.ipynb_checkpoints/check_model-checkpoint.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Using TensorFlow backend.\n"
     ]
    }
   ],
   "source": [
    "from keras.models import Sequential\n",
    "from keras.layers import Dense\n",
    "from keras import regularizers\n",
    "from keras.optimizers import Adam, RMSprop\n",
    "from loss import cemse_loss\n",
    "from keras import losses\n",
    "import numpy as np\n",
    "import tensorflow as tf\n",
    "import matplotlib.pyplot as plt\n",
    "\n",
    "custom_loss = cemse_loss()\n",
    "\n",
    "model = Sequential()\n",
    "model.add(Dense(10, use_bias=True, bias_initializer='random_uniform', activation='relu', kernel_regularizer=regularizers.l2(0.01), input_dim=18))\n",
    "model.add(Dense(10, use_bias=True, bias_initializer='random_uniform', activation='relu', kernel_regularizer=regularizers.l2(0.01)))\n",
    "model.add(Dense(9 + 1, use_bias=True, bias_initializer='random_uniform', activation='linear', kernel_regularizer=regularizers.l2(0.01)))\n",
    "model.compile(loss=custom_loss,\n",
    "    optimizer=Adam(lr=0.01))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "state = [1,1,0,0,0,0,0,0,0,0,0,0,-1,-1,0,0,0,0]\n",
    "target = 1\n",
    "next_move = [0,0,0.5,0,0,0.2,0.3,0,0]\n",
    "\n",
    "y_true = np.array([[target] + next_move])\n",
    "\n",
    "training_states = np.array([state])\n",
    "training_targets = y_true"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "fit = model.fit(training_states, training_targets, epochs=1, verbose=0)\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "{'loss': [3.0224502086639404]}"
      ]
     },
     "execution_count": 4,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "fit.history"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 200,
   "metadata": {},
   "outputs": [],
   "source": [
    "for i in range(100):\n",
    "    fit = model.fit(training_states, training_targets, epochs=1, verbose=0)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [],
   "source": [
    "y_pred = model.predict(np.array([state]))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [],
   "source": [
    "v = tf.slice(tf.constant(y_pred, dtype='float64'), [0,0], [-1,1])\n",
    "z = tf.slice(tf.constant(y_true, dtype='float64'), [0,0], [-1,1])\n",
    "\n",
    "p = tf.slice(tf.constant(y_pred, dtype='float64'), [0,1], [-1,-1])\n",
    "pi = tf.slice(tf.constant(y_true, dtype='float64'), [0,1], [-1,-1])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {},
   "outputs": [
    {
     "ename": "TypeError",
     "evalue": "Input 'y' of 'NotEqual' Op has type int32 that does not match type float64 of argument 'x'.",
     "output_type": "error",
     "traceback": [
      "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
      "\u001b[0;31mTypeError\u001b[0m                                 Traceback (most recent call last)",
      "\u001b[0;32m<ipython-input-20-821feb4fb351>\u001b[0m in \u001b[0;36m<module>\u001b[0;34m()\u001b[0m\n\u001b[1;32m      1\u001b[0m \u001b[0mzero\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mtf\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mconstant\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;36m0\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m----> 2\u001b[0;31m \u001b[0mloc\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mtf\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mnot_equal\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mpi\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mzero\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m      3\u001b[0m \u001b[0msess\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mrun\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mloc\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;32m/Users/davidfoster/.virtualenvs/deepreinforcement/lib/python2.7/site-packages/tensorflow/python/ops/gen_math_ops.pyc\u001b[0m in \u001b[0;36mnot_equal\u001b[0;34m(x, y, name)\u001b[0m\n\u001b[1;32m   2851\u001b[0m   \u001b[0;32mif\u001b[0m \u001b[0m_ctx\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0min_graph_mode\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m   2852\u001b[0m     _, _, _op = _op_def_lib._apply_op_helper(\n\u001b[0;32m-> 2853\u001b[0;31m         \"NotEqual\", x=x, y=y, name=name)\n\u001b[0m\u001b[1;32m   2854\u001b[0m     \u001b[0m_result\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0m_op\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0moutputs\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m   2855\u001b[0m     \u001b[0m_inputs_flat\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0m_op\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0minputs\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;32m/Users/davidfoster/.virtualenvs/deepreinforcement/lib/python2.7/site-packages/tensorflow/python/framework/op_def_library.pyc\u001b[0m in \u001b[0;36m_apply_op_helper\u001b[0;34m(self, op_type_name, name, **keywords)\u001b[0m\n\u001b[1;32m    544\u001b[0m                   \u001b[0;34m\"%s type %s of argument '%s'.\"\u001b[0m \u001b[0;34m%\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    545\u001b[0m                   (prefix, dtypes.as_dtype(attrs[input_arg.type_attr]).name,\n\u001b[0;32m--> 546\u001b[0;31m                    inferred_from[input_arg.type_attr]))\n\u001b[0m\u001b[1;32m    547\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    548\u001b[0m           \u001b[0mtypes\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0;34m[\u001b[0m\u001b[0mvalues\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mdtype\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;31mTypeError\u001b[0m: Input 'y' of 'NotEqual' Op has type int32 that does not match type float64 of argument 'x'."
