Product centricity-Copy1.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "%matplotlib inline\n",
    "import tensorflow as tf\n",
    "import numpy as np\n",
    "from sklearn.preprocessing import OneHotEncoder\n",
    "import os\n",
    "import matplotlib.pyplot as plt\n",
    "import time"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": [
    "def csv_to_numpy_array(filePath, delimiter):\n",
    "    return np.genfromtxt(filePath, delimiter=delimiter, dtype=float,skip_header=1)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Read the train and test data"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": [
    "trainX = csv_to_numpy_array(\"prod_centricity/trainX.csv\", delimiter=\",\")\n",
    "trainY = csv_to_numpy_array(\"prod_centricity/trainY.csv\", delimiter=\",\")\n",
    "testX = csv_to_numpy_array(\"prod_centricity/testX.csv\", delimiter=\",\")\n",
    "testY = csv_to_numpy_array(\"prod_centricity/testY.csv\", delimiter=\",\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": [
    "#########################\n",
    "### GLOBAL PARAMETERS ###\n",
    "#########################\n",
    "\n",
    "# DATA SET PARAMETERS\n",
    "# Get our dimensions for our different variables and placeholders:\n",
    "# numFeatures = the number of words extracted from each email\n",
    "numFeatures = 4 #trainX.shape[1]\n",
    "# numLabels = number of classes we are predicting (here just 2: Ham or Spam)\n",
    "numLabels = trainY.shape[1]\n",
    "\n",
    "# TRAINING SESSION PARAMETERS\n",
    "# number of times we iterate through training data\n",
    "# tensorboard shows that accuracy plateaus at ~25k epochs\n",
    "numEpochs = 27000\n",
    "# a smarter learning rate for gradientOptimizer\n",
    "learningRate = tf.train.exponential_decay(learning_rate=0.008,\n",
    "                                          global_step= 1,\n",
    "                                          decay_steps=trainX.shape[0],\n",
    "                                          decay_rate= 0.95,\n",
    "                                          staircase=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": [
    "####################\n",
    "### PLACEHOLDERS ###\n",
    "####################\n",
    "\n",
    "# X = X-matrix / feature-matrix / data-matrix... It's a tensor to hold our email\n",
    "# data. 'None' here means that we can hold any number of emails\n",
    "X = tf.placeholder(tf.float32, [None, numFeatures])\n",
    "# yGold = Y-matrix / label-matrix / labels... This will be our correct answers\n",
    "# matrix. Every row has either [1,0] for SPAM or [0,1] for HAM. 'None' here \n",
    "# means that we can hold any number of emails\n",
    "yGold = tf.placeholder(tf.float32, [None, numLabels])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "#################\n",
    "### VARIABLES ###\n",
    "#################\n",
    "\n",
    "# Values are randomly sampled from a Gaussian with a standard deviation of:\n",
    "#     sqrt(6 / (numInputNodes + numOutputNodes + 1))\n",
    "\n",
    "weights = tf.Variable(tf.random_normal([numFeatures,numLabels],\n",
    "                                       mean=0,\n",
    "                                       stddev=(np.sqrt(6/numFeatures+\n",
    "                                                         numLabels+1)),\n",
    "                                       name=\"weights\"))\n",
    "\n",
    "bias = tf.Variable(tf.random_normal([1,numLabels],\n",
    "                                    mean=0,\n",
    "                                    stddev=(np.sqrt(6/numFeatures+numLabels+1)),\n",
    "                                    name=\"bias\"))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "######################\n",
    "### PREDICTION OPS ###\n",
    "######################\n",
    "\n",
    "# INITIALIZE our weights and biases\n",
    "init_OP = tf.initialize_all_variables()\n",
    "\n",
    "# PREDICTION ALGORITHM i.e. FEEDFORWARD ALGORITHM\n",
    "apply_weights_OP = tf.matmul(X, weights, name=\"apply_weights\")\n",
    "add_bias_OP = tf.add(apply_weights_OP, bias, name=\"add_bias\") \n",
