MC_with_edward.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "import edward as ed\n",
    "import matplotlib.pyplot as plt\n",
    "import numpy as np\n",
    "import seaborn as sns\n",
    "import tensorflow as tf\n",
    "import mobilenet_v1\n",
    "CLASS_OF_INTEREST = 10\n",
    "sess = tf.Session()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Get some pretrained tensorflow network for our needs\n",
    "\n",
    "\n",
    "$y = f(x)$, where y --- prediction, $f(\\cdot)$ -- NN, x -- image(for example)"
   ]
  },
  {
   "cell_type": "raw",
   "metadata": {},
   "source": [
    "!wget http://download.tensorflow.org/models/mobilenet_v1_1.0_224_2017_06_14.tar.gz\n",
    "!mkdir -p net\n",
    "!tar -xvf mobilenet_v1_1.0_224_2017_06_14.tar.gz -C net"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "INFO:tensorflow:Restoring parameters from ./mobilenet_v1_1.0_224.ckpt\n"
     ]
    }
   ],
   "source": [
    "x = tf.Variable(tf.random_normal([224, 224, 3], stddev=0.35),name=\"input\")\n",
    "with tf.contrib.slim.arg_scope(mobilenet_v1.mobilenet_v1_arg_scope()):\n",
    "    logits, net = mobilenet_v1.mobilenet_v1(tf.expand_dims(x, axis=0), is_training=False, num_classes=1001)\n",
    "    \n",
    "init_fn = tf.contrib.slim.assign_from_checkpoint_fn('./mobilenet_v1_1.0_224.ckpt', \n",
    "                                                        tf.contrib.slim.get_model_variables())\n",
    "init_fn(sess)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Construct from this network probability dension function for Edwardlib\n",
    "\n",
    "1. We need density function\n",
    "\n",
    "2. Some how need to define density from classifier output (wow so heuristic). I.e. if we want to sample images that belong to class 1:\n",
    "\n",
    "$$p(x) = \\frac{1}{Z} \\left( f(x) - 1 + \\epsilon\\right)_+$$\n",
    "\n",
    "Or smth more smooth."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": [
    "from tensorflow.python.framework import dtypes\n",
    "from edward.models import RandomVariable\n",
    "from tensorflow.contrib.distributions import Distribution\n",
    "from tensorflow.python.ops.distributions import distribution\n",
    "EPS = 1e-20\n",
    "CONST = 50\n",
    "class NNRandomVariable(RandomVariable, Distribution):\n",
    "    def __init__(self, *args, **kwargs):\n",
    "        super(NNRandomVariable, self).__init__(*args, **kwargs)\n",
    "        self.eps = 0.1\n",
    "        \n",
    "    def _log_prob(self, value):\n",
    "        with tf.contrib.slim.arg_scope(mobilenet_v1.mobilenet_v1_arg_scope()):\n",
    "            logits, net = mobilenet_v1.mobilenet_v1(tf.expand_dims(value, axis=0), reuse=True,\n",
    "                                                    is_training=False, num_classes=1001)\n",
    "\n",
    "        init_fn = tf.contrib.slim.assign_from_checkpoint_fn('./mobilenet_v1_1.0_224.ckpt', \n",
    "                                                                tf.contrib.slim.get_model_variables())\n",
    "        init_fn(sess)\n",
    "        classifier = tf.sigmoid(logits)[:, CLASS_OF_INTEREST]\n",
    "        return -CONST * (classifier - 1 + self.eps)\n",
    "\n",
    "    def _sample_n(self, n, seed=None):\n",
    "        return tf.random_normal(dtype=dtypes.float32, shape=[n, 224, 224, 3])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": [
    "NN = NNRandomVariable(dtype=dtypes.float32,\n",
    "                      validate_args=False,\n",
    "                      allow_nan_stats=True,\n",
    "                      reparameterization_type=distribution.NOT_REPARAMETERIZED)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "INFO:tensorflow:Restoring parameters from ./mobilenet_v1_1.0_224.ckpt\n"
     ]
    }
   ],
   "source": [
    "a = tf.Variable(tf.random_normal([224, 224, 3], stddev=0.35),name=\"a\")\n",
