-inner-loop test accuracy: 0.59375, meta_reg : 3.989481\n","Iteration 3720: pre-inner-loop train accuracy: 0.29062, post-inner-loop test accuracy: 0.59375, meta_reg : 3.989280\n","Iteration 3730: pre-inner-loop train accuracy: 0.34062, post-inner-loop test accuracy: 0.59688, meta_reg : 3.989245\n","Iteration 3740: pre-inner-loop train accuracy: 0.33437, post-inner-loop test accuracy: 0.58438, meta_reg : 3.989989\n","Iteration 3750: pre-inner-loop train accuracy: 0.25625, post-inner-loop test accuracy: 0.60625, meta_reg : 3.989647\n","Meta-validation pre-inner-loop train accuracy: 0.05313, meta-validation post-inner-loop test accuracy: 0.25938\n","Iteration 3760: pre-inner-loop train accuracy: 0.30000, post-inner-loop test accuracy: 0.59688, meta_reg : 3.989405\n","Iteration 3770: pre-inner-loop train accuracy: 0.32812, post-inner-loop test accuracy: 0.57187, meta_reg : 3.989483\n","Iteration 3780: pre-inner-loop train accuracy: 0.33125, post-inner-loop test accuracy: 0.63750, meta_reg : 3.990100\n","Iteration 3790: pre-inner-loop train accuracy: 0.32188, post-inner-loop test accuracy: 0.60625, meta_reg : 3.989388\n","Iteration 3800: pre-inner-loop train accuracy: 0.31875, post-inner-loop test accuracy: 0.64375, meta_reg : 3.989078\n","Meta-validation pre-inner-loop train accuracy: 0.10000, meta-validation post-inner-loop test accuracy: 0.23125\n","Iteration 3810: pre-inner-loop train accuracy: 0.33437, post-inner-loop test accuracy: 0.57812, meta_reg : 3.989679\n","Iteration 3820: pre-inner-loop train accuracy: 0.36250, post-inner-loop test accuracy: 0.61875, meta_reg : 3.989188\n","Iteration 3830: pre-inner-loop train accuracy: 0.32812, post-inner-loop test accuracy: 0.59375, meta_reg : 3.989884\n","Iteration 3840: pre-inner-loop train accuracy: 0.27813, post-inner-loop test accuracy: 0.62500, meta_reg : 3.989360\n","Iteration 3850: pre-inner-loop train accuracy: 0.32500, post-inner-loop test accuracy: 0.63750, meta_reg : 3.989467\n","Meta-validation pre-inner-loop train accuracy: 0.04688, meta-validation post-inner-loop test accuracy: 0.26562\n","Iteration 3860: pre-inner-loop train accuracy: 0.33437, post-inner-loop test accuracy: 0.58750, meta_reg : 3.989858\n","Iteration 3870: pre-inner-loop train accuracy: 0.31250, post-inner-loop test accuracy: 0.61875, meta_reg : 3.989644\n","Iteration 3880: pre-inner-loop train accuracy: 0.34688, post-inner-loop test accuracy: 0.65625, meta_reg : 3.989711\n","Iteration 3890: pre-inner-loop train accuracy: 0.29375, post-inner-loop test accuracy: 0.59062, meta_reg : 3.988889\n","Iteration 3900: pre-inner-loop train accuracy: 0.34375, post-inner-loop test accuracy: 0.57812, meta_reg : 3.990235\n","Meta-validation pre-inner-loop train accuracy: 0.04375, meta-validation post-inner-loop test accuracy: 0.21875\n","Iteration 3910: pre-inner-loop train accuracy: 0.32187, post-inner-loop test accuracy: 0.63750, meta_reg : 3.989048\n","Iteration 3920: pre-inner-loop train accuracy: 0.31250, post-inner-loop test accuracy: 0.59062, meta_reg : 3.989522\n","Iteration 3930: pre-inner-loop train accuracy: 0.32500, post-inner-loop test accuracy: 0.61563, meta_reg : 3.989568\n","Iteration 3940: pre-inner-loop train accuracy: 0.29375, post-inner-loop test accuracy: 0.57812, meta_reg : 3.989413\n","Iteration 3950: pre-inner-loop train accuracy: 0.40000, post-inner-loop test accuracy: 0.63438, meta_reg : 3.989695\n","Meta-validation