     ]
    }
   ],
   "source": [
    "zero = tf.constant(0, dtype=tf.float64)\n",
    "loc = tf.not_equal(pi, zero)\n",
    "sess.run(loc)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [],
   "source": [
    "loss = losses.mean_squared_error(z, tf.tanh(v)) + tf.nn.softmax_cross_entropy_with_logits(labels = tf.boolean_mask(pi,loc), logits = tf.boolean_mask(p,loc))\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[ 1.42451725]\n",
      "[ 0.5  0.2  0.3]\n",
      "[ 0.09151755 -0.57005888  0.28418985]\n"
     ]
    }
   ],
   "source": [
    "sess = tf.Session()\n",
    "print(sess.run(loss))\n",
    "print(sess.run(tf.boolean_mask(pi,loc)))\n",
    "print(sess.run(tf.boolean_mask(p,loc)))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[[ 0.   0.   0.5  0.   0.   0.2  0.3  0.   0. ]]\n",
      "[[ 0.01390975  0.02305907  0.09151755  0.01204331  0.34318545 -0.57005888\n",
      "   0.28418985  0.05954683  0.06024369]]\n",
      "[[ 0.10593089  0.10690453  0.11447938  0.10573336  0.14723981  0.05907562\n",
      "   0.13880457  0.11087728  0.11095457]]\n",
      "2.24187418139\n",
      "[ 2.24187418]\n"
     ]
    }
   ],
   "source": [
    "print(sess.run(pi))\n",
    "print(sess.run(p))\n",
    "softmax = tf.nn.softmax(p)\n",
    "\n",
    "print(sess.run(softmax))\n",
    "total_loss = tf.reduce_mean(-tf.reduce_sum(pi * tf.log(softmax), [1]))\n",
    "\n",
    "print(sess.run(total_loss))\n",
    "print(sess.run(tf.nn.softmax_cross_entropy_with_logits(labels = pi, logits = p)))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 210,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([[ 3.01843262, -3.35358572, -2.37631893,  3.34004378, -2.99087596,\n",
       "        -2.9094696 ,  2.45611715,  2.80245829, -2.63021517, -2.40210104]], dtype=float32)"
      ]
     },
     "execution_count": 210,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "y_pred"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 208,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAJ4AAAD8CAYAAACGuR0qAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\ngg==\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x11d687190>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAX8AAABECAYAAABpjjW9AAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x11de63710>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAPgAAAD8CAYAAABaQGkdAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x11e75ca90>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAX8AAABECAYAAABpjjW9AAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nrkJggg==\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x11d26b350>