    "init = tf.global_variables_initializer()\n",
    "init.run(session=sess)\n",
    "res = tf.gradients(NN._log_prob(value=a), a)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": [
    "NN = ed.models.Normal(loc=tf.ones([224, 224, 3]), scale=2.0 * tf.ones([224, 224, 3]))\n",
    "empiric = ed.models.Empirical(params=tf.Variable(2. + tf.random_normal([500, 224, 224, 3])))\n",
    "init = tf.global_variables_initializer()\n",
    "init.run(session=sess)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {
    "collapsed": false,
    "scrolled": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "500\n",
      " 10/500 [  2%]                                ETA: 88s | Acceptance Rate: 0.000 \b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\n",
      "1.9997\n",
      "Loss -1.86162\n",
      "\n",
      " 20/500 [  4%] █                              ETA: 53s | Acceptance Rate: 0.000\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\n",
      "1.99962\n",
      "Loss -1.86162\n",
      "\n",
      " 30/500 [  6%] █                              ETA: 41s | Acceptance Rate: 0.000\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\n",
      "1.99954\n",
      "Loss -1.86162\n",
      "\n",
      " 40/500 [  8%] ██                             ETA: 35s | Acceptance Rate: 0.000\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\n",
      "1.99947\n",
      "Loss -1.86162\n",
      "\n",
      " 50/500 [ 10%] ███                            ETA: 31s | Acceptance Rate: 0.000\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\n",
      "1.99941\n",
      "Loss -1.86162\n",
      "\n",
      " 60/500 [ 12%] ███                            ETA: 28s | Acceptance Rate: 0.000\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\n",
      "1.99933\n",
      "Loss -1.86162\n",
      "\n",
      " 70/500 [ 14%] ████                           ETA: 26s | Acceptance Rate: 0.000\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\n",
      "1.99927\n",
      "Loss -1.86162\n",
      "\n",
      " 80/500 [ 16%] ████                           ETA: 25s | Acceptance Rate: 0.000\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\n",
      "1.99922\n",
      "Loss -1.86162\n",
      "\n",
      " 90/500 [ 18%] █████                          ETA: 24s | Acceptance Rate: 0.000\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\n",
      "1.99917\n",
      "Loss -1.86162\n",
      "\n",
      "100/500 [ 20%] ██████                         ETA: 22s | Acceptance Rate: 0.000\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\n",
      "1.9991\n",
      "Loss -1.86162\n",
      "\n",
      "110/500 [ 22%] ██████                         ETA: 21s | Acceptance Rate: 0.000\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\n",
      "1.99904\n",
      "Loss -1.86162\n",
      "\n",
      "120/500 [ 24%] ███████                        ETA: 20s | Acceptance Rate: 0.000\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\n",
      "1.99897\n",
      "Loss -1.86162\n",
      "\n",
      "130/500 [ 26%] ███████                        ETA: 20s | Acceptance Rate: 0.000\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\n",
      "1.99891\n",
      "Loss -1.86162\n",
      "\n",
      "140/500 [ 28%] ████████                       ETA: 19s | Acceptance Rate: 0.000\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\n",
      "1.99889\n",
      "Loss -1.86162\n",
      "\n",
      "150/500 [ 30%] █████████                      ETA: 19s | Acceptance Rate: 0.000\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\n",
      "1.99881\n",
      "Loss -1.86162\n",
      "\n",
      "160/500 [ 32%] █████████                      ETA: 18s | Acceptance Rate: 0.000\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\n",
      "1.99876\n",
