CAAE.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/home/zhangj/anaconda2/envs/Python35/lib/python3.5/site-packages/h5py/__init__.py:36: FutureWarning: Conversion of the second argument of issubdtype from `float` to `np.floating` is deprecated. In future, it will be treated as `np.float64 == np.dtype(float).type`.\n",
      "  from ._conv import register_converters as _register_converters\n",
      "Using TensorFlow backend.\n"
     ]
    }
   ],
   "source": [
    "#-*- coding: utf-8 -*-\n",
    "from aae_utils import set_gpu_usage,get_train_test,sample_fault,sample_from_prior\n",
    "from data import data_loader\n",
    "\n",
    "import keras.backend as K\n",
    "from keras.layers import Input,Dense,Merge,merge\n",
    "from keras.layers.core import Activation\n",
    "from keras.models import Model,Sequential\n",
    "from keras.optimizers import Adam\n",
    "from keras.layers.normalization import BatchNormalization\n",
    "from keras.utils.generic_utils import Progbar\n",
    "from collections import defaultdict\n",
    "import pickle\n",
    "import numpy as np\n",
    "import tensorflow as tf\n",
    "import matplotlib.patches as mpatches\n",
    "from math import sin,cos,sqrt\n",
    "\n",
    "import matplotlib.pyplot as plt\n",
    "from matplotlib.pyplot import cm\n",
    "import matplotlib\n",
    "matplotlib.pyplot.switch_backend('agg')\n",
    "from mpl_toolkits.mplot3d import Axes3D\n",
    "import os\n",
    "import errno\n",
    "\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "def set_gpu_usage(ratio=0.9,gpu=1):\n",
    "    config=tf.ConfigProto(gpu_options=tf.GPUOptions(\n",
    "                                        per_process_gpu_memory_fraction=ratio,\n",
    "                                        allow_growth=True),\n",
    "                          device_count={'GPU':gpu},\n",
    "                          allow_soft_placement=True)\n",
    "    sess=tf.Session(config=config)\n",
    "    K.set_session(sess)\n",
    "\n",
    "\n",
    "def make_sure_path_exists(path):\n",
    "    try:\n",
    "        os.makedirs(path)\n",
    "    except OSError as exception:\n",
    "        if exception.errno != errno.EEXIST:\n",
    "            raise\n",
    "def get_normalized_vector(v):\n",
    "    v=v/(1e-20+K.max(K.abs(v),axis=1,keepdims=True))\n",
    "    v_2=K.sum(v**2,axis=1,keepdims=True)\n",
    "    return v / K.sqrt(1e-6+v_2)\n",
    "\n",
    "# sample function of z_p\n",
    "\n",
    "def make_trainable(net, val):\n",
    "    net.trainable = val\n",
    "    for l in net.layers:\n",
    "        l.trainable = val"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "latent_dim = 2\n",
    "z_prior='gaussian'\n",
    "lamb=-20\n",
    "recon_obj_type='QE'\n",
    "initial_learning_rate=0.003\n",
    "learning_rate_decay=1.0\n",
    "num_epochs=20\n",
    "batch_size=10\n",
    "save_filename='trained_model'\n",
    "seed=1\n",
    "path='./datasets'\n",
    "\n",
    "opt_gen=Adam(lr=0.00001,decay=0.99)\n",
    "opt_rec=Adam(lr=0.001,decay=0.99)\n",
    "opt_disc=Adam(lr=0.02,decay=0.99)\n",
    "\n",
    "make_sure_path_exists(path)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Data_preprocess"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "the shape of x_train:\n",
      "(2034, 52)\n",
      "the shape of x_test:\n",
      "(226, 52)\n",
      "the shape of y_train:\n",
      "(2034, 52)\n",
      "the shape of y_test:\n",
      "(226, 52)\n",
      "[[1.]\n",
      " [0.]\n",
      " [0.]\n",
      " [0.]\n",
      " [1.]\n",
      " [1.]\n",
      " [0.]\n",
      " [1.]\n",
      " [0.]\n",
      " [1.]\n",
      " [1.]\n",
      " [1.]\n",
      " [1.]\n",
      " [0.]\n",
      " [0.]\n",
      " [0.]\n",
      " [1.]\n",
      " [1.]\n",
      " [1.]\n",
      " [1.]\n",
      " [1.]\n",
      " [1.]\n",
      " [1.]\n",
      " [0.]\n",
      " [1.]\n",
      " [1.]\n",
      " [1.]\n",
      " [1.]\n",
      " [1.]\n",
      " [1.]\n",
      " [1.]\n",
      " [0.]\n",
      " [1.]\n",
      " [1.]\n",
      " [1.]\n",
      " [1.]\n",
      " [1.]\n",
      " [1.]\n",
      " [1.]\n",
      " [0.]\n",
      " [1.]\n",
      " [1.]\n",
      " [1.]\n",
      " [1.]\n",
      " [1.]\n",
      " [1.]\n",
      " [1.]\n",
      " [0.]\n",
      " [1.]\n",
      " [0.]\n",
      " [1.]\n",
      " [1.]\n",
      " [1.]\n",
      " [1.]\n",
      " [1.]\n",
      " [1.]\n",
      " [1.]\n",
      " [0.]\n",
      " [1.]\n",
      " [0.]\n",
      " [1.]\n",
      " [1.]\n",
      " [1.]\n",
      " [1.]\n",
      " [1.]\n",
      " [1.]\n",
      " [0.]\n",
      " [0.]\n",
      " [1.]\n",
      " [1.]\n",
      " [0.]\n",
      " [0.]\n",
      " [0.]\n",
      " [1.]\n",
      " [1.]\n",
      " [1.]\n",
      " [0.]\n",
      " [0.]\n",
      " [1.]\n",
      " [0.]\n",
      " [1.]\n",
      " [0.]\n",
      " [1.]\n",
      " [0.]\n",
      " [0.]\n",
      " [1.]\n",
      " [1.]\n",
      " [1.]\n",
      " [1.]\n",
      " [1.]\n",
      " [1.]\n",
      " [1.]\n",
      " [1.]\n",
      " [1.]\n",
      " [1.]\n",
      " [1.]\n",
      " [1.]\n",
      " [1.]\n",
      " [0.]\n",
      " [0.]\n",
      " [0.]\n",
      " [1.]\n",
      " [1.]\n",
      " [1.]\n",
      " [1.]\n",
      " [1.]\n",
      " [1.]\n",
      " [1.]\n",
      " [1.]\n",
      " [1.]\n",
      " [1.]\n",
      " [1.]\n",
      " [0.]\n",
      " [1.]\n",
      " [1.]\n",
      " [1.]\n",
      " [1.]\n",
      " [0.]\n",
      " [0.]\n",
      " [1.]\n",
      " [1.]\n",
      " [1.]\n",
      " [1.]\n",
      " [0.]\n",
      " [0.]\n",
      " [1.]\n",
      " [1.]\n",
      " [0.]\n",
      " [1.]\n",