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAPgAAAD8CAYAAABaQGkdAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz\nAAALEgAACxIB0t1+/AAAADl0RVh0U29mdHdhcmUAbWF0cGxvdGxpYiB2ZXJzaW9uIDIuMS4wLCBo\ndHRwOi8vbWF0cGxvdGxpYi5vcmcvpW3flQAAC6BJREFUeJzt3d9rnvUZx/HPx6RV08ra2G3YtNhs\nOqUoostELchQGTrFdjBGBR3zpDDmTwTRnfgPiOiBCMUfB7PTg1qcOPEHqAM9KMa2UNsqdFX7e3aL\nrbWbNjXXDpJBFZvnTvP9eue59n6B0MTbq5dp3rmf58mdu44IAcjplLYXAFAPgQOJETiQGIEDiRE4\nkBiBA4kROJAYgQOJETiQWG+NofPnnxlnDSwuPrfHXxWfKUlW+av5ekf/U3ymJI3O6qsyV3KluVwp\nWcOePXs0MjLS8Q+tSuBnDSzWn9e9Vnzu3J7Pi8+UpFljXxafuWD/5uIzJWn/wp9WmTtW6cFcb4xW\nmRsu/wUpuugB7YoVyxsd1z3/RwCmjMCBxAgcSIzAgcQIHEiMwIHEGgVu+1rbH9jebvu+2ksBKKNj\n4LZ7JD0q6TpJSyXdZHtp7cUATF+TM/ilkrZHxI6IOCrpWUnNvssOoFVNAh+QtOu4t3dPvO9rbK+y\nPWx7+OCn/yq1H4BpKPYiW0SsjoihiBiaN//MUmMBTEOTwPdIOv4nRxZNvA/ADNck8HcknWt70PZs\nSSslvVB3LQAldPxpsog4Zvs2Sa9I6pH0ZERsqb4ZgGlr9OOiEfGSpJcq7wKgMK5kAxIjcCAxAgcS\nI3AgMQIHEqty08XTjh3WT/a/XnzuwR+eX3ymJB3qWVB85vD864vPlKSzdKDK3DnHDlWZ+1lPnasa\nd31e/s/snLm7Oh90EkZjdvGZ0fAuuJzBgcQIHEiMwIHECBxIjMCBxAgcSIzAgcQIHEiMwIHECBxI\njMCBxAgcSIzAgcQIHEiMwIHECBxIjMCBxAgcSIzAgcQIHEiMwIHEqtxVdXRWn/YN/KzG6Cpm68vi\nMxedvr/4zJq+6J1bZW6Nj60k/Xhu9/wN1j3+qvhMKxodxxkcSIzAgcQIHEiMwIHECBxIjMCBxDoG\nbnux7Tdsb7W9xfad38ViAKavyffBj0m6JyI22D5D0ru2X4uIrZV3AzBNHc/gEbEvIjZM/PqwpG2S\nBmovBmD6pvQc3PYSSRdLWl9jGQBlNQ7c9lxJz0m6KyI++5Z/v8r2sO3hkZGRkjsCOEmNArc9S+Nx\nr4mIdd92TESsjoihiBjq7+8vuSOAk9TkVXRLekLStoh4qP5KAEppcgZfJukWSVfZ3jTxzy8r7wWg\ngI7fJouItyT5O9gFQGFcyQYkRuBAYgQOJEbgQGIEDiRW5aaLknRKlL/R3KGxecVnStLrW88sPnPF\nhTuKz5Skz8fOqDJ3z+fzq8z9+T/+VGVufLK3+My3Lrq3+ExJen/X7OIzD31xaqPjOIMDiRE4kBiB\nA4kROJAYgQOJETiQGIEDiRE4kBiBA4kROJAYgQOJETiQGIEDiRE4kBiBA4kROJAYgQOJETiQGIED\niRE4kBiBA4lVuavq6Pa/a9fy3xSfe/+PHi4+U5J+e/uVxWduPe/64jMl6cnf/7XK3Hk/qPO1fmj0\n7SpzTz2z/F1gX337WPGZkjS4pPxdVZviDA4kRuBAYgQOJEbgQGIEDiRG4EBiBA4k1jhw2z22N9p+\nseZCAMqZyhn8Tknbai0CoLxGgdteJOl6SY/XXQdASU3P4A9LulfS2IkOsL3K9rDt4YNf1bnkD8DU\ndAzc9g2SPomIdyc7LiJWR8RQRAzN66lyiTuAKWpyBl8m6UbbH0l6VtJVtp+uuhWAIjoGHhH3R8Si\niFgiaaWk1yPi5uqbAZg2vg8OJDalJ8sR8aakN6tsAqA4zuBAYgQOJEbgQGIEDiRG4EBijojiQy+8\n8MJ4/vm/FJ8bcvGZkmSV/xjU4hNfLTwtwdf6rrJixXJt3ry5YxD8qQKJETiQGIEDiRE4kBiBA4kR\nOJAYgQOJETiQGIEDiRE4kBiBA4kROJAYgQOJETiQGIEDiRE4kBiBA4kROJAYgQOJETiQGIEDiXXV\nX+TdTXc/raXb7n7aTXeBrbXrmHqqzG2iuz5bAEwJgQOJETiQGIEDiRE4kBiBA4k1Ctz2PNtrbb9v\ne5vty2svBmD6mn4f/BFJL0fEr23PltRXcScAhXQM3Pb3JF0p6XeSFBFHJR2tuxaAEpo8RB+UdEDS\nU7Y32n7c9pzKewEooEngvZIukfRYRFws6Yik+755kO1VtodtD4+MjBReE8DJaBL4bkm7I2L9xNtr\nNR7810TE6ogYioih/v7+kjsCOEkdA4+I/ZJ22T5v4l1XS9padSsARTR9Ff12SWsmXkHfIenWeisB\nKKVR4BGxSdJQ5V0AFMaVbEBiBA4kRuBAYgQOJEbgQGIEDiRW5a6qh4+eqr99PFh87i8WbSk+U5I+\n9YLiM8/ZtKb4TEnae8mvqsydc/RglblHZs+rMvfDIwuLz1wyZ1/xmZI0VuE8GnKj4ziDA4kROJAY\ngQOJETiQGIEDiRE4kBiBA4kROJAYgQOJETiQGIEDiRE4kBiBA4kROJAYgQOJETiQGIEDiRE4kBiB\nA4kROJBYlZsunjH7S1159kfF536pvuIzJalP/y4+s9bNEWupdXPEpjcHnKrBOXurzK3hFI0Vn2lF\nw98bQFoEDiRG4EBiBA4kRuBAYgQOJEbgQGKNArd9t+0ttt+z/Yzt02ovBmD6OgZue0DSHZKGIuIC\nST2SVtZeDMD0NX2I3ivpdNu9kvokdc9lRMD/sY6BR8QeSQ9K2ilpn6RDEfHqN4+zvcr2sO3hkZGR\n8psCmLImD9HnS1ouaVDSQklzbN/8zeMiYnVEDEXEUH9/f/lNAUxZk4fo10j6MCIORMSopHWSrqi7\nFoASmgS+U9JltvtsW9LVkrbVXQtACU2eg6+XtFbSBkmbJ/6b1ZX3AlBAo58Hj4gHJD1QeRcAhXEl\nG5AYgQOJETiQGIEDiRE4kFiVu6pK0in6qvjM4OuRXOEOnZI0pp4qc5ve/TOzNj8GFAMkRuBAYgQO\nJEbgQGIEDiRG4EBiBA4kRuBAYgQOJEbgQGIEDiRG4EBiBA4kRuBAYgQOJEbgQGIEDiRG4EBiBA4k\nRuBAYgQOJOaI8nd8tH1A0scNDl0g6Z/FF6inm/btpl2l7tp3Jux6dkR8v9NBVQJvyvZwRAy1tsAU\nddO+3bSr1F37dtOuPEQHEiNwILG2A1/d8u8/Vd20bzftKnXXvl2za6vPwQHU1fYZHEBFrQVu+1rb\nH9jebvu+tvboxPZi22/Y3mp7i+07296pCds9tjfafrHtXSZje57ttbbft73N9uVt7zQZ23dPfB68\nZ/sZ26e1vdNkWgncdo+kRyVdJ2mppJtsL21jlwaOSbonIpZKukzSH2bwrse7U9K2tpdo4BFJL0fE\n+ZIu0gze2faApDskDUXEBZJ6JK1sd6vJtXUGv1TS9ojYERFHJT0raXlLu0wqIvZFxIaJXx/W+Cfg\nQLtbTc72IknXS3q87V0mY/t7kq6U9IQkRcTRiDjY7lYd9Uo63XavpD5Je1veZ1JtBT4gaddxb+/W\nDI9GkmwvkXSxpPXtbtLRw5LulSr9ZeLlDEo6IOmpiacTj9ue0/ZSJxIReyQ9KGmnpH2SDkXEq+1u\nNTleZGvI9lxJz0m6KyI+a3ufE7F9g6RPIuLdtndpoFfSJZIei4iLJR2RNJNfj5mv8Ueag5IWSppj\n++Z2t5pcW4HvkbT4uLcXTbxvRrI9S+Nxr4mIdW3v08EySTfa/kjjT32usv10uyud0G5JuyPif4+I\n1mo8+JnqGkkfRsSBiBiVtE7SFS3vNKm2An9H0rm2B23P1vgLFS+0tMukbFvjzxG3RcRDbe/TSUTc\nHxGLImKJxj+ur0fEjDzLRMR+SbtsnzfxrqslbW1xpU52SrrMdt/E58XVmsEvCkrjD5G+cxFxzPZt\nkl7R+CuRT0bEljZ2aWCZpFskbba9aeJ9f4yIl1rcKZPbJa2Z+EK/Q9KtLe9zQhGx3vZaSRs0/t2V\njZrhV7VxJRuQGC+yAYkROJAYgQOJETiQGIEDiRE4kBiBA4kROJDYfwGBbYhPclscCgAAAABJRU5E\nrkJggg==\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x11d5d7590>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAX8AAABECAYAAABpjjW9AAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\ngg==\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x11de21e50>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "l1 = model.layers[0].get_weights()[0]\n",
    "l1_bias = [model.layers[0].get_weights()[1]]\n",
    "l2 = model.layers[1].get_weights()[0]\n",
    "l2_bias = [model.layers[1].get_weights()[1]]\n",
    "l3 = model.layers[2].get_weights()[0]\n",
    "l3_bias = [model.layers[2].get_weights()[1]]\n",
    "plt.imshow(l1, cmap='coolwarm')\n",
    "plt.show()\n",
    "plt.imshow(l1_bias, cmap='coolwarm')\n",
    "plt.show()\n",
    "plt.imshow(l2, cmap='coolwarm')\n",
    "plt.show()\n",
    "plt.imshow(l2_bias, cmap='coolwarm')\n",
    "plt.show()\n",
    "plt.imshow(l3, cmap='coolwarm')\n",
    "plt.show()\n",
    "plt.imshow(l3_bias, cmap='coolwarm')\n",