    "activation_OP = tf.nn.sigmoid(add_bias_OP, name=\"activation\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": [
    "#####################\n",
    "### EVALUATION OP ###\n",
    "#####################\n",
    "\n",
    "# COST FUNCTION i.e. MEAN SQUARED ERROR\n",
    "cost_OP = tf.nn.l2_loss(activation_OP-yGold, name=\"squared_error_cost\")\n",
    "cost_OP1 = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(add_bias_OP,yGold))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "#######################\n",
    "### OPTIMIZATION OP ###\n",
    "#######################\n",
    "\n",
    "# OPTIMIZATION ALGORITHM i.e. GRADIENT DESCENT\n",
    "training_OP = tf.train.GradientDescentOptimizer(learningRate).minimize(cost_OP)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {
    "collapsed": true
   },
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "WARNING:tensorflow:Passing a `GraphDef` to the SummaryWriter is deprecated. Pass a `Graph` object instead, such as `sess.graph`.\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "step 0, training accuracy 0.270142\n",
      "step 0, cost 96.8647\n",
      "step 0, change in cost 96.8647\n",
      "step 10, training accuracy 0.459716\n",
      "step 10, cost 79.9514\n",
      "step 10, change in cost 16.9133\n",
      "step 20, training accuracy 0.464455\n",
      "step 20, cost 77.043\n",
      "step 20, change in cost 2.90842\n",
      "step 30, training accuracy 0.469194\n",
      "step 30, cost 74.5681\n",
      "step 30, change in cost 2.47488\n",
      "step 40, training accuracy 0.478673\n",
      "step 40, cost 72.4668\n",
      "step 40, change in cost 2.10131\n",
      "step 50, training accuracy 0.483412\n",
      "step 50, cost 70.7135\n",
      "step 50, change in cost 1.75333\n",
      "step 60, training accuracy 0.488152\n",
      "step 60, cost 69.2601\n",
      "step 60, change in cost 1.45343\n",
      "step 70, training accuracy 0.49763\n",
      "step 70, cost 68.0427\n",
      "step 70, change in cost 1.2174\n",
      "step 80, training accuracy 0.50237\n",
      "step 80, cost 66.9908\n",
      "step 80, change in cost 1.0519\n",
      "step 90, training accuracy 0.50237\n",
      "step 90, cost 66.0308\n",
      "step 90, change in cost 0.95993\n",
      "step 100, training accuracy 0.50237\n",
      "step 100, cost 65.082\n",
      "step 100, change in cost 0.948837\n",
      "step 110, training accuracy 0.507109\n",
      "step 110, cost 64.0439\n",
      "step 110, change in cost 1.03815\n",
      "step 120, training accuracy 0.549763\n",
      "step 120, cost 62.775\n",
      "step 120, change in cost 1.26888\n",
      "step 130, training accuracy 0.57346\n",
      "step 130, cost 61.0716\n",
      "step 130, change in cost 1.70335\n",
      "step 140, training accuracy 0.611374\n",
      "step 140, cost 58.7052\n",
      "step 140, change in cost 2.36642\n",
      "step 150, training accuracy 0.654028\n",
      "step 150, cost 55.6549\n",
      "step 150, change in cost 3.0503\n",
      "step 160, training accuracy 0.677725\n",
      "step 160, cost 52.3331\n",
      "step 160, change in cost 3.32179\n",
      "step 170, training accuracy 0.687204\n",
      "step 170, cost 49.2056\n",
      "step 170, change in cost 3.12752\n",
      "step 180, training accuracy 0.696682\n",
      "step 180, cost 46.4471\n",
      "step 180, change in cost 2.75846\n",
      "step 190, training accuracy 0.71564\n",
      "step 190, cost 44.0567\n",
      "step 190, change in cost 2.39046\n",
      "step 200, training accuracy 0.753555\n",
      "step 200, cost 41.9875\n",
      "step 200, change in cost 2.06912\n",
      "step 210, training accuracy 0.758294\n",
      "step 210, cost 40.191\n",
      "step 210, change in cost 1.7965\n",
      "step 220, training accuracy 0.772512\n",
      "step 220, cost 38.6247\n",
      "step 220, change in cost 1.56636\n",
      "step 230, training accuracy 0.800948\n",
      "step 230, cost 37.2524\n",
      "step 230, change in cost 1.37234\n",
      "step 240, training accuracy 0.796209\n",
      "step 240, cost 36.0435\n",
      "step 240, change in cost 1.20885\n",