      " [0.]\n",
      " [1.]\n",
      " [1.]\n",
      " [1.]\n",
      " [1.]\n",
      " [0.]\n",
      " [0.]\n",
      " [1.]\n",
      " [1.]\n",
      " [0.]\n",
      " [1.]\n",
      " [1.]\n",
      " [0.]\n",
      " [1.]\n",
      " [0.]\n",
      " [1.]\n",
      " [0.]\n",
      " [0.]\n",
      " [1.]\n",
      " [1.]\n",
      " [0.]\n",
      " [1.]\n",
      " [1.]\n",
      " [1.]\n",
      " [0.]\n",
      " [1.]\n",
      " [1.]\n",
      " [0.]\n",
      " [0.]\n",
      " [1.]\n",
      " [1.]\n",
      " [1.]\n",
      " [1.]\n",
      " [1.]\n",
      " [1.]\n",
      " [0.]\n",
      " [1.]\n",
      " [1.]\n",
      " [0.]\n",
      " [1.]\n",
      " [0.]\n",
      " [1.]\n",
      " [0.]\n",
      " [1.]\n",
      " [1.]\n",
      " [1.]\n",
      " [0.]\n",
      " [1.]\n",
      " [1.]\n",
      " [1.]\n",
      " [1.]\n",
      " [1.]\n",
      " [1.]\n",
      " [1.]\n",
      " [0.]\n",
      " [1.]\n",
      " [1.]\n",
      " [1.]\n",
      " [0.]\n",
      " [0.]\n",
      " [0.]\n",
      " [1.]\n",
      " [1.]\n",
      " [1.]\n",
      " [1.]\n",
      " [0.]\n",
      " [1.]\n",
      " [0.]\n",
      " [0.]\n",
      " [1.]\n",
      " [0.]\n",
      " [1.]\n",
      " [0.]\n",
      " [0.]\n",
      " [1.]\n",
      " [1.]\n",
      " [1.]\n",
      " [1.]\n",
      " [1.]\n",
      " [1.]\n",
      " [1.]\n",
      " [1.]\n",
      " [0.]\n",
      " [1.]\n",
      " [1.]\n",
      " [1.]\n",
      " [1.]\n",
      " [1.]\n",
      " [0.]\n",
      " [0.]\n",
      " [1.]\n",
      " [1.]\n",
      " [0.]\n",
      " [1.]\n",
      " [1.]\n",
      " [0.]\n",
      " [0.]\n",
      " [0.]\n",
      " [1.]\n",
      " [0.]\n",
      " [0.]\n",
      " [0.]\n",
      " [1.]\n",
      " [1.]\n",
      " [0.]\n",
      " [0.]\n",
      " [0.]\n",
      " [0.]\n",
      " [1.]\n",
      " [1.]\n",
      " [0.]\n",
      " [1.]\n",
      " [1.]\n",
      " [1.]\n",
      " [1.]\n",
      " [1.]\n",
      " [0.]\n",
      " [1.]\n",
      " [1.]\n",
      " [1.]\n",
      " [1.]\n",
      " [0.]\n",
      " [1.]\n",
      " [0.]\n",
      " [1.]\n",
      " [0.]\n",
      " [1.]\n",
      " [0.]\n",
      " [1.]\n",
      " [1.]\n",
      " [0.]\n",
      " [0.]\n",
      " [1.]\n",
      " [0.]\n",
      " [1.]\n",
      " [1.]\n",
      " [0.]\n",
      " [1.]\n",
      " [0.]\n",
      " [1.]\n",
      " [1.]\n",
      " [1.]\n",
      " [0.]\n",
      " [0.]\n",
      " [0.]\n",
      " [1.]\n",
      " [1.]\n",
      " [1.]\n",
      " [0.]\n",
      " [0.]\n",
      " [1.]\n",
      " [1.]\n",
      " [1.]\n",
      " [0.]\n",
      " [1.]\n",
      " [0.]\n",
      " [0.]\n",
      " [1.]\n",
      " [0.]\n",
      " [1.]\n",
      " [1.]\n",
      " [1.]\n",
      " [0.]\n",
      " [1.]\n",
      " [1.]\n",
      " [0.]\n",
      " [0.]\n",
      " [1.]\n",
      " [1.]\n",
      " [0.]\n",
      " [0.]\n",
      " [1.]\n",
      " [0.]\n",
      " [0.]\n",
      " [0.]\n",
      " [1.]\n",
      " [1.]\n",
      " [1.]\n",
      " [0.]\n",
      " [1.]\n",
      " [1.]\n",
      " [1.]\n",
      " [0.]\n",
      " [0.]\n",
      " [0.]\n",
      " [1.]\n",
      " [0.]\n",
      " [0.]\n",
      " [1.]\n",
      " [0.]\n",
      " [1.]\n",
      " [1.]\n",
      " [1.]\n",
      " [0.]\n",
      " [1.]\n",
      " [1.]\n",
      " [0.]\n",
      " [0.]\n",
      " [0.]\n",
      " [0.]\n",
      " [0.]\n",
      " [1.]\n",
      " [1.]\n",
      " [1.]\n",
      " [1.]\n",
      " [1.]\n",
      " [1.]\n",
      " [1.]\n",
      " [0.]\n",
      " [1.]\n",
      " [1.]\n",
      " [1.]\n",
      " [1.]\n",
      " [1.]\n",
      " [1.]\n",
      " [1.]\n",
      " [1.]\n",
      " [1.]\n",
      " [1.]\n",
      " [1.]\n",
      " [0.]\n",
      " [1.]\n",
      " [1.]\n",
      " [1.]\n",
      " [0.]\n",
      " [1.]\n",
      " [0.]\n",
      " [1.]\n",
      " [1.]\n",
      " [1.]\n",
      " [1.]\n",
      " [1.]\n",
      " [1.]\n",
      " [1.]\n",
      " [1.]\n",
      " [0.]\n",
      " [1.]\n",
      " [0.]\n",
      " [0.]\n",
      " [0.]\n",
      " [0.]\n",
      " [1.]\n",
      " [1.]\n",
      " [1.]\n",
      " [0.]\n",
      " [1.]\n",
      " [1.]\n",
      " [1.]\n",
      " [0.]\n",
      " [0.]\n",
      " [0.]\n",
      " [1.]\n",
      " [1.]\n",
      " [1.]\n",
      " [1.]\n",
      " [0.]\n",
      " [1.]\n",
      " [1.]\n",
      " [0.]\n",
      " [1.]\n",
      " [1.]\n",
      " [0.]\n",
      " [1.]\n",
      " [1.]\n",
      " [0.]\n",
      " [1.]\n",
      " [0.]\n",
      " [1.]\n",
      " [1.]\n",
      " [1.]\n",
      " [0.]\n",
      " [0.]\n",
      " [1.]\n",
      " [1.]\n",
      " [1.]\n",
      " [0.]\n",
      " [0.]\n",
      " [1.]\n",
      " [1.]\n",
      " [0.]\n",
      " [1.]\n",
      " [0.]\n",
      " [1.]\n",
      " [1.]\n",
      " [1.]\n",
      " [1.]\n",
      " [0.]\n",
      " [0.]\n",
      " [1.]\n",
      " [1.]\n",
      " [1.]\n",
      " [1.]\n",
      " [0.]\n",
      " [1.]\n",
      " [1.]\n",
      " [1.]\n",
      " [0.]\n",
      " [0.]\n",
      " [1.]\n",
      " [1.]\n",
      " [1.]\n",
      " [1.]\n",
      " [0.]\n",
      " [0.]\n",
      " [1.]\n",
      " [1.]\n",
      " [0.]\n",
      " [1.]\n",
      " [0.]\n",
      " [0.]\n",
      " [0.]\n",
      " [1.]\n",
      " [1.]\n",
      " [0.]\n",
      " [1.]\n",
      " [0.]\n",
      " [1.]\n",
      " [1.]\n",
      " [1.]\n",
      " [1.]\n",
      " [1.]\n",
      " [1.]\n",
      " [0.]\n",
      " [1.]\n",
      " [1.]\n",
      " [1.]\n",
      " [1.]\n",
      " [1.]\n",
      " [1.]\n",
      " [1.]\n",
      " [1.]\n",
      " [1.]\n",
      " [0.]\n",
      " [0.]\n",