pre-inner-loop train accuracy: 0.04062, meta-validation post-inner-loop test accuracy: 0.21875\n","Iteration 3960: pre-inner-loop train accuracy: 0.36250, post-inner-loop test accuracy: 0.61250, meta_reg : 3.989546\n","Iteration 3970: pre-inner-loop train accuracy: 0.30313, post-inner-loop test accuracy: 0.61875, meta_reg : 3.989048\n","Iteration 3980: pre-inner-loop train accuracy: 0.39687, post-inner-loop test accuracy: 0.65000, meta_reg : 3.990520\n","Iteration 3990: pre-inner-loop train accuracy: 0.25313, post-inner-loop test accuracy: 0.61563, meta_reg : 3.988897\n","Iteration 4000: pre-inner-loop train accuracy: 0.30937, post-inner-loop test accuracy: 0.60625, meta_reg : 3.989229\n","Meta-validation pre-inner-loop train accuracy: 0.05938, meta-validation post-inner-loop test accuracy: 0.26875\n","Iteration 4010: pre-inner-loop train accuracy: 0.32500, post-inner-loop test accuracy: 0.57812, meta_reg : 3.990021\n","Iteration 4020: pre-inner-loop train accuracy: 0.30625, post-inner-loop test accuracy: 0.59688, meta_reg : 3.988541\n","Iteration 4030: pre-inner-loop train accuracy: 0.31875, post-inner-loop test accuracy: 0.60938, meta_reg : 3.989859\n","Iteration 4040: pre-inner-loop train accuracy: 0.34375, post-inner-loop test accuracy: 0.64688, meta_reg : 3.989820\n","Iteration 4050: pre-inner-loop train accuracy: 0.31563, post-inner-loop test accuracy: 0.63750, meta_reg : 3.989272\n","Meta-validation pre-inner-loop train accuracy: 0.02812, meta-validation post-inner-loop test accuracy: 0.25000\n","Iteration 4060: pre-inner-loop train accuracy: 0.34688, post-inner-loop test accuracy: 0.65625, meta_reg : 3.989852\n","Iteration 4070: pre-inner-loop train accuracy: 0.32187, post-inner-loop test accuracy: 0.68125, meta_reg : 3.989505\n","Iteration 4080: pre-inner-loop train accuracy: 0.30937, post-inner-loop test accuracy: 0.60000, meta_reg : 3.989701\n","Iteration 4090: pre-inner-loop train accuracy: 0.30000, post-inner-loop test accuracy: 0.61563, meta_reg : 3.989353\n","Iteration 4100: pre-inner-loop train accuracy: 0.28438, post-inner-loop test accuracy: 0.66562, meta_reg : 3.989179\n","Meta-validation pre-inner-loop train accuracy: 0.03750, meta-validation post-inner-loop test accuracy: 0.24375\n","Iteration 4110: pre-inner-loop train accuracy: 0.33125, post-inner-loop test accuracy: 0.63750, meta_reg : 3.989442\n","Iteration 4120: pre-inner-loop train accuracy: 0.33750, post-inner-loop test accuracy: 0.64062, meta_reg : 3.989832\n","Iteration 4130: pre-inner-loop train accuracy: 0.31250, post-inner-loop test accuracy: 0.61875, meta_reg : 3.989269\n","Iteration 4140: pre-inner-loop train accuracy: 0.36250, post-inner-loop test accuracy: 0.63125, meta_reg : 3.989846\n","Iteration 4150: pre-inner-loop train accuracy: 0.30625, post-inner-loop test accuracy: 0.62813, meta_reg : 3.989058\n","Meta-validation pre-inner-loop train accuracy: 0.04375, meta-validation post-inner-loop test accuracy: 0.21250\n","Iteration 4160: pre-inner-loop train accuracy: 0.29375, post-inner-loop test accuracy: 0.64688, meta_reg : 3.990063\n","Iteration 4170: pre-inner-loop train accuracy: 0.32188, post-inner-loop test accuracy: 0.63750, meta_reg : 3.989180\n","Iteration 4180: pre-inner-loop train accuracy: 0.37500, post-inner-loop test accuracy: 0.61250, meta_reg : 3.990364\n","Iteration 4190: pre-inner-loop train accuracy: 0.33125, post-inner-loop test