      "Loss -1.86162\n",
      "\n",
      "170/500 [ 34%] ██████████                     ETA: 17s | Acceptance Rate: 0.000\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\n",
      "1.99867\n",
      "Loss -1.86162\n",
      "\n",
      "180/500 [ 36%] ██████████                     ETA: 17s | Acceptance Rate: 0.000\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\n",
      "1.99863\n",
      "Loss -1.86162\n",
      "\n",
      "190/500 [ 38%] ███████████                    ETA: 16s | Acceptance Rate: 0.000\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\n",
      "1.99858\n",
      "Loss -1.86162\n",
      "\n",
      "200/500 [ 40%] ████████████                   ETA: 15s | Acceptance Rate: 0.000\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\n",
      "1.99853\n",
      "Loss -1.86162\n",
      "\n",
      "210/500 [ 42%] ████████████                   ETA: 15s | Acceptance Rate: 0.000\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\n",
      "1.99843\n",
      "Loss -1.86162\n",
      "\n",
      "220/500 [ 44%] █████████████                  ETA: 14s | Acceptance Rate: 0.000\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\n",
      "1.99837\n",
      "Loss -1.86162\n",
      "\n",
      "230/500 [ 46%] █████████████                  ETA: 13s | Acceptance Rate: 0.000\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\n",
      "1.99833\n",
      "Loss -1.86162\n",
      "\n",
      "240/500 [ 48%] ██████████████                 ETA: 13s | Acceptance Rate: 0.000\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\n",
      "1.99826\n",
      "Loss -1.86162\n",
      "\n",
      "250/500 [ 50%] ███████████████                ETA: 12s | Acceptance Rate: 0.000\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\n",
      "1.9982\n",
      "Loss -1.86162\n",
      "\n",
      "260/500 [ 52%] ███████████████                ETA: 12s | Acceptance Rate: 0.000\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\n",
      "1.99813\n",
      "Loss -1.86162\n",
      "\n",
      "270/500 [ 54%] ████████████████               ETA: 11s | Acceptance Rate: 0.000\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\n",
      "1.99812\n",
      "Loss -1.86162\n",
      "\n",
      "280/500 [ 56%] ████████████████               ETA: 11s | Acceptance Rate: 0.000\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\n",
      "1.99806\n",
      "Loss -1.86162\n",
      "\n",
      "290/500 [ 57%] █████████████████              ETA: 10s | Acceptance Rate: 0.000\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\n",
      "1.99803\n",
      "Loss -1.86162\n",
      "\n",
      "300/500 [ 60%] ██████████████████             ETA: 10s | Acceptance Rate: 0.000\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\n",
      "1.99797\n",
      "Loss -1.86162\n",
      "\n",
      "310/500 [ 62%] ██████████████████             ETA: 9s | Acceptance Rate: 0.000 \b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\n",
      "1.99788\n",
      "Loss -1.86162\n",
      "\n",
      "320/500 [ 64%] ███████████████████            ETA: 9s | Acceptance Rate: 0.000\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\n",
      "1.99783\n",
      "Loss -1.86162\n",
      "\n",
      "330/500 [ 66%] ███████████████████            ETA: 8s | Acceptance Rate: 0.000\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\n",
      "1.9978\n",
      "Loss -1.86162\n",
      "\n",
      "340/500 [ 68%] ████████████████████           ETA: 8s | Acceptance Rate: 0.000\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\n",
      "1.99773\n",
      "Loss -1.86162\n",
      "\n",
      "350/500 [ 70%] █████████████████████          ETA: 7s | Acceptance Rate: 0.000\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\n",
      "1.99767\n",
      "Loss -1.86162\n",