      " [0.]\n",
      " [1.]\n",
      " [0.]\n",
      " [1.]\n",
      " [0.]\n",
      " [0.]\n",
      " [1.]\n",
      " [1.]\n",
      " [1.]\n",
      " [1.]\n",
      " [1.]\n",
      " [1.]\n",
      " [1.]\n",
      " [0.]\n",
      " [0.]\n",
      " [0.]\n",
      " [0.]\n",
      " [1.]\n",
      " [1.]\n",
      " [1.]\n",
      " [1.]\n",
      " [0.]\n",
      " [1.]\n",
      " [1.]\n",
      " [0.]\n",
      " [1.]\n",
      " [0.]\n",
      " [1.]\n",
      " [1.]\n",
      " [1.]\n",
      " [1.]\n",
      " [1.]\n",
      " [1.]\n",
      " [1.]\n",
      " [0.]\n",
      " [1.]\n",
      " [1.]]\n"
     ]
    }
   ],
   "source": [
    "set_gpu_usage(ratio=0.9,gpu=0)\n",
    "\n",
    "datapath = os.getcwd() + '/datasets'\n",
    "\n",
    "\n",
    "# train,x_min,x_max = data_loader(datapath)\n",
    "# x_train,x_test = get_train_test(train,train_ratio)\n",
    "DataFiles={'normal':'d00_te.dat','fault1':'d01_te.dat','fault2':'d02_te.dat',\n",
    "           'fault3':'d03_te.dat','fault4':'d04_te.dat','fault5':'d05_te.dat',\n",
    "           'fault6':'d06_te.dat','fault7':'d07_te.dat','fault8':'d08_te.dat',\n",
    "           'fault9':'d09_te.dat','fault10':'d10_te.dat','fault11':'d11_te.dat',\n",
    "           'fault12':'d12_te.dat','fault13':'d13_te.dat','fault14':'d14_te.dat',\n",
    "           'fault15':'d15_te.dat','fault16':'d16_te.dat','fault17':'d17_te.dat',\n",
    "           'fault18':'d18_te.dat','fault19':'d19_te.dat','fault20':'d20_te.dat',\n",
    "           'fault21':'d21_te.dat'}\n",
    "\n",
    "datapath = os.getcwd() + '/datasets/data_960'\n",
    "xx=np.genfromtxt(datapath + '/d00.dat',dtype=np.float32)\n",
    "yy=np.genfromtxt(datapath + '/d00_te.dat',dtype=np.float32)\n",
    "fault_1_path = datapath + '/d01_te.dat'\n",
    "fault = np.genfromtxt(fault_1_path,dtype=None)\n",
    "fault = fault[160:960]\n",
    "np.random.seed(0)\n",
    "np.random.shuffle(fault)\n",
    "#for j in range(2,22):\n",
    "    #if j < 10:\n",
    "        #test_path = datapath + '/d0' + str(j) + '_te.dat'\n",
    "    #else:  \n",
    "        #test_path = datapath + '/d' + str(j) + '_te.dat'\n",
    "    #fault_data = np.genfromtxt(test_path,dtype = 'float32')\n",
    "    #fault_data = fault_data[160:960]\n",
    "    #np.random.seed(0)\n",
    "    #np.random.shuffle(fault_data)\n",
    "    #fault_data = fault_data[0:10]\n",
    "    #fault = np.row_stack((fault,fault_data))\n",
    "x_fault = fault\n",
    "\n",
    "train_data=np.row_stack((xx.T,yy,x_fault))   #将数据矩阵x和y合成新的数据矩阵用于训练\n",
    "#np.random.shuffle(train_data)\n",
    "n=train_data.shape[0]\n",
    "y=np.zeros((n,1))\n",
    "\n",
    "for i in range(1,n):\n",
    "    if i<1460:\n",
    "        y[i,0]=1\n",
    "    else:\n",
    "        y[i,0]=0\n",
    "\n",
    "data_ratio=1\n",
    "X_Normal=train_data\n",
    "train_data_min=np.min(train_data,axis=0)\n",
    "train_data_max=np.max(train_data,axis=0)\n",
    "Data_normalization=(train_data-train_data_min)/(train_data_max-train_data_min)\n",
    "#print(X_Normal.shape[0])\n",
    "index=np.arange(X_Normal.shape[0])\n",
    "np.random.shuffle(index)\n",
    "data_size=int(train_data.shape[0]*data_ratio)\n",
    "X_normal=X_Normal[index[:data_size]]\n",
    "Y=y[index[:data_size]]\n",
    "#print(X_normal.shape[0])\n",
    "X_min=np.min(X_normal,axis=0)\n",
    "X_max=np.max(X_normal,axis=0)\n",
    "#训练集大小\n",
    "X=X_normal\n",
    "train_ratio=0.9\n",
    "X_normalization=(X-X_min)/(X_max-X_min)  \n",
    "#np.random.shuffle(X_normalization)\n",
    "indices=np.arange(X.shape[0])\n",
    "np.random.shuffle(indices)\n",
    "train_size=int(X.shape[0]*train_ratio)    \n",
    "x_train=X_normalization[indices[:train_size]]\n",
    "x_test=X_normalization[indices[train_size:]]\n",
    "y_train=Y[indices[:train_size]]\n",
    "y_test=Y[indices[train_size:]]\n",
    "\n",
    "print('the shape of x_train:')\n",
    "print(np.shape(x_train))\n",
    "print('the shape of x_test:')\n",
    "print(np.shape(x_test))\n",
    "print('the shape of y_train:')\n",
    "print(np.shape(x_train))\n",
    "print('the shape of y_test:')\n",
    "print(np.shape(x_test))\n",
    "print(y_train[0:500])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Model config"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/home/zhangj/anaconda2/envs/Python35/lib/python3.5/site-packages/ipykernel_launcher.py:3: UserWarning: The `merge` function is deprecated and will be removed after 08/2017. Use instead layers from `keras.layers.merge`, e.g. `add`, `concatenate`, etc.\n",
      "  This is separate from the ipykernel package so we can avoid doing imports until\n",
      "/home/zhangj/anaconda2/envs/Python35/lib/python3.5/site-packages/keras/legacy/layers.py:464: UserWarning: The `Merge` layer is deprecated and will be removed after 08/2017. Use instead layers from `keras.layers.merge`, e.g. `add`, `concatenate`, etc.\n",
      "  name=name)\n",
      "/home/zhangj/anaconda2/envs/Python35/lib/python3.5/site-packages/ipykernel_launcher.py:7: UserWarning: The `merge` function is deprecated and will be removed after 08/2017. Use instead layers from `keras.layers.merge`, e.g. `add`, `concatenate`, etc.\n",
      "  import sys\n"
     ]
    }
   ],
   "source": [
    "X=Input(shape=[x_train.shape[1]])\n",
    "cond=Input(shape=[y_train.shape[1]])\n",
    "inputs=merge([X,cond],mode='concat',concat_axis=1)\n",
    "h1=Dense(200,activation='relu')(inputs)\n",
    "h2=Dense(200,activation='relu')(h1)\n",
    "z=Dense(latent_dim)(h2)\n",
    "z_cond=merge([z,cond],mode='concat',concat_axis=1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [],