accuracy: 0.63437, meta_reg : 3.989405\n","Iteration 4200: pre-inner-loop train accuracy: 0.32812, post-inner-loop test accuracy: 0.63437, meta_reg : 3.989208\n","Meta-validation pre-inner-loop train accuracy: 0.06562, meta-validation post-inner-loop test accuracy: 0.21875\n","Iteration 4210: pre-inner-loop train accuracy: 0.36875, post-inner-loop test accuracy: 0.65312, meta_reg : 3.989812\n","Iteration 4220: pre-inner-loop train accuracy: 0.36250, post-inner-loop test accuracy: 0.64062, meta_reg : 3.989415\n","Iteration 4230: pre-inner-loop train accuracy: 0.33750, post-inner-loop test accuracy: 0.69687, meta_reg : 3.989722\n","Iteration 4240: pre-inner-loop train accuracy: 0.32812, post-inner-loop test accuracy: 0.63437, meta_reg : 3.989479\n","Iteration 4250: pre-inner-loop train accuracy: 0.35000, post-inner-loop test accuracy: 0.64062, meta_reg : 3.989273\n","Meta-validation pre-inner-loop train accuracy: 0.07500, meta-validation post-inner-loop test accuracy: 0.23125\n","Iteration 4260: pre-inner-loop train accuracy: 0.37500, post-inner-loop test accuracy: 0.65000, meta_reg : 3.989641\n","Iteration 4270: pre-inner-loop train accuracy: 0.34375, post-inner-loop test accuracy: 0.61250, meta_reg : 3.989409\n","Iteration 4280: pre-inner-loop train accuracy: 0.29375, post-inner-loop test accuracy: 0.61875, meta_reg : 3.989106\n","Iteration 4290: pre-inner-loop train accuracy: 0.33750, post-inner-loop test accuracy: 0.64688, meta_reg : 3.989773\n","Iteration 4300: pre-inner-loop train accuracy: 0.30312, post-inner-loop test accuracy: 0.59687, meta_reg : 3.989095\n","Meta-validation pre-inner-loop train accuracy: 0.03438, meta-validation post-inner-loop test accuracy: 0.30000\n","Iteration 4310: pre-inner-loop train accuracy: 0.33125, post-inner-loop test accuracy: 0.63125, meta_reg : 3.989513\n","Iteration 4320: pre-inner-loop train accuracy: 0.34375, post-inner-loop test accuracy: 0.63438, meta_reg : 3.989279\n","Iteration 4330: pre-inner-loop train accuracy: 0.32188, post-inner-loop test accuracy: 0.65000, meta_reg : 3.989419\n","Iteration 4340: pre-inner-loop train accuracy: 0.27500, post-inner-loop test accuracy: 0.59063, meta_reg : 3.989001\n","Iteration 4350: pre-inner-loop train accuracy: 0.35000, post-inner-loop test accuracy: 0.67188, meta_reg : 3.989499\n","Meta-validation pre-inner-loop train accuracy: 0.08438, meta-validation post-inner-loop test accuracy: 0.26875\n","Iteration 4360: pre-inner-loop train accuracy: 0.33750, post-inner-loop test accuracy: 0.65625, meta_reg : 3.988746\n","Iteration 4370: pre-inner-loop train accuracy: 0.31563, post-inner-loop test accuracy: 0.62813, meta_reg : 3.988569\n","Iteration 4380: pre-inner-loop train accuracy: 0.33125, post-inner-loop test accuracy: 0.64062, meta_reg : 3.989212\n","Iteration 4390: pre-inner-loop train accuracy: 0.35938, post-inner-loop test accuracy: 0.68125, meta_reg : 3.989274\n","Iteration 4400: pre-inner-loop train accuracy: 0.35000, post-inner-loop test accuracy: 0.59688, meta_reg : 3.990217\n","Meta-validation pre-inner-loop train accuracy: 0.07812, meta-validation post-inner-loop test accuracy: 0.21875\n","Iteration 4410: pre-inner-loop train accuracy: 0.38125, post-inner-loop test accuracy: 0.67188, meta_reg : 3.989420\n","Iteration 4420: pre-inner-loop train accuracy: 0.40625, post-inner-loop test accuracy: 0.69063, meta_reg : 3.990341\n","Iteration 4430: pre-inner-loop train