    "plt.show()\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 209,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([[  1.63326238e-03,   3.16449344e-01,   5.86463630e-01,\n",
       "         -2.01987638e-03,  -8.03918540e-02,   3.22499429e-03,\n",
       "         -6.76295953e-04,   2.31270387e-04,  -3.93889584e-02,\n",
       "          2.98810769e-02],\n",
       "       [  3.74790933e-03,   3.19886327e-01,   4.97435331e-01,\n",
       "          3.93377207e-02,   2.33290941e-01,   1.85876188e-03,\n",
       "          1.30566055e-04,  -1.97542031e-05,   4.09340054e-01,\n",
       "          2.29218248e-02],\n",
       "       [ -1.79458119e-03,  -7.36567250e-04,  -1.68336974e-03,\n",
       "          7.71252380e-05,  -7.27214792e-04,  -1.60337469e-04,\n",
       "          1.77326205e-03,  -5.33297658e-04,  -4.45837097e-04,\n",
       "          6.06847927e-04],\n",
       "       [  4.75142151e-04,   1.94792147e-03,  -7.74591172e-07,\n",
       "          1.94060453e-03,   2.14025122e-03,   6.62030216e-05,\n",
       "          1.41475582e-04,  -5.88265713e-04,  -1.95807265e-03,\n",
       "         -9.69951361e-05],\n",
       "       [  2.84593494e-04,  -7.40125368e-04,   7.38236587e-04,\n",
       "         -1.95917371e-03,   1.32156827e-04,  -1.22703519e-03,\n",
       "         -4.18331986e-03,   2.72949343e-04,  -1.80080751e-04,\n",
       "          6.15868601e-04],\n",
       "       [ -1.69476116e-05,  -9.67179483e-04,   1.40215707e-06,\n",
       "          2.37035274e-04,   1.08089461e-03,   6.38876925e-04,\n",
       "         -3.07213422e-03,   1.75490987e-03,   1.50778258e-04,\n",
       "         -1.87554397e-03],\n",
       "       [ -7.81350536e-04,   7.89463229e-07,   1.02843063e-04,\n",
       "         -3.00052529e-03,   3.01247201e-04,  -1.14033069e-03,\n",
       "          3.54149306e-05,  -1.68544566e-03,  -2.89741135e-03,\n",
       "         -4.90532373e-04],\n",
       "       [  4.61248169e-03,   5.95118268e-04,  -4.35080286e-03,\n",
       "          4.57971240e-04,   2.50914192e-04,   9.95574752e-04,\n",
       "         -1.81498544e-04,   3.75931663e-03,  -8.47270639e-05,\n",
       "          9.55671072e-04],\n",
       "       [  4.09419695e-03,  -6.91802008e-04,   3.37542850e-04,\n",
       "          2.89478805e-04,  -7.90580816e-04,   1.16744905e-03,\n",
       "          1.89099199e-04,   7.25450300e-05,   5.39307366e-04,\n",
       "         -6.75208285e-04],\n",
       "       [ -7.41342519e-05,   1.89158600e-03,  -1.59170682e-04,\n",
       "         -1.13746384e-03,   1.88069663e-03,   3.51590024e-05,\n",