   "source": [
    "decoder_hidden1=Dense(200,activation='relu')\n",
    "decoder_hidden2=Dense(200,activation='relu')\n",
    "decoder_out=Dense(52,activation='sigmoid')\n",
    "\n",
    "h_p1=decoder_hidden1(z_cond)\n",
    "h_p2=decoder_hidden2(h_p1)\n",
    "outputs=decoder_out(h_p2)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [],
   "source": [
    "autoencoder=Model([X,cond],outputs)\n",
    "encoder=Model([X,cond],z)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/home/zhangj/anaconda2/envs/Python35/lib/python3.5/site-packages/ipykernel_launcher.py:3: UserWarning: The `merge` function is deprecated and will be removed after 08/2017. Use instead layers from `keras.layers.merge`, e.g. `add`, `concatenate`, etc.\n",
      "  This is separate from the ipykernel package so we can avoid doing imports until\n",
      "/home/zhangj/anaconda2/envs/Python35/lib/python3.5/site-packages/keras/legacy/layers.py:464: UserWarning: The `Merge` layer is deprecated and will be removed after 08/2017. Use instead layers from `keras.layers.merge`, e.g. `add`, `concatenate`, etc.\n",
      "  name=name)\n"
     ]
    }
   ],
   "source": [
    "d_cond=Input(shape=[y_train.shape[1]])\n",
    "d_z=Input(shape=[latent_dim])\n",
    "d_inputs=merge([d_z,d_cond],mode='concat',concat_axis=1)\n",
    "d_h1=decoder_hidden1(d_inputs)\n",
    "d_h2=decoder_hidden2(d_h1)\n",
    "d_out=decoder_out(d_h2)\n",
    "decoder=Model([d_z,d_cond],d_out)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [],
   "source": [
    "def build_discriminator(input,output_dim=2):\n",
    "    h=Dense(50)(input)\n",
    "    h=Activation('relu')(h)\n",
    "    h=Dense(50)(h)\n",
    "    h=Activation('relu')(h)\n",
    "    h=Dense(1)(h)\n",
    "    h=Activation('sigmoid',name='discriminator_output')(h)\n",
    "    return h\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [],
   "source": [
    "discriminator_hidden1=Dense(50,activation='relu')\n",
    "discriminator_hidden2=Dense(50,activation='relu')\n",
    "discriminator_out=Dense(1,activation='sigmoid')\n",
    "\n",
    "d_p=discriminator_hidden1(z)\n",
    "labels=discriminator_out(d_p)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [],
   "source": [
    "input_z=Input(shape=[latent_dim],name='z_input')\n",
    "#D_cond=Input(shape=[y_train.shape[1]])\n",
    "#D_inputs=merge([input_z,D_cond],mode='concat',concat_axis=1)\n",
    "D_h1=discriminator_hidden1(input_z)\n",
    "D_h2=discriminator_hidden2(D_h1)\n",
    "D_out=discriminator_out(D_h2)\n",
    "\n",
    "discriminator=Model(input_z,D_out)\n",
    "G_out=discriminator(z)\n",
    "discriminator_x=Model([X,cond],G_out)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Target Distribution p(z)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [],
   "source": [
    "\n",
    "def gaussian_mixture(batch_size, n_dim=2, n_labels=2,\n",
    "                     x_var=0.5, y_var=0.1, label_indices=None):\n",
    "    # borrow from:\n",
    "    # https://github.com/nicklhy/AdversarialAutoEncoder/blob/master/data_factory.py#L40\n",
    "    if n_dim % 2 != 0:\n",
    "        raise Exception(\"n_dim must be a multiple of 2.\")\n",
    "\n",
    "    def sample(x, y, label, n_labels):\n",
    "        shift = 1.4\n",
    "        if label >= n_labels:\n",
    "            label =  np.random.randint(0, n_labels)\n",
    "        r = 2.0 * np.pi / float(n_labels) * float(label)\n",
    "        new_x = x * cos(r) - y * sin(r)\n",
    "        new_y = x * sin(r) + y * cos(r)\n",
    "        new_x += shift * cos(r)\n",
    "        new_y += shift * sin(r)\n",
    "        return np.array([new_x, new_y]).reshape((2,))\n",
    "\n",
    "    x = np.random.normal(0, x_var, (batch_size, n_dim // 2))\n",
    "    y = np.random.normal(0, y_var, (batch_size, n_dim // 2))\n",
    "    z = np.empty((batch_size, n_dim), dtype=np.float32)\n",
    "    for batch in range(batch_size):\n",
    "        for zi in range(n_dim // 2):\n",
    "            if label_indices is not None:\n",
    "                z[batch, zi*2:zi*2+2] = sample(x[batch, zi], y[batch, zi], label_indices[batch], n_labels)\n",
    "            else:\n",
    "                z[batch, zi*2:zi*2+2] = sample(x[batch, zi], y[batch, zi], np.random.randint(0, n_labels), n_labels)\n",
    "\n",
    "    return z\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [],
   "source": [
    "z=gaussian_mixture(20000)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<matplotlib.collections.PathCollection at 0x7ff61858ed30>"
      ]
     },
     "execution_count": 14,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "xs, ys = np.array(z).T\n",
    "plt.figure()\n",
    "plt.scatter(xs, ys, s=2, color='b')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [],
   "source": [
    "def reconstruction_loss():        \n",
    "    if recon_obj_type == 'QE':\n",
    "        return 'mean_squared_error'\n",
    "    elif recon_obj_type == 'CE':\n",
    "        return 'binary_crossentropy'\n",
    "def rs_loss(y_true,y_pred):\n",
    "    recon=K.sum(K.binary_crossentropy(y_pred,y_true),axis=1)\n",
    "    return recon\n",
    "# build model\n",
    "autoencoder.compile(optimizer=opt_rec,\n",
    "                     loss='mean_squared_error')\n",
    "\n",
    "discriminator.compile(optimizer=opt_disc,\n",
    "                     loss='binary_crossentropy')\n",
    "\n",
    "discriminator_x.compile(optimizer=opt_gen,\n",
    "                     loss='binary_crossentropy')\n",
    "\n",
    "\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Train the model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1 of 20\n",