      "\n",
      "360/500 [ 72%] █████████████████████          ETA: 7s | Acceptance Rate: 0.000\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\n",
      "1.99763\n",
      "Loss -1.86162\n",
      "\n",
      "370/500 [ 74%] ██████████████████████         ETA: 6s | Acceptance Rate: 0.000\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\n",
      "1.9976\n",
      "Loss -1.86162\n",
      "\n",
      "380/500 [ 76%] ██████████████████████         ETA: 5s | Acceptance Rate: 0.000\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\n",
      "1.99754\n",
      "Loss -1.86162\n",
      "\n",
      "390/500 [ 78%] ███████████████████████        ETA: 5s | Acceptance Rate: 0.000\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\n",
      "1.99745\n",
      "Loss -1.86162\n",
      "\n",
      "400/500 [ 80%] ████████████████████████       ETA: 4s | Acceptance Rate: 0.000\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\n",
      "1.99739\n",
      "Loss -1.86162\n",
      "\n",
      "410/500 [ 82%] ████████████████████████       ETA: 4s | Acceptance Rate: 0.000\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\n",
      "1.99732\n",
      "Loss -1.86162\n",
      "\n",
      "420/500 [ 84%] █████████████████████████      ETA: 3s | Acceptance Rate: 0.000\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\n",
      "1.99725\n",
      "Loss -1.86162\n",
      "\n",
      "430/500 [ 86%] █████████████████████████      ETA: 3s | Acceptance Rate: 0.000\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\n",
      "1.99718\n",
      "Loss -1.86162\n",
      "\n",
      "440/500 [ 88%] ██████████████████████████     ETA: 2s | Acceptance Rate: 0.000\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\n",
      "1.99714\n",
      "Loss -1.86162\n",
      "\n",
      "450/500 [ 90%] ███████████████████████████    ETA: 2s | Acceptance Rate: 0.000\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\n",
      "1.99707\n",
      "Loss -1.86162\n",
      "\n",
      "460/500 [ 92%] ███████████████████████████    ETA: 1s | Acceptance Rate: 0.000\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\n",
      "1.99701\n",
      "Loss -1.86162\n",
      "\n",
      "470/500 [ 94%] ████████████████████████████   ETA: 1s | Acceptance Rate: 0.000\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\n",
      "1.99694\n",
      "Loss -1.86162\n",
      "\n",
      "480/500 [ 96%] ████████████████████████████   ETA: 1s | Acceptance Rate: 0.000\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\n",
      "1.99686\n",
      "Loss -1.86162\n",
      "\n",
      "490/500 [ 98%] █████████████████████████████  ETA: 0s | Acceptance Rate: 0.000\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\n",
      "1.99682\n",
      "Loss -1.86162\n",
      "\n",
      "500/500 [100%] ██████████████████████████████ Elapsed: 25s | Acceptance Rate: 0.000\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\n",
      "\n",
      "1.99676\n",
      "Loss -1.86162\n",
      "\n"
     ]
    }
   ],
   "source": [
    "with tf.name_scope(\"inference\"):\n",
    "    inference = ed.inferences.HMC({NN: empiric})\n",
    "    inference.initialize()\n",
    "    sess = ed.get_session()\n",
    "    tf.global_variables_initializer().run()\n",
    "\n",
    "    print(inference.n_iter)\n",
    "    for i in range(inference.n_iter):\n",
    "        info_dict = inference.update()\n",
    "        inference.print_progress(info_dict)\n",
    "        t = info_dict['t']\n",
    "        if t % 100 == 0:\n",
    "            print()\n",
    "            pics = sess.run(empiric.get_variables())[0]\n",
    "            print(pics.mean())\n",
    "            print('Loss', sess.run(NN.log_prob(pics[1])).mean())\n",
    "            print()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([[[[ 0.64244258,  0.95129108,  0.31718183],\n",