       "         -1.20351347e-03,   1.02841621e-03,  -2.52223050e-04,\n",
       "         -5.71289391e-04],\n",
       "       [ -3.75415402e-05,  -1.55794260e-06,  -4.55596577e-03,\n",
       "          4.38807532e-03,   7.76244910e-04,   5.12352388e-04,\n",
       "          3.08484916e-04,   4.21317643e-04,   4.49988898e-03,\n",
       "         -1.38257499e-04],\n",
       "       [ -3.62292049e-05,  -8.12688522e-05,   3.45208042e-04,\n",
       "         -1.95627520e-03,  -9.88077256e-04,   1.78457424e-03,\n",
       "         -1.65861542e-03,   3.94518982e-04,   7.82087562e-04,\n",
       "          8.21604772e-06],\n",
       "       [ -3.63035873e-03,  -2.94502497e-01,  -3.34966332e-01,\n",
       "          3.71005340e-03,   4.57581818e-01,  -1.37329195e-03,\n",
       "          1.70347223e-04,  -1.86926778e-03,  -7.26208016e-02,\n",
       "         -2.85171028e-02],\n",
       "       [ -1.72299042e-03,  -3.17751557e-01,  -4.82379079e-01,\n",
       "          6.70907944e-02,  -8.06102678e-02,   2.82194203e-04,\n",
       "         -1.10312772e-04,   5.95521415e-05,  -4.93049175e-01,\n",
       "         -1.72092691e-02],\n",
       "       [  1.96311681e-04,   1.97618129e-03,   1.12727087e-03,\n",
       "         -1.62933444e-04,  -7.34310132e-04,  -3.87770357e-04,\n",
       "         -2.34144216e-04,  -2.02960437e-05,  -1.01767248e-03,\n",
       "         -3.51970448e-05],\n",
       "       [ -1.83264507e-04,  -5.12971485e-04,  -1.33233465e-04,\n",
       "          1.89479289e-03,  -9.85731254e-04,   5.49553020e-04,\n",
       "         -4.88846053e-05,   5.93018951e-04,  -1.70807238e-04,\n",
       "         -4.25323175e-04],\n",
       "       [ -1.05608627e-03,  -1.45533343e-03,  -3.77990725e-03,\n",
       "          4.42105997e-03,   5.00839844e-04,  -1.96086336e-03,\n",
       "          9.53009934e-04,   1.91947818e-03,   5.23374649e-04,\n",
       "          2.58291839e-05],\n",
       "       [ -1.82355300e-03,  -3.29327933e-03,   9.11658135e-05,\n",
       "          1.93507108e-03,  -1.61876704e-03,  -2.67736427e-03,\n",
       "         -4.41248249e-03,  -8.72806646e-04,   3.77288810e-03,\n",
       "         -8.40242938e-05]], dtype=float32)"
      ]
     },
     "execution_count": 209,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "model.layers[0].get_weights()[0]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "deepreinforcement",
   "language": "python",
   "name": "deepreinforcement"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 2
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython2",
   "version": "2.7.13"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}