      "\r",
      "  0/203 [..............................] - ETA: 0s"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/home/zhangj/anaconda2/envs/Python35/lib/python3.5/site-packages/keras/engine/training.py:973: UserWarning: Discrepancy between trainable weights and collected trainable weights, did you set `model.trainable` without calling `model.compile` after ?\n",
      "  'Discrepancy between trainable weights and collected trainable'\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "201/203 [============================>.] - ETA: 0s\n",
      "Testing for epoch 1:\n",
      "latent z\n",
      "2\n",
      "          generative_loss  recon_loss       disc_loss       \n",
      "train     1.367            0.028            0.403           \n",
      "test      1.687            0.026            0.380           \n",
      "Epoch 2 of 20\n",
      "199/203 [============================>.] - ETA: 0s\n",
      "Testing for epoch 2:\n",
      "latent z\n",
      "2\n",
      "          generative_loss  recon_loss       disc_loss       \n",
      "train     1.796            0.026            0.328           \n",
      "test      1.862            0.025            0.335           \n",
      "Epoch 3 of 20\n",
      "199/203 [============================>.] - ETA: 0s\n",
      "Testing for epoch 3:\n",
      "latent z\n",
      "2\n",
      "          generative_loss  recon_loss       disc_loss       \n",
      "train     1.940            0.025            0.299           \n",
      "test      1.984            0.024            0.322           \n",
      "Epoch 4 of 20\n",
      "201/203 [============================>.] - ETA: 0s\n",
      "Testing for epoch 4:\n",
      "latent z\n",
      "2\n",
      "          generative_loss  recon_loss       disc_loss       \n",
      "train     2.032            0.024            0.284           \n",
      "test      2.055            0.023            0.301           \n",
      "Epoch 5 of 20\n",
      "201/203 [============================>.] - ETA: 0s\n",
      "Testing for epoch 5:\n",
      "latent z\n",
      "2\n",
      "          generative_loss  recon_loss       disc_loss       \n",
      "train     2.104            0.023            0.269           \n",
      "test      2.120            0.023            0.299           \n",
      "Epoch 6 of 20\n",
      "198/203 [============================>.] - ETA: 0s\n",
      "Testing for epoch 6:\n",
      "latent z\n",
      "2\n",
      "          generative_loss  recon_loss       disc_loss       \n",
      "train     2.155            0.023            0.260           \n",
      "test      2.165            0.023            0.280           \n",
      "Epoch 7 of 20\n",
      "197/203 [============================>.] - ETA: 0s\n",
      "Testing for epoch 7:\n",
      "latent z\n",
      "2\n",
      "          generative_loss  recon_loss       disc_loss       \n",
      "train     2.190            0.023            0.255           \n",
      "test      2.190            0.022            0.277           \n",
      "Epoch 8 of 20\n",
      "200/203 [============================>.] - ETA: 0s\n",
      "Testing for epoch 8:\n",
      "latent z\n",
      "2\n",
      "          generative_loss  recon_loss       disc_loss       \n",
      "train     2.221            0.023            0.247           \n",
      "test      2.233            0.022            0.273           \n",
      "Epoch 9 of 20\n",
      "202/203 [============================>.] - ETA: 0s\n",
      "Testing for epoch 9:\n",
      "latent z\n",
      "2\n",
      "          generative_loss  recon_loss       disc_loss       \n",
      "train     2.248            0.022            0.247           \n",
      "test      2.244            0.022            0.259           \n",
      "Epoch 10 of 20\n",
      "202/203 [============================>.] - ETA: 0s\n",
      "Testing for epoch 10:\n",
      "latent z\n",
      "2\n",
      "          generative_loss  recon_loss       disc_loss       \n",
      "train     2.273            0.022            0.237           \n",
      "test      2.273            0.022            0.258           \n",
      "Epoch 11 of 20\n",
      "202/203 [============================>.] - ETA: 0s\n",
      "Testing for epoch 11:\n",
      "latent z\n",
      "2\n",
      "          generative_loss  recon_loss       disc_loss       \n",
      "train     2.297            0.022            0.233           \n",
      "test      2.292            0.022            0.253           \n",
      "Epoch 12 of 20\n",
      "198/203 [============================>.] - ETA: 0s\n",
      "Testing for epoch 12:\n",
      "latent z\n",
      "2\n",
      "          generative_loss  recon_loss       disc_loss       \n",
      "train     2.322            0.022            0.226           \n",
      "test      2.317            0.022            0.245           \n",
      "Epoch 13 of 20\n",
      "199/203 [============================>.] - ETA: 0s\n",
      "Testing for epoch 13:\n",
      "latent z\n",
      "2\n",
      "          generative_loss  recon_loss       disc_loss       \n",
      "train     2.336            0.022            0.227           \n",
      "test      2.330            0.022            0.246           \n",
      "Epoch 14 of 20\n",
      "199/203 [============================>.] - ETA: 0s\n",
      "Testing for epoch 14:\n",
      "latent z\n",
      "2\n",
      "          generative_loss  recon_loss       disc_loss       \n",
      "train     2.353            0.022            0.223           \n",
      "test      2.347            0.022            0.247           \n",
      "Epoch 15 of 20\n",