       "         [ 1.59534585,  2.31182766,  1.9686718 ],\n",
       "         [ 1.03903389,  0.80012167,  3.36637402],\n",
       "         ..., \n",
       "         [ 3.20131755,  2.30033064, -0.11945844],\n",
       "         [ 1.51400733,  2.95512009,  0.23125696],\n",
       "         [ 2.4529705 ,  3.06067133,  2.29629016]],\n",
       "\n",
       "        [[ 2.30872321,  2.78408146,  4.01946545],\n",
       "         [ 1.94570696,  0.77270246,  1.17918098],\n",
       "         [ 2.21998596,  3.08780885, -1.11257005],\n",
       "         ..., \n",
       "         [ 0.19757986,  0.53924358,  1.1882987 ],\n",
       "         [ 3.54100609,  1.08081138,  0.68205297],\n",
       "         [ 2.11315465,  2.02942657,  0.50478816]],\n",
       "\n",
       "        [[ 3.22460723,  2.05025673,  1.87231219],\n",
       "         [ 0.11530101,  1.67652059,  1.71445334],\n",
       "         [ 4.32448292,  2.1193316 , -1.06634498],\n",
       "         ..., \n",
       "         [ 1.88726795,  3.01932049,  0.27497375],\n",
       "         [ 2.95477009,  1.94157791,  1.66287732],\n",
       "         [ 2.23451829,  1.56608987,  1.07996511]],\n",
       "\n",
       "        ..., \n",
       "        [[ 2.14906311,  1.2876451 ,  0.0203402 ],\n",
       "         [ 1.44096565,  0.96881437,  3.01865244],\n",
       "         [ 1.43783927,  2.49392438,  1.27035975],\n",
       "         ..., \n",
       "         [ 2.52390528,  3.06158066,  0.91184509],\n",
       "         [ 1.56479883,  3.17456341,  1.6508944 ],\n",
       "         [ 2.12321448,  2.63687086,  3.73630047]],\n",
       "\n",
       "        [[ 2.97932458,  1.98850513,  1.95648587],\n",
       "         [ 1.69649208,  3.27238417,  0.88261819],\n",
       "         [ 2.96720171,  1.83968425,  3.4610157 ],\n",
       "         ..., \n",
       "         [ 2.22180176,  0.76730096,  2.24792266],\n",
       "         [ 2.82381058,  0.909132  ,  2.98192954],\n",
       "         [ 2.05385637,  3.96989989,  0.15474737]],\n",
       "\n",
       "        [[ 1.87214994,  2.96182323,  3.67447519],\n",
       "         [ 2.65571308,  4.03195477,  2.6825285 ],\n",
       "         [ 1.30321074,  1.0339582 ,  2.99070239],\n",
       "         ..., \n",
       "         [ 1.59551942,  1.59794652,  4.90345097],\n",
       "         [ 3.04819083,  1.41315031,  3.1602881 ],\n",
       "         [ 1.23057604,  1.32200074,  2.64875221]]],\n",
       "\n",
       "\n",
       "       [[[ 0.64244258,  0.95129108,  0.31718183],\n",
       "         [ 1.59534585,  2.31182766,  1.9686718 ],\n",
       "         [ 1.03903389,  0.80012167,  3.36637402],\n",
       "         ..., \n",
       "         [ 3.20131755,  2.30033064, -0.11945844],\n",
       "         [ 1.51400733,  2.95512009,  0.23125696],\n",
       "         [ 2.4529705 ,  3.06067133,  2.29629016]],\n",
       "\n",
       "        [[ 2.30872321,  2.78408146,  4.01946545],\n",
       "         [ 1.94570696,  0.77270246,  1.17918098],\n",
       "         [ 2.21998596,  3.08780885, -1.11257005],\n",
       "         ..., \n",
       "         [ 0.19757986,  0.53924358,  1.1882987 ],\n",
       "         [ 3.54100609,  1.08081138,  0.68205297],\n",
       "         [ 2.11315465,  2.02942657,  0.50478816]],\n",
       "\n",
       "        [[ 3.22460723,  2.05025673,  1.87231219],\n",
       "         [ 0.11530101,  1.67652059,  1.71445334],\n",
       "         [ 4.32448292,  2.1193316 , -1.06634498],\n",
       "         ..., \n",
       "         [ 1.88726795,  3.01932049,  0.27497375],\n",
       "         [ 2.95477009,  1.94157791,  1.66287732],\n",
       "         [ 2.23451829,  1.56608987,  1.07996511]],\n",
       "\n",
       "        ..., \n",
       "        [[ 2.14906311,  1.2876451 ,  0.0203402 ],\n",