      "step 250, training accuracy 0.800948\n",
      "step 250, cost 34.9726\n",
      "step 250, change in cost 1.07091\n",
      "step 260, training accuracy 0.810427\n",
      "step 260, cost 34.0185\n",
      "step 260, change in cost 0.954071\n",
      "step 270, training accuracy 0.815166\n",
      "step 270, cost 33.1641\n",
      "step 270, change in cost 0.854412\n",
      "step 280, training accuracy 0.824645\n",
      "step 280, cost 32.3955\n",
      "step 280, change in cost 0.768646\n",
      "step 290, training accuracy 0.834123\n",
      "step 290, cost 31.7014\n",
      "step 290, change in cost 0.694103\n",
      "step 300, training accuracy 0.838863\n",
      "step 300, cost 31.0727\n",
      "step 300, change in cost 0.628672\n",
      "step 310, training accuracy 0.843602\n",
      "step 310, cost 30.5019\n",
      "step 310, change in cost 0.570824\n",
      "step 320, training accuracy 0.853081\n",
      "step 320, cost 29.9825\n",
      "step 320, change in cost 0.519384\n",
      "step 330, training accuracy 0.853081\n",
      "step 330, cost 29.509\n",
      "step 330, change in cost 0.473507\n",
      "step 340, training accuracy 0.853081\n",
      "step 340, cost 29.0765\n",
      "step 340, change in cost 0.432495\n",
      "step 350, training accuracy 0.853081\n",
      "step 350, cost 28.6807\n",
      "step 350, change in cost 0.395809\n",
      "step 360, training accuracy 0.843602\n",
      "step 360, cost 28.3177\n",
      "step 360, change in cost 0.36301\n",
      "step 370, training accuracy 0.843602\n",
      "step 370, cost 27.984\n",
      "step 370, change in cost 0.333622\n",
      "step 380, training accuracy 0.848341\n",
      "step 380, cost 27.6767\n",
      "step 380, change in cost 0.307323\n",
      "step 390, training accuracy 0.848341\n",
      "step 390, cost 27.393\n",
      "step 390, change in cost 0.28376\n",
      "step 400, training accuracy 0.848341\n",
      "step 400, cost 27.1303\n",
      "step 400, change in cost 0.262606\n",
      "step 410, training accuracy 0.848341\n",
      "step 410, cost 26.8867\n",
      "step 410, change in cost 0.243626\n",
      "step 420, training accuracy 0.848341\n",
      "step 420, cost 26.6602\n",
      "step 420, change in cost 0.226543\n",
      "step 430, training accuracy 0.848341\n",
      "step 430, cost 26.449\n",
      "step 430, change in cost 0.211155\n",
      "step 440, training accuracy 0.848341\n",
      "step 440, cost 26.2518\n",
      "step 440, change in cost 0.19726\n",
      "step 450, training accuracy 0.848341\n",
      "step 450, cost 26.0671\n",
      "step 450, change in cost 0.184702\n",
      "step 460, training accuracy 0.848341\n",
      "step 460, cost 25.8937\n",
      "step 460, change in cost 0.173315\n",
      "step 470, training accuracy 0.848341\n",
      "step 470, cost 25.7308\n",
      "step 470, change in cost 0.162971\n",
      "step 480, training accuracy 0.848341\n",
      "step 480, cost 25.5772\n",
      "step 480, change in cost 0.153561\n",
      "step 490, training accuracy 0.853081\n",
      "step 490, cost 25.4322\n",
      "step 490, change in cost 0.14497\n",
      "step 500, training accuracy 0.853081\n",
      "step 500, cost 25.2951\n",
      "step 500, change in cost 0.137123\n",
      "step 510, training accuracy 0.853081\n",
      "step 510, cost 25.1652\n",
      "step 510, change in cost 0.129925\n",
      "step 520, training accuracy 0.853081\n",
      "step 520, cost 25.0419\n",
      "step 520, change in cost 0.123306\n",
      "step 530, training accuracy 0.853081\n",
      "step 530, cost 24.9247\n",
      "step 530, change in cost 0.117216\n",
      "step 540, training accuracy 0.853081\n",
      "step 540, cost 24.8131\n",
      "step 540, change in cost 0.111593\n",
      "step 550, training accuracy 0.853081\n",
      "step 550, cost 24.7067\n",
      "step 550, change in cost 0.106388\n",
      "step 560, training accuracy 0.853081\n",
      "step 560, cost 24.6051\n",
      "step 560, change in cost 0.101561\n",
      "step 570, training accuracy 0.853081\n",
      "step 570, cost 24.5081\n",