      "197/203 [============================>.] - ETA: 0s\n",
      "Testing for epoch 15:\n",
      "latent z\n",
      "2\n",
      "          generative_loss  recon_loss       disc_loss       \n",
      "train     2.365            0.022            0.223           \n",
      "test      2.354            0.021            0.234           \n",
      "Epoch 16 of 20\n",
      "201/203 [============================>.] - ETA: 0s\n",
      "Testing for epoch 16:\n",
      "latent z\n",
      "2\n",
      "          generative_loss  recon_loss       disc_loss       \n",
      "train     2.374            0.022            0.217           \n",
      "test      2.372            0.021            0.230           \n",
      "Epoch 17 of 20\n",
      "201/203 [============================>.] - ETA: 0s\n",
      "Testing for epoch 17:\n",
      "latent z\n",
      "2\n",
      "          generative_loss  recon_loss       disc_loss       \n",
      "train     2.391            0.022            0.217           \n",
      "test      2.383            0.021            0.241           \n",
      "Epoch 18 of 20\n",
      "200/203 [============================>.] - ETA: 0s\n",
      "Testing for epoch 18:\n",
      "latent z\n",
      "2\n",
      "          generative_loss  recon_loss       disc_loss       \n",
      "train     2.404            0.022            0.211           \n",
      "test      2.399            0.021            0.229           \n",
      "Epoch 19 of 20\n",
      "199/203 [============================>.] - ETA: 0s\n",
      "Testing for epoch 19:\n",
      "latent z\n",
      "2\n",
      "          generative_loss  recon_loss       disc_loss       \n",
      "train     2.423            0.022            0.208           \n",
      "test      2.416            0.021            0.232           \n",
      "Epoch 20 of 20\n",
      "198/203 [============================>.] - ETA: 0s\n",
      "Testing for epoch 20:\n",
      "latent z\n",
      "2\n",
      "          generative_loss  recon_loss       disc_loss       \n",
      "train     2.435            0.021            0.207           \n",
      "test      2.428            0.021            0.232           \n"
     ]
    }
   ],
   "source": [
    "# In[57]:\n",
    "nb_batches=int(x_train.shape[0]/batch_size)\n",
    "\n",
    "# train\n",
    "train_history = defaultdict(list)\n",
    "test_history = defaultdict(list)\n",
    "\n",
    "with open(path+'/train.log','w') as f:\n",
    "    f.write('train_begin\\n' )\n",
    "    \n",
    "for epoch in range(num_epochs):\n",
    "    print('Epoch {} of {}'.format(epoch+1,num_epochs))\n",
    "    progress_bar=Progbar(target=nb_batches)\n",
    "\n",
    "    epoch_gen_loss=[]\n",
    "    epoch_disc_loss=[]\n",
    "    epoch_rec_loss=[]\n",
    "\n",
    "    for index in range(nb_batches):\n",
    "        progress_bar.update(index)\n",
    "        # get a batch of real images\n",
    "        x_batch=x_train[index*batch_size:(index+1)*batch_size]\n",
    "        y_batch=y_train[index*batch_size:(index+1)*batch_size]\n",
    "        \n",
    "        z_real=gaussian_mixture(batch_size)\n",
    "        for i in range(batch_size):\n",
    "            if y_batch[i,0]==1:\n",
    "                    z_real[i,0]=np.abs(z_real[i,0])\n",
    "            else:\n",
    "                    z_real[i,0]=-1*np.abs(z_real[i,0])\n",
    "        #z_real=z_real[np.newaxis,:]\n",
    "        make_trainable(autoencoder,True)\n",
    "        _rec=autoencoder.train_on_batch([x_batch,y_batch],x_batch)\n",
    "        epoch_rec_loss.append(_rec)\n",
    "        \n",
    "        # train the discriminator \n",
    "        make_trainable(autoencoder,False)          \n",
    "        make_trainable(discriminator,True)\n",
    "        \n",
    "        z_fake=encoder.predict([x_batch,y_batch])\n",
    "        z_batch=np.concatenate([z_real,z_fake],axis=0)\n",
    "        labels_batch=np.concatenate([np.zeros((batch_size,1)),np.ones((batch_size,1))],axis=0)\n",
    "        _disc=discriminator.train_on_batch([z_batch],labels_batch)\n",
    "        epoch_disc_loss.append(_disc)\n",
    "\n",
    "        # train the generator\n",
    "        make_trainable(discriminator,False)\n",
    "        make_trainable(autoencoder,True)            \n",
    "        \n",
    "        labels_x_batch=np.zeros((batch_size,1))\n",
    "        _gen=discriminator_x.train_on_batch([x_batch,y_batch],labels_x_batch)\n",
    "        epoch_gen_loss.append(_gen)\n",
    "\n",
    "    print('\\nTesting for epoch {}:'.format(epoch+1))\n",
    "\n",
    "    # evaluate the testing loss\n",
    "    n_test=x_test.shape[0]\n",
    "    z_real_test=gaussian_mixture(n_test)\n",
    "    for i in range(n_test):\n",
    "        if y_test[i,0]==1:\n",
    "                z_real_test[i,0]=np.abs(z_real_test[i,0])\n",
    "        else:\n",
    "                z_real_test[i,0]=-1*np.abs(z_real_test[i,0])\n",
    "    #z_real_test=gaussian_mixture(n_test)\n",
    "    z_fake_test=encoder.predict([x_test,y_test])\n",
    "    \n",
    "    labels_test=np.concatenate([np.zeros((n_test,1)),np.ones((n_test,1))],axis=0)    \n",
    "    z_test=np.concatenate([z_real_test,z_fake_test],axis=0)    \n",
    "    \n",
    "    test_rec_loss  = autoencoder.test_on_batch([x_test,y_test],x_test)\n",
    "    test_disc_loss = discriminator.test_on_batch([z_test],labels_test)\n",
    "    test_gen_loss  = discriminator_x.test_on_batch([x_test,y_test],np.zeros((n_test,1)))\n",
    "\n",
    "    train_rec_loss  = np.mean(np.array(epoch_rec_loss),axis=0)\n",
    "    train_disc_loss = np.mean(np.array(epoch_disc_loss), axis=0)\n",