accuracy: 0.32812, post-inner-loop test accuracy: 0.60938, meta_reg : 3.989453\n","Iteration 4440: pre-inner-loop train accuracy: 0.35000, post-inner-loop test accuracy: 0.60000, meta_reg : 3.989580\n","Iteration 4450: pre-inner-loop train accuracy: 0.34375, post-inner-loop test accuracy: 0.63750, meta_reg : 3.989443\n","Meta-validation pre-inner-loop train accuracy: 0.07500, meta-validation post-inner-loop test accuracy: 0.21875\n","Iteration 4460: pre-inner-loop train accuracy: 0.31875, post-inner-loop test accuracy: 0.65625, meta_reg : 3.989375\n","Iteration 4470: pre-inner-loop train accuracy: 0.33750, post-inner-loop test accuracy: 0.60938, meta_reg : 3.989135\n","Iteration 4480: pre-inner-loop train accuracy: 0.31250, post-inner-loop test accuracy: 0.66562, meta_reg : 3.989125\n","Iteration 4490: pre-inner-loop train accuracy: 0.32812, post-inner-loop test accuracy: 0.67500, meta_reg : 3.989994\n","Iteration 4500: pre-inner-loop train accuracy: 0.39375, post-inner-loop test accuracy: 0.63125, meta_reg : 3.988969\n","Meta-validation pre-inner-loop train accuracy: 0.05313, meta-validation post-inner-loop test accuracy: 0.24375\n","Iteration 4510: pre-inner-loop train accuracy: 0.32812, post-inner-loop test accuracy: 0.65625, meta_reg : 3.989725\n","Iteration 4520: pre-inner-loop train accuracy: 0.37500, post-inner-loop test accuracy: 0.63125, meta_reg : 3.989129\n","Iteration 4530: pre-inner-loop train accuracy: 0.31875, post-inner-loop test accuracy: 0.63437, meta_reg : 3.989032\n","Iteration 4540: pre-inner-loop train accuracy: 0.33438, post-inner-loop test accuracy: 0.62500, meta_reg : 3.989602\n","Iteration 4550: pre-inner-loop train accuracy: 0.40000, post-inner-loop test accuracy: 0.70625, meta_reg : 3.989554\n","Meta-validation pre-inner-loop train accuracy: 0.03125, meta-validation post-inner-loop test accuracy: 0.20625\n","Iteration 4560: pre-inner-loop train accuracy: 0.30937, post-inner-loop test accuracy: 0.63125, meta_reg : 3.988796\n","Iteration 4570: pre-inner-loop train accuracy: 0.31875, post-inner-loop test accuracy: 0.66562, meta_reg : 3.990000\n","Iteration 4580: pre-inner-loop train accuracy: 0.33125, post-inner-loop test accuracy: 0.67813, meta_reg : 3.988779\n","Iteration 4590: pre-inner-loop train accuracy: 0.35000, post-inner-loop test accuracy: 0.62500, meta_reg : 3.989278\n","Iteration 4600: pre-inner-loop train accuracy: 0.31875, post-inner-loop test accuracy: 0.67188, meta_reg : 3.989104\n","Meta-validation pre-inner-loop train accuracy: 0.05000, meta-validation post-inner-loop test accuracy: 0.24375\n","Iteration 4610: pre-inner-loop train accuracy: 0.32188, post-inner-loop test accuracy: 0.66875, meta_reg : 3.988751\n","Iteration 4620: pre-inner-loop train accuracy: 0.34375, post-inner-loop test accuracy: 0.66562, meta_reg : 3.989446\n","Iteration 4630: pre-inner-loop train accuracy: 0.39062, post-inner-loop test accuracy: 0.66562, meta_reg : 3.989555\n","Iteration 4640: pre-inner-loop train accuracy: 0.39062, post-inner-loop test accuracy: 0.64062, meta_reg : 3.989176\n","Iteration 4650: pre-inner-loop train accuracy: 0.35938, post-inner-loop test accuracy: 0.70312, meta_reg : 3.989502\n","Meta-validation pre-inner-loop train accuracy: 0.06563, meta-validation post-inner-loop test accuracy: 0.26250\n","Iteration 4660: pre-inner-loop train accuracy: 0.35000, post-inner-loop test accuracy: 0.61563, meta_reg : 