      "step 570, change in cost 0.0970783\n",
      "step 580, training accuracy 0.853081\n",
      "step 580, cost 24.4152\n",
      "step 580, change in cost 0.0928898\n",
      "step 590, training accuracy 0.853081\n",
      "step 590, cost 24.3262\n",
      "step 590, change in cost 0.0889893\n",
      "step 600, training accuracy 0.853081\n",
      "step 600, cost 24.2408\n",
      "step 600, change in cost 0.0853329\n",
      "step 610, training accuracy 0.853081\n",
      "step 610, cost 24.1589\n",
      "step 610, change in cost 0.0819111\n",
      "step 620, training accuracy 0.853081\n",
      "step 620, cost 24.0802\n",
      "step 620, change in cost 0.0786934\n",
      "step 630, training accuracy 0.853081\n",
      "step 630, cost 24.0046\n",
      "step 630, change in cost 0.075655\n",
      "step 640, training accuracy 0.853081\n",
      "step 640, cost 23.9318\n",
      "step 640, change in cost 0.0728073\n",
      "step 650, training accuracy 0.853081\n",
      "step 650, cost 23.8617\n",
      "step 650, change in cost 0.0701084\n",
      "step 660, training accuracy 0.853081\n",
      "step 660, cost 23.7941\n",
      "step 660, change in cost 0.0675545\n",
      "step 670, training accuracy 0.853081\n",
      "step 670, cost 23.729\n",
      "step 670, change in cost 0.0651398\n",
      "step 680, training accuracy 0.853081\n",
      "step 680, cost 23.6661\n",
      "step 680, change in cost 0.0628452\n",
      "step 690, training accuracy 0.853081\n",
      "step 690, cost 23.6055\n",
      "step 690, change in cost 0.0606651\n",
      "step 700, training accuracy 0.85782\n",
      "step 700, cost 23.5469\n",
      "step 700, change in cost 0.0585957\n",
      "step 710, training accuracy 0.862559\n",
      "step 710, cost 23.4902\n",
      "step 710, change in cost 0.0566196\n",
      "step 720, training accuracy 0.862559\n",
      "step 720, cost 23.4355\n",
      "step 720, change in cost 0.0547409\n",
      "step 730, training accuracy 0.862559\n",
      "step 730, cost 23.3826\n",
      "step 730, change in cost 0.0529461\n",
      "step 740, training accuracy 0.862559\n",
      "step 740, cost 23.3313\n",
      "step 740, change in cost 0.0512333\n",
      "step 750, training accuracy 0.867299\n",
      "step 750, cost 23.2817\n",
      "step 750, change in cost 0.0495949\n",
      "step 760, training accuracy 0.867299\n",
      "step 760, cost 23.2337\n",
      "step 760, change in cost 0.0480232\n",
      "step 770, training accuracy 0.867299\n",
      "step 770, cost 23.1872\n",
      "step 770, change in cost 0.0465317\n",
      "step 780, training accuracy 0.867299\n",
      "step 780, cost 23.1421\n",
      "step 780, change in cost 0.0450802\n",
      "step 790, training accuracy 0.867299\n",
      "step 790, cost 23.0984\n",
      "step 790, change in cost 0.0437164\n",
      "step 800, training accuracy 0.867299\n",
      "step 800, cost 23.056\n",
      "step 800, change in cost 0.0423889\n",
      "step 810, training accuracy 0.867299\n",
      "step 810, cost 23.0149\n",
      "step 810, change in cost 0.0411167\n",
      "step 820, training accuracy 0.867299\n",
      "step 820, cost 22.975\n",
      "step 820, change in cost 0.0399017\n",
      "step 830, training accuracy 0.867299\n",
      "step 830, cost 22.9362\n",
      "step 830, change in cost 0.0387344\n",
      "step 840, training accuracy 0.867299\n",
      "step 840, cost 22.8986\n",
      "step 840, change in cost 0.037611\n",
      "step 850, training accuracy 0.867299\n",
      "step 850, cost 22.8621\n",
      "step 850, change in cost 0.0365295\n",
      "step 860, training accuracy 0.867299\n",
      "step 860, cost 22.8266\n",
      "step 860, change in cost 0.0354824\n",
      "step 870, training accuracy 0.867299\n",
      "step 870, cost 22.7921\n",
      "step 870, change in cost 0.0344868\n",
      "step 880, training accuracy 0.867299\n",
      "step 880, cost 22.7586\n",
      "step 880, change in cost 0.0335255\n",
      "step 890, training accuracy 0.867299\n",
      "step 890, cost 22.726\n",
      "step 890, change in cost 0.0325928\n",
      "step 900, training accuracy 0.867299\n",