       "         [ 1.44096565,  0.96881437,  3.01865244],\n",
       "         [ 1.43783927,  2.49392438,  1.27035975],\n",
       "         ..., \n",
       "         [ 2.52390528,  3.06158066,  0.91184509],\n",
       "         [ 1.56479883,  3.17456341,  1.6508944 ],\n",
       "         [ 2.12321448,  2.63687086,  3.73630047]],\n",
       "\n",
       "        [[ 2.97932458,  1.98850513,  1.95648587],\n",
       "         [ 1.69649208,  3.27238417,  0.88261819],\n",
       "         [ 2.96720171,  1.83968425,  3.4610157 ],\n",
       "         ..., \n",
       "         [ 2.22180176,  0.76730096,  2.24792266],\n",
       "         [ 2.82381058,  0.909132  ,  2.98192954],\n",
       "         [ 2.05385637,  3.96989989,  0.15474737]],\n",
       "\n",
       "        [[ 1.87214994,  2.96182323,  3.67447519],\n",
       "         [ 2.65571308,  4.03195477,  2.6825285 ],\n",
       "         [ 1.30321074,  1.0339582 ,  2.99070239],\n",
       "         ..., \n",
       "         [ 1.59551942,  1.59794652,  4.90345097],\n",
       "         [ 3.04819083,  1.41315031,  3.1602881 ],\n",
       "         [ 1.23057604,  1.32200074,  2.64875221]]],\n",
       "\n",
       "\n",
       "       [[[ 0.64244258,  0.95129108,  0.31718183],\n",
       "         [ 1.59534585,  2.31182766,  1.9686718 ],\n",
       "         [ 1.03903389,  0.80012167,  3.36637402],\n",
       "         ..., \n",
       "         [ 3.20131755,  2.30033064, -0.11945844],\n",
       "         [ 1.51400733,  2.95512009,  0.23125696],\n",
       "         [ 2.4529705 ,  3.06067133,  2.29629016]],\n",
       "\n",
       "        [[ 2.30872321,  2.78408146,  4.01946545],\n",
       "         [ 1.94570696,  0.77270246,  1.17918098],\n",
       "         [ 2.21998596,  3.08780885, -1.11257005],\n",
       "         ..., \n",
       "         [ 0.19757986,  0.53924358,  1.1882987 ],\n",
       "         [ 3.54100609,  1.08081138,  0.68205297],\n",
       "         [ 2.11315465,  2.02942657,  0.50478816]],\n",
       "\n",
       "        [[ 3.22460723,  2.05025673,  1.87231219],\n",
       "         [ 0.11530101,  1.67652059,  1.71445334],\n",
       "         [ 4.32448292,  2.1193316 , -1.06634498],\n",
       "         ..., \n",
       "         [ 1.88726795,  3.01932049,  0.27497375],\n",
       "         [ 2.95477009,  1.94157791,  1.66287732],\n",
       "         [ 2.23451829,  1.56608987,  1.07996511]],\n",
       "\n",
       "        ..., \n",
       "        [[ 2.14906311,  1.2876451 ,  0.0203402 ],\n",
       "         [ 1.44096565,  0.96881437,  3.01865244],\n",
       "         [ 1.43783927,  2.49392438,  1.27035975],\n",
       "         ..., \n",
       "         [ 2.52390528,  3.06158066,  0.91184509],\n",
       "         [ 1.56479883,  3.17456341,  1.6508944 ],\n",
       "         [ 2.12321448,  2.63687086,  3.73630047]],\n",
       "\n",
       "        [[ 2.97932458,  1.98850513,  1.95648587],\n",
       "         [ 1.69649208,  3.27238417,  0.88261819],\n",
       "         [ 2.96720171,  1.83968425,  3.4610157 ],\n",
       "         ..., \n",
       "         [ 2.22180176,  0.76730096,  2.24792266],\n",
       "         [ 2.82381058,  0.909132  ,  2.98192954],\n",
       "         [ 2.05385637,  3.96989989,  0.15474737]],\n",
       "\n",
       "        [[ 1.87214994,  2.96182323,  3.67447519],\n",
       "         [ 2.65571308,  4.03195477,  2.6825285 ],\n",
       "         [ 1.30321074,  1.0339582 ,  2.99070239],\n",
       "         ..., \n",
       "         [ 1.59551942,  1.59794652,  4.90345097],\n",
       "         [ 3.04819083,  1.41315031,  3.1602881 ],\n",
       "         [ 1.23057604,  1.32200074,  2.64875221]]],\n",
       "\n",
       "\n",
       "       ..., \n",
       "       [[[ 0.64244258,  0.95129108,  0.31718183],\n",
       "         [ 1.59534585,  2.31182766,  1.9686718 ],\n",