    "    train_gen_loss  = np.mean(np.array(epoch_gen_loss),axis=0)\n",
    "\n",
    "    train_history['generative'].append(train_gen_loss)\n",
    "    train_history['discriminator'].append(train_disc_loss)\n",
    "    train_history['reconstruction'].append(train_rec_loss)\n",
    "\n",
    "    test_history['generative'].append(test_gen_loss)\n",
    "    test_history['discriminator'].append(test_disc_loss)\n",
    "    test_history['reconstruction'].append(test_rec_loss)\n",
    "\n",
    "    encoded_x=encoder.predict([x_test,y_test])\n",
    "    print('latent z')\n",
    "    print(latent_dim)\n",
    "    if (epoch %5==4) & (latent_dim==2):\n",
    "        _y=np.zeros((n_test,1))\n",
    "        #plot_latent_variable(path,encoded_x,_y,epoch=epoch)       \n",
    "    ROW_FMT_1='{0:8s}  {1:<15s}  {2:<15s}  {3:<15s} '\n",
    "    ROW_FMT_2='{0:8s}  {1:<15.3f}  {2:<15.3f}  {3:<15.3f} '\n",
    "    \n",
    "    str1=ROW_FMT_1.format('','generative_loss','recon_loss','disc_loss')\n",
    "    str2=ROW_FMT_2.format('train', train_history['generative'][-1],\n",
    "                     train_history['reconstruction'][-1],\n",
    "                     train_history['discriminator'][-1])\n",
    "    str3=ROW_FMT_2.format('test',  test_history['generative'][-1],\n",
    "                     test_history['reconstruction'][-1],\n",
    "                     test_history['discriminator'][-1])\n",
    "    \n",
    "    print(str1+'\\n'+str2+'\\n'+str3)\n",
    "    with open(path+'/train.log','a') as f:\n",
    "        f.write('\\nepoch{}:\\n'.format(epoch))\n",
    "        f.write(str1+'\\n'+str2+'\\n'+str3)\n",
    "    # save weights every 5 epoch\n",
    "    if epoch%5==4:\n",
    "        autoencoder.save_weights(path+'/params_combined_model_epoch_{0:03d}.hdf5'.format(epoch),True)\n",
    "\n",
    "pickle.dump({'train':train_history,'test': test_history},open(path+'/aae_history.pkl','wb'))\n",
    "# sample from learned distribution\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Visualization of latent space"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[1. 0. 0. ... 0. 0. 1.]\n",
      "[[-3.705173    0.2671201 ]\n",
      " [-2.7167888   0.11469984]\n",
      " [-3.061196    0.2254113 ]\n",
      " ...\n",
      " [-2.7466662   0.27055952]\n",
      " [-3.0460238   0.2625288 ]\n",
      " [-3.500211    0.28139824]]\n"
     ]
    }
   ],
   "source": [
    "latent_z=encoder.predict([x_train,y_train])\n",
    "print(y_train[:,0])\n",
    "latent_z_test=encoder.predict([x_test,y_test])\n",
    "print(latent_z)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### training data"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<matplotlib.collections.PathCollection at 0x7ff6081bc208>"
      ]
     },
     "execution_count": 18,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "xs, ys = np.array(latent_z).T\n",
    "plt.figure()\n",
    "# plt.title(\"round {}: {} in latent space\".format(model.step, title))\n",
    "kwargs = {'alpha': 0.8}\n",
    "label_list=[int(x) for x in y_train[:,0]]\n",
    "classes = set(label_list)\n",
    "if classes:\n",
    "    colormap = plt.cm.rainbow(np.linspace(0, 1, len(classes)))\n",
    "    kwargs['c'] = [colormap[i] for i in label_list]\n",
    "    # make room for legend\n",
    "    ax = plt.subplot(111, aspect='equal')\n",
    "    box = ax.get_position()\n",
    "    ax.set_position([box.x0, box.y0, box.width * 0.8, box.height])\n",
    "    handles = [mpatches.Circle((0,0), label=class_, color=colormap[i])\n",
    "              for i, class_ in enumerate(classes)]\n",
    "    ax.legend(handles=handles, shadow=True, bbox_to_anchor=(1.05, 0.45),\n",
    "              fancybox=True, loc='center left')\n",
    "\n",
    "plt.scatter(xs, ys, s=2, **kwargs)\n",
    "\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### testing data"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<matplotlib.collections.PathCollection at 0x7ff60829f898>"
      ]
     },
     "execution_count": 19,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "xs, ys = np.array(latent_z_test).T\n",
    "\n",
    "plt.figure()\n",
    "# plt.title(\"round {}: {} in latent space\".format(model.step, title))\n",
    "kwargs = {'alpha': 0.8}\n",
    "label_list=[int(x) for x in y_test[:,0]]\n",
    "classes = set(label_list)\n",
    "if classes:\n",
    "    colormap = plt.cm.rainbow(np.linspace(0, 1, len(classes)))\n",
    "    kwargs['c'] = [colormap[i] for i in label_list]\n",
    "\n",
    "    # make room for legend\n",
    "    ax = plt.subplot(111, aspect='equal')\n",
    "    box = ax.get_position()\n",
    "    ax.set_position([box.x0, box.y0, box.width * 0.8, box.height])\n",
    "    handles = [mpatches.Circle((0,0), label=class_, color=colormap[i])\n",
    "              for i, class_ in enumerate(classes)]\n",
    "    ax.legend(handles=handles, shadow=True, bbox_to_anchor=(1.05, 0.45),\n",
    "              fancybox=True, loc='center left')\n",
    "\n",
    "plt.scatter(xs, ys, s=2, **kwargs)\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Sample from p(z) and evaluate"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Reconstruction:\n",
      "mean: 0.0\n",
      "variance: 1.0974776653869511\n",
      "Generation:\n",
      "mean: 0.5258402606643879\n",
      "variance: 0.0025428343\n"
     ]
    }
   ],
   "source": [
    "new_size=50000\n",
    "c=np.ones((new_size,1))\n",
    "d=np.zeros((new_size,1))\n",
    "z_sample_1=gaussian_mixture(new_size)\n",
    "z_sample_2=gaussian_mixture(new_size)\n",
    "for i in range(new_size):\n",
    "    z_sample_1[i,0]=np.abs(z_sample_1[i,0])\n",