3.989651\n","Iteration 4670: pre-inner-loop train accuracy: 0.39375, post-inner-loop test accuracy: 0.66875, meta_reg : 3.989443\n","Iteration 4680: pre-inner-loop train accuracy: 0.34688, post-inner-loop test accuracy: 0.62500, meta_reg : 3.989077\n","Iteration 4690: pre-inner-loop train accuracy: 0.39688, post-inner-loop test accuracy: 0.65000, meta_reg : 3.989091\n","Iteration 4700: pre-inner-loop train accuracy: 0.40000, post-inner-loop test accuracy: 0.65625, meta_reg : 3.989141\n","Meta-validation pre-inner-loop train accuracy: 0.06875, meta-validation post-inner-loop test accuracy: 0.25000\n","Iteration 4710: pre-inner-loop train accuracy: 0.34688, post-inner-loop test accuracy: 0.63437, meta_reg : 3.989647\n","Iteration 4720: pre-inner-loop train accuracy: 0.37187, post-inner-loop test accuracy: 0.61563, meta_reg : 3.989372\n","Iteration 4730: pre-inner-loop train accuracy: 0.38750, post-inner-loop test accuracy: 0.66562, meta_reg : 3.989111\n","Iteration 4740: pre-inner-loop train accuracy: 0.31563, post-inner-loop test accuracy: 0.66875, meta_reg : 3.988605\n","Iteration 4750: pre-inner-loop train accuracy: 0.35000, post-inner-loop test accuracy: 0.61250, meta_reg : 3.989202\n","Meta-validation pre-inner-loop train accuracy: 0.04375, meta-validation post-inner-loop test accuracy: 0.22500\n","Iteration 4760: pre-inner-loop train accuracy: 0.33750, post-inner-loop test accuracy: 0.60312, meta_reg : 3.988825\n","Iteration 4770: pre-inner-loop train accuracy: 0.33125, post-inner-loop test accuracy: 0.68125, meta_reg : 3.989015\n","Iteration 4780: pre-inner-loop train accuracy: 0.39687, post-inner-loop test accuracy: 0.66562, meta_reg : 3.988895\n","Iteration 4790: pre-inner-loop train accuracy: 0.38438, post-inner-loop test accuracy: 0.69062, meta_reg : 3.989648\n","Iteration 4800: pre-inner-loop train accuracy: 0.35938, post-inner-loop test accuracy: 0.62813, meta_reg : 3.988847\n","Meta-validation pre-inner-loop train accuracy: 0.05000, meta-validation post-inner-loop test accuracy: 0.22500\n","Iteration 4810: pre-inner-loop train accuracy: 0.32500, post-inner-loop test accuracy: 0.65938, meta_reg : 3.989333\n","Iteration 4820: pre-inner-loop train accuracy: 0.34375, post-inner-loop test accuracy: 0.63437, meta_reg : 3.989321\n","Iteration 4830: pre-inner-loop train accuracy: 0.32500, post-inner-loop test accuracy: 0.68437, meta_reg : 3.989227\n","Iteration 4840: pre-inner-loop train accuracy: 0.39062, post-inner-loop test accuracy: 0.68437, meta_reg : 3.989952\n","Iteration 4850: pre-inner-loop train accuracy: 0.34688, post-inner-loop test accuracy: 0.64688, meta_reg : 3.989069\n","Meta-validation pre-inner-loop train accuracy: 0.03125, meta-validation post-inner-loop test accuracy: 0.24687\n","Iteration 4860: pre-inner-loop train accuracy: 0.35313, post-inner-loop test accuracy: 0.66250, meta_reg : 3.989266\n","Iteration 4870: pre-inner-loop train accuracy: 0.35938, post-inner-loop test accuracy: 0.68125, meta_reg : 3.989110\n","Iteration 4880: pre-inner-loop train accuracy: 0.40625, post-inner-loop test accuracy: 0.66875, meta_reg : 3.989439\n","Iteration 4890: pre-inner-loop train accuracy: 0.36562, post-inner-loop test accuracy: 0.66562, meta_reg : 3.989397\n","Iteration 4900: pre-inner-loop train accuracy: 0.33125, post-inner-loop test accuracy: 0.68125, meta_reg : 3.988878\n","Meta-validation pre-inner-loop train accuracy: 