       "         [ 1.03903389,  0.80012167,  3.36637402],\n",
       "         ..., \n",
       "         [ 3.20131755,  2.30033064, -0.11945844],\n",
       "         [ 1.51400733,  2.95512009,  0.23125696],\n",
       "         [ 2.4529705 ,  3.06067133,  2.29629016]],\n",
       "\n",
       "        [[ 2.30872321,  2.78408146,  4.01946545],\n",
       "         [ 1.94570696,  0.77270246,  1.17918098],\n",
       "         [ 2.21998596,  3.08780885, -1.11257005],\n",
       "         ..., \n",
       "         [ 0.19757986,  0.53924358,  1.1882987 ],\n",
       "         [ 3.54100609,  1.08081138,  0.68205297],\n",
       "         [ 2.11315465,  2.02942657,  0.50478816]],\n",
       "\n",
       "        [[ 3.22460723,  2.05025673,  1.87231219],\n",
       "         [ 0.11530101,  1.67652059,  1.71445334],\n",
       "         [ 4.32448292,  2.1193316 , -1.06634498],\n",
       "         ..., \n",
       "         [ 1.88726795,  3.01932049,  0.27497375],\n",
       "         [ 2.95477009,  1.94157791,  1.66287732],\n",
       "         [ 2.23451829,  1.56608987,  1.07996511]],\n",
       "\n",
       "        ..., \n",
       "        [[ 2.14906311,  1.2876451 ,  0.0203402 ],\n",
       "         [ 1.44096565,  0.96881437,  3.01865244],\n",
       "         [ 1.43783927,  2.49392438,  1.27035975],\n",
       "         ..., \n",
       "         [ 2.52390528,  3.06158066,  0.91184509],\n",
       "         [ 1.56479883,  3.17456341,  1.6508944 ],\n",
       "         [ 2.12321448,  2.63687086,  3.73630047]],\n",
       "\n",
       "        [[ 2.97932458,  1.98850513,  1.95648587],\n",
       "         [ 1.69649208,  3.27238417,  0.88261819],\n",
       "         [ 2.96720171,  1.83968425,  3.4610157 ],\n",
       "         ..., \n",
       "         [ 2.22180176,  0.76730096,  2.24792266],\n",
       "         [ 2.82381058,  0.909132  ,  2.98192954],\n",
       "         [ 2.05385637,  3.96989989,  0.15474737]],\n",
       "\n",
       "        [[ 1.87214994,  2.96182323,  3.67447519],\n",
       "         [ 2.65571308,  4.03195477,  2.6825285 ],\n",
       "         [ 1.30321074,  1.0339582 ,  2.99070239],\n",
       "         ..., \n",
       "         [ 1.59551942,  1.59794652,  4.90345097],\n",
       "         [ 3.04819083,  1.41315031,  3.1602881 ],\n",
       "         [ 1.23057604,  1.32200074,  2.64875221]]],\n",
       "\n",
       "\n",
       "       [[[ 0.64244258,  0.95129108,  0.31718183],\n",
       "         [ 1.59534585,  2.31182766,  1.9686718 ],\n",
       "         [ 1.03903389,  0.80012167,  3.36637402],\n",
       "         ..., \n",
       "         [ 3.20131755,  2.30033064, -0.11945844],\n",
       "         [ 1.51400733,  2.95512009,  0.23125696],\n",
       "         [ 2.4529705 ,  3.06067133,  2.29629016]],\n",
       "\n",
       "        [[ 2.30872321,  2.78408146,  4.01946545],\n",
       "         [ 1.94570696,  0.77270246,  1.17918098],\n",
       "         [ 2.21998596,  3.08780885, -1.11257005],\n",
       "         ..., \n",
       "         [ 0.19757986,  0.53924358,  1.1882987 ],\n",
       "         [ 3.54100609,  1.08081138,  0.68205297],\n",
       "         [ 2.11315465,  2.02942657,  0.50478816]],\n",
       "\n",
       "        [[ 3.22460723,  2.05025673,  1.87231219],\n",
       "         [ 0.11530101,  1.67652059,  1.71445334],\n",
       "         [ 4.32448292,  2.1193316 , -1.06634498],\n",
       "         ..., \n",
       "         [ 1.88726795,  3.01932049,  0.27497375],\n",
       "         [ 2.95477009,  1.94157791,  1.66287732],\n",
       "         [ 2.23451829,  1.56608987,  1.07996511]],\n",
       "\n",
       "        ..., \n",
       "        [[ 2.14906311,  1.2876451 ,  0.0203402 ],\n",
       "         [ 1.44096565,  0.96881437,  3.01865244],\n",
       "         [ 1.43783927,  2.49392438,  1.27035975],\n",