    "    z_sample_2[i,0]=-1*np.abs(z_sample_2[i,0])\n",
    "new_sample_N=decoder.predict([z_sample_1,c])\n",
    "new_sample_F=decoder.predict([z_sample_2,d])\n",
    "sess=tf.Session()\n",
    "sess.run(tf.global_variables_initializer())\n",
    "x_normal = x_train[0:1314]\n",
    "x_mean=np.mean(x_normal,axis=0)\n",
    "x_hat = autoencoder.predict([x_normal,c])\n",
    "def loss(a,b):\n",
    "    result = np.dot((a-b),(a-b))\n",
    "    return result\n",
    "def cal_var(data,mean):\n",
    "    total_var = []\n",
    "    for i in range(data.shape[0]):\n",
    "        var = loss(data[i],mean)\n",
    "        total_var.append(var)\n",
    "    return total_var\n",
    "new_sample_N1=new_sample_N[0:1314]\n",
    "new_sample_mean=np.mean(new_sample_N1,axis=0)\n",
    "print('Reconstruction:')\n",
    "print('mean:',loss(x_mean,x_mean))\n",
    "print('variance:',np.mean(cal_var(x_normal,x_mean)))\n",
    "print('Generation:')\n",
    "print('mean:',loss(new_sample_mean,x_mean))\n",
    "print('variance:',np.mean(cal_var(new_sample_N1,new_sample_mean)))\n",
    "np.savetxt(path + '/sample_fault1.txt',new_sample_N)\n",
    "\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### sampling distribution"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "metadata": {},
   "outputs": [],
   "source": [
    "label_list=[]\n",
    "for i in range(2*new_size):\n",
    "    if i<new_size:\n",
    "        label_list.append(1)\n",
    "    else:\n",
    "        label_list.append(0)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<matplotlib.collections.PathCollection at 0x7ff608370f28>"
      ]
     },
     "execution_count": 25,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "sample_z=np.concatenate([z_sample_1,z_sample_2],axis=0)\n",
    "xs, ys = np.array(sample_z).T\n",
    "\n",
    "plt.figure()\n",
    "# plt.title(\"round {}: {} in latent space\".format(model.step, title))\n",
    "kwargs = {'alpha': 0.8}\n",
    "#label_list=[int(x) for x in y_train[:,0]]\n",
    "classes = set(label_list)\n",
    "if classes:\n",
    "    colormap = plt.cm.rainbow(np.linspace(0, 1, len(classes)))\n",
    "    kwargs['c'] = [colormap[i] for i in label_list]\n",
    "\n",
    "    # make room for legend\n",
    "    ax = plt.subplot(111, aspect='equal')\n",
    "    box = ax.get_position()\n",
    "    ax.set_position([box.x0, box.y0, box.width * 0.8, box.height])\n",
    "    handles = [mpatches.Circle((0,0), label=class_, color=colormap[i])\n",
    "              for i, class_ in enumerate(classes)]\n",
    "    ax.legend(handles=handles, shadow=True, bbox_to_anchor=(1.05, 0.45),\n",
    "              fancybox=True, loc='center left')\n",
    "\n",
    "plt.scatter(xs, ys, s=2, **kwargs)\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "metadata": {},
   "outputs": [],
   "source": [
    "x_G=new_sample_N[0:5000]\n",
    "x_F=new_sample_F[0:5000]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "metadata": {},
   "outputs": [],
   "source": [
    "g=np.ones((5000,1))\n",
    "f=np.zeros((5000,1))\n",
    "x_g=encoder.predict([x_G,g])\n",
    "x_f=encoder.predict([x_F,f])\n",
    "sample=np.concatenate([x_g,x_f],axis=0)\n",
    "label_list=[]\n",
    "for i in range(10000):\n",
    "    if i<5000:\n",
    "        label_list.append(1)\n",
    "    else:\n",
    "        label_list.append(0)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<matplotlib.collections.PathCollection at 0x7ff54c084860>"
      ]
     },
     "execution_count": 28,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "xs, ys = np.array(sample).T\n",
    "\n",
    "plt.figure()\n",
    "# plt.title(\"round {}: {} in latent space\".format(model.step, title))\n",
    "kwargs = {'alpha': 0.8}\n",
    "#label_list=[int(x) for x in y_train[:,0]]\n",
    "classes = set(label_list)\n",
    "if classes:\n",
    "    colormap = plt.cm.rainbow(np.linspace(0, 1, len(classes)))\n",
    "    kwargs['c'] = [colormap[i] for i in label_list]\n",
    "\n",
    "    # make room for legend\n",
    "    ax = plt.subplot(111, aspect='equal')\n",
    "    box = ax.get_position()\n",
    "    ax.set_position([box.x0, box.y0, box.width * 0.8, box.height])\n",
    "    handles = [mpatches.Circle((0,0), label=class_, color=colormap[i])\n",
    "              for i, class_ in enumerate(classes)]\n",
    "    ax.legend(handles=handles, shadow=True, bbox_to_anchor=(1.05, 0.45),\n",
    "              fancybox=True, loc='center left')\n",
    "\n",
    "plt.scatter(xs, ys, s=1, **kwargs)\n",
    "#ax.set_xlim([-3.5, -2])\n",
    "#ax.set_ylim([0.2, 0.25])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 30,
   "metadata": {},
   "outputs": [],
   "source": [
    "from sklearn.decomposition import PCA\n",
    "pca=PCA(n_components=2)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 31,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<matplotlib.collections.PathCollection at 0x7ff5146bcfd0>"
      ]
     },
     "execution_count": 31,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 600x400 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "x_G=pca.fit_transform(x_G[0:1314])\n",
    "x_F=pca.fit_transform(x_F[0:1314])\n",
    "x_N=pca.fit_transform(x_normal)\n",
    "matplotlib.rcParams['xtick.direction'] = 'in'\n",
    "matplotlib.rcParams['ytick.direction'] = 'in'\n",
    "\n",
    "plt.rcParams['figure.dpi'] = 100\n",
    "plt.scatter(x_N[:,0],x_N[:,1],color='r',s=2)\n",
    "plt.scatter(x_G[:,0],x_G[:,1],color='b',s=2)\n",
    "plt.scatter(x_F[:,0],x_F[:,1],color='g',s=2)\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.5.4"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}