0.05625, meta-validation post-inner-loop test accuracy: 0.27188\n","Iteration 4910: pre-inner-loop train accuracy: 0.40000, post-inner-loop test accuracy: 0.65937, meta_reg : 3.989212\n","Iteration 4920: pre-inner-loop train accuracy: 0.39688, post-inner-loop test accuracy: 0.67188, meta_reg : 3.989285\n","Iteration 4930: pre-inner-loop train accuracy: 0.35625, post-inner-loop test accuracy: 0.67500, meta_reg : 3.989210\n","Iteration 4940: pre-inner-loop train accuracy: 0.34062, post-inner-loop test accuracy: 0.62813, meta_reg : 3.988665\n","Iteration 4950: pre-inner-loop train accuracy: 0.41875, post-inner-loop test accuracy: 0.65938, meta_reg : 3.989571\n","Meta-validation pre-inner-loop train accuracy: 0.05938, meta-validation post-inner-loop test accuracy: 0.21563\n","Iteration 4960: pre-inner-loop train accuracy: 0.36875, post-inner-loop test accuracy: 0.62187, meta_reg : 3.989474\n","Iteration 4970: pre-inner-loop train accuracy: 0.34062, post-inner-loop test accuracy: 0.64062, meta_reg : 3.988740\n","Iteration 4980: pre-inner-loop train accuracy: 0.33438, post-inner-loop test accuracy: 0.65625, meta_reg : 3.989201\n","Iteration 4990: pre-inner-loop train accuracy: 0.39062, post-inner-loop test accuracy: 0.64062, meta_reg : 3.989298\n","Saving to  /content/project_data/data/checkpoint/cls_20.mbs_16.k_shot_1.inner_numstep_1.inner_updatelr_0.04.learn_inner_update_lr_False.train_typememorization/model4999\n","Mean meta-test accuracy/loss, stddev, and confidence intervals\n","(0.27875, 0.090384014, 0.007232227359414484)\n","pre accuracy stats\n","mean 0.04966667 std 0.047520757\n","post accuracy stats\n","mean 0.27875 std 0.090384014\n"],"name":"stdout"}]},{"cell_type":"code","metadata":{"id":"dYZPCDCsCSsu","colab":{"base_uri":"https://localhost:8080/"},"executionInfo":{"status":"ok","timestamp":1606186719697,"user_tz":360,"elapsed":372,"user":{"displayName":"Pankaj Rajak","photoUrl":"https://lh3.googleusercontent.com/a-/AOh14GiTXe8B8vdrh4uTyUPqEtb-tRwPltWOZurP8-lxog=s64","userId":"04274727635571437779"}},"outputId":"d1676484-902a-4067-e944-9f04f98e450d"},"source":["result_path = '{}/MAML-n_20-k_1_lr_0-4_out_0-0025_memo_ratio3'.format(SYM_PATH)\n","if not os.path.exists(result_path):\n","  %mkdir $result_path"],"execution_count":null,"outputs":[{"output_type":"stream","text":["[5, 22]\n","[5, 2]\n","[6, 7]\n","[18, 4]\n","[14, 5]\n","[8, 17]\n","[0, 13]\n","[6, 3]\n","[16, 15]\n","[3, 12]\n"],"name":"stdout"}]},{"cell_type":"code","metadata":{"id":"HZpSyDXoAkT8","colab":{"base_uri":"https://localhost:8080/","height":397},"executionInfo":{"status":"ok","timestamp":1606186724511,"user_tz":360,"elapsed":2394,"user":{"displayName":"Pankaj Rajak","photoUrl":"https://lh3.googleusercontent.com/a-/AOh14GiTXe8B8vdrh4uTyUPqEtb-tRwPltWOZurP8-lxog=s64","userId":"04274727635571437779"}},"outputId":"13137900-1a62-4e5b-a2ca-babf18fcbd75"},"source":["import matplotlib.pyplot as plt\n","\n","fig,(ax1, ax2) = plt.subplots(ncols=2,figsize=(20,6))\n","\n","ax1.plot(step_train_trj,pre_train_accuracy_trj,label='pre-accuracy',c='r')\n","ax1.plot(step_train_trj,post_train_accuracy_trj,label='post-accuracy',c='b')\n","\n","ax2.plot(step_num_trj,pre_val_accuracy_trj,label='pre-accuracy',c='r')\n","ax2.plot(step_num_trj,post_val_accuracy_trj,label='post-accuracy',c='b')\n","\n","ax1.legend(fontsize=16)\n","ax2.legend(fontsize=16)\n","ax1.set_ylim(0.0,1.0)\n","ax2.set_ylim(0.0,1.0)\n","ax1.set_ylabel('Accuracy',weight='bold',fontsize=16)\n","ax1.set_xlabel('Iterations',weight='bold',fontsize=16)\n","ax2.set_ylabel('Accuracy',weight='bold',fontsize=16)\n","ax2.set_xlabel('Iterations',weight='bold',fontsize=16)\n","img_path = '{}/MAML-n_20-k-1_lr_0-4-memo.png'.format(result_path)\n","plt.savefig(img_path,dpi=300)\n","plt.show()"],"execution_count":null,"outputs":[{"output_type":"display_data","data":{"image/png":"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\n","text/plain":["<Figure size 1440x432 with 2 Axes>"]},"metadata":{"tags":[],"needs_background":"light"}}]},{"cell_type":"code","metadata":{"id":"ArQTy51fCRDe","colab":{"base_uri":"https://localhost:8080/","height":397},"executionInfo":{"status":"ok","timestamp":1606186738127,"user_tz":360,"elapsed":2221,"user":{"displayName":"Pankaj Rajak","photoUrl":"https://lh3.googleusercontent.com/a-/AOh14GiTXe8B8vdrh4uTyUPqEtb-tRwPltWOZurP8-lxog=s64","userId":"04274727635571437779"}},"outputId":"8c049f3e-b130-4d6d-aa89-c251b9077942"},"source":["fig,(ax1, ax2) = plt.subplots(ncols=2,figsize=(20,6))\n","\n","ax1.plot(step_train_trj,np.asarray(post_train_accuracy_trj)-np.asarray(pre_train_accuracy_trj),label='post_accuracy - pre_accuracy',c='b')\n","ax2.plot(step_num_trj,np.asarray(post_val_accuracy_trj)-np.asarray(pre_val_accuracy_trj),label='post_accuracy - pre_accuracy',c='b')\n","\n","ax1.legend(fontsize=16)\n","ax2.legend(fontsize=16)\n","ax1.set_ylim(0.0,1.0)\n","ax2.set_ylim(0.0,1.0)\n","ax1.set_ylabel('Accuracy Difference',weight='bold',fontsize=16)\n","ax1.set_xlabel('Iterations',weight='bold',fontsize=16)\n","ax2.set_ylabel('Accuracy Difference',weight='bold',fontsize=16)\n","ax2.set_xlabel('Iterations',weight='bold',fontsize=16)\n","img_path = '{}/MAML-n_20-k-1_lr_0-04-memo-diff.png'.format(result_path)\n","plt.savefig(img_path,dpi=300)\n","plt.show()"],"execution_count":null,"outputs":[{"output_type":"display_data","data":{"image/png":"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\n","text/plain":["<Figure size 1440x432 with 2 Axes>"]},"metadata":{"tags":[],"needs_background":"light"}}]},{"cell_type":"code","metadata":{"id":"EOChjSsKStx8","colab":{"base_uri":"https://localhost:8080/"},"executionInfo":{"status":"ok","timestamp":1606186438301,"user_tz":360,"elapsed":2179594,"user":{"displayName":"Pankaj Rajak","photoUrl":"https://lh3.googleusercontent.com/a-/AOh14GiTXe8B8vdrh4uTyUPqEtb-tRwPltWOZurP8-lxog=s64","userId":"04274727635571437779"}},"outputId":"bae4c55b-5f7b-4212-d4bf-905e1fbcd53a"},"source":["means,stds,ci95=run_maml(n_way=20, k_shot=1, inner_update_lr=0.04, num_inner_updates=1,meta_train=False,meta_lr=0.0025,\n","                         meta_train_k_shot=1,logdir=log_path,train_type='memorization')"],"execution_count":null,"outputs":[{"output_type":"stream","text":["Number of training examples. :  1200\n","Number of validation examples:  100\n","Number of test examples      :  323\n","Training Type:  memorization\n","Restoring model weights from  /content/project_data/data/checkpoint/cls_20.mbs_25.k_shot_1.inner_numstep_1.inner_updatelr_0.04.learn_inner_update_lr_False.train_typememorization/model4999\n","Mean meta-test accuracy/loss, stddev, and confidence intervals\n","(0.165, 0.083765544, 0.006702639466786973)\n","pre accuracy stats\n","mean 0.04625 std 0.04670943\n","post accuracy stats\n","mean 0.165 std 0.083765544\n"],"name":"stdout"}]}]}