       "         ..., \n",
       "         [ 2.52390528,  3.06158066,  0.91184509],\n",
       "         [ 1.56479883,  3.17456341,  1.6508944 ],\n",
       "         [ 2.12321448,  2.63687086,  3.73630047]],\n",
       "\n",
       "        [[ 2.97932458,  1.98850513,  1.95648587],\n",
       "         [ 1.69649208,  3.27238417,  0.88261819],\n",
       "         [ 2.96720171,  1.83968425,  3.4610157 ],\n",
       "         ..., \n",
       "         [ 2.22180176,  0.76730096,  2.24792266],\n",
       "         [ 2.82381058,  0.909132  ,  2.98192954],\n",
       "         [ 2.05385637,  3.96989989,  0.15474737]],\n",
       "\n",
       "        [[ 1.87214994,  2.96182323,  3.67447519],\n",
       "         [ 2.65571308,  4.03195477,  2.6825285 ],\n",
       "         [ 1.30321074,  1.0339582 ,  2.99070239],\n",
       "         ..., \n",
       "         [ 1.59551942,  1.59794652,  4.90345097],\n",
       "         [ 3.04819083,  1.41315031,  3.1602881 ],\n",
       "         [ 1.23057604,  1.32200074,  2.64875221]]],\n",
       "\n",
       "\n",
       "       [[[ 0.64244258,  0.95129108,  0.31718183],\n",
       "         [ 1.59534585,  2.31182766,  1.9686718 ],\n",
       "         [ 1.03903389,  0.80012167,  3.36637402],\n",
       "         ..., \n",
       "         [ 3.20131755,  2.30033064, -0.11945844],\n",
       "         [ 1.51400733,  2.95512009,  0.23125696],\n",
       "         [ 2.4529705 ,  3.06067133,  2.29629016]],\n",
       "\n",
       "        [[ 2.30872321,  2.78408146,  4.01946545],\n",
       "         [ 1.94570696,  0.77270246,  1.17918098],\n",
       "         [ 2.21998596,  3.08780885, -1.11257005],\n",
       "         ..., \n",
       "         [ 0.19757986,  0.53924358,  1.1882987 ],\n",
       "         [ 3.54100609,  1.08081138,  0.68205297],\n",
       "         [ 2.11315465,  2.02942657,  0.50478816]],\n",
       "\n",
       "        [[ 3.22460723,  2.05025673,  1.87231219],\n",
       "         [ 0.11530101,  1.67652059,  1.71445334],\n",
       "         [ 4.32448292,  2.1193316 , -1.06634498],\n",
       "         ..., \n",
       "         [ 1.88726795,  3.01932049,  0.27497375],\n",
       "         [ 2.95477009,  1.94157791,  1.66287732],\n",
       "         [ 2.23451829,  1.56608987,  1.07996511]],\n",
       "\n",
       "        ..., \n",
       "        [[ 2.14906311,  1.2876451 ,  0.0203402 ],\n",
       "         [ 1.44096565,  0.96881437,  3.01865244],\n",
       "         [ 1.43783927,  2.49392438,  1.27035975],\n",
       "         ..., \n",
       "         [ 2.52390528,  3.06158066,  0.91184509],\n",
       "         [ 1.56479883,  3.17456341,  1.6508944 ],\n",
       "         [ 2.12321448,  2.63687086,  3.73630047]],\n",
       "\n",
       "        [[ 2.97932458,  1.98850513,  1.95648587],\n",
       "         [ 1.69649208,  3.27238417,  0.88261819],\n",
       "         [ 2.96720171,  1.83968425,  3.4610157 ],\n",
       "         ..., \n",
       "         [ 2.22180176,  0.76730096,  2.24792266],\n",
       "         [ 2.82381058,  0.909132  ,  2.98192954],\n",
       "         [ 2.05385637,  3.96989989,  0.15474737]],\n",
       "\n",
       "        [[ 1.87214994,  2.96182323,  3.67447519],\n",
       "         [ 2.65571308,  4.03195477,  2.6825285 ],\n",
       "         [ 1.30321074,  1.0339582 ,  2.99070239],\n",
       "         ..., \n",
       "         [ 1.59551942,  1.59794652,  4.90345097],\n",
       "         [ 3.04819083,  1.41315031,  3.1602881 ],\n",
       "         [ 1.23057604,  1.32200074,  2.64875221]]]], dtype=float32)"
      ]
     },
     "execution_count": 8,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "sess.run(empiric.get_variables())[0]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.6.0"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}