      "step 900, cost 22.6943\n",
      "step 900, change in cost 0.0317097\n",
      "step 910, training accuracy 0.867299\n",
      "step 910, cost 22.6635\n",
      "step 910, change in cost 0.0308399\n",
      "step 920, training accuracy 0.867299\n",
      "step 920, cost 22.6334\n",
      "step 920, change in cost 0.0300121\n",
      "step 930, training accuracy 0.867299\n",
      "step 930, cost 22.6042\n",
      "step 930, change in cost 0.0292168\n",
      "step 940, training accuracy 0.867299\n",
      "step 940, cost 22.5758\n",
      "step 940, change in cost 0.0284405\n",
      "step 950, training accuracy 0.867299\n",
      "step 950, cost 22.5481\n",
      "step 950, change in cost 0.0277042\n",
      "step 960, training accuracy 0.867299\n",
      "step 960, cost 22.5211\n",
      "step 960, change in cost 0.0269833\n",
      "step 970, training accuracy 0.867299\n",
      "step 970, cost 22.4948\n",
      "step 970, change in cost 0.0262871\n",
      "step 980, training accuracy 0.867299\n",
      "step 980, cost 22.4692\n",
      "step 980, change in cost 0.0256252\n",
      "step 990, training accuracy 0.867299\n",
      "step 990, cost 22.4442\n",
      "step 990, change in cost 0.0249767\n",
      "step 1000, training accuracy 0.867299\n",
      "step 1000, cost 22.4199\n",
      "step 1000, change in cost 0.0243549\n",
      "step 1010, training accuracy 0.867299\n",
      "step 1010, cost 22.3961\n",
      "step 1010, change in cost 0.0237465\n",
      "step 1020, training accuracy 0.867299\n",
      "step 1020, cost 22.3729\n",
      "step 1020, change in cost 0.0231743\n",
      "step 1030, training accuracy 0.867299\n",
      "step 1030, cost 22.3503\n",
      "step 1030, change in cost 0.0226021\n",
      "step 1040, training accuracy 0.867299\n",
      "step 1040, cost 22.3283\n",
      "step 1040, change in cost 0.0220642\n",
      "step 1050, training accuracy 0.867299\n",
      "step 1050, cost 22.3067\n",
      "step 1050, change in cost 0.0215378\n",
      "step 1060, training accuracy 0.867299\n",
      "step 1060, cost 22.2857\n",
      "step 1060, change in cost 0.0210247\n",
      "step 1070, training accuracy 0.867299\n",
      "step 1070, cost 22.2652\n",
      "step 1070, change in cost 0.0205364\n",
      "step 1080, training accuracy 0.867299\n",
      "step 1080, cost 22.2451\n",
      "step 1080, change in cost 0.0200615\n",
      "step 1090, training accuracy 0.867299\n",
      "step 1090, cost 22.2255\n",
      "step 1090, change in cost 0.019598\n",
      "step 1100, training accuracy 0.867299\n",
      "step 1100, cost 22.2064\n",
      "step 1100, change in cost 0.0191536\n",
      "step 1110, training accuracy 0.872038\n",
      "step 1110, cost 22.1876\n",
      "step 1110, change in cost 0.0187168\n",
      "step 1120, training accuracy 0.872038\n",
      "step 1120, cost 22.1693\n",
      "step 1120, change in cost 0.0183029\n",
      "step 1130, training accuracy 0.872038\n",
      "step 1130, cost 22.1514\n",
      "step 1130, change in cost 0.0178909\n",
      "step 1140, training accuracy 0.872038\n",
      "step 1140, cost 22.1339\n",
      "step 1140, change in cost 0.0174999\n",
      "step 1150, training accuracy 0.872038\n",
      "step 1150, cost 22.1168\n",
      "step 1150, change in cost 0.0171204\n",
      "step 1160, training accuracy 0.872038\n",
      "step 1160, cost 22.1001\n",
      "step 1160, change in cost 0.0167484\n",
      "step 1170, training accuracy 0.872038\n",
      "step 1170, cost 22.0837\n",
      "step 1170, change in cost 0.0163918\n",
      "step 1180, training accuracy 0.872038\n",
      "step 1180, cost 22.0676\n",
      "step 1180, change in cost 0.0160408\n",
      "step 1190, training accuracy 0.872038\n",
      "step 1190, cost 22.0519\n",
      "step 1190, change in cost 0.0157032\n",
      "step 1200, training accuracy 0.872038\n",
      "step 1200, cost 22.0366\n",
      "step 1200, change in cost 0.015377\n",
      "step 1210, training accuracy 0.872038\n",
      "step 1210, cost 22.0215\n",
      "step 1210, change in cost 0.0150585\n",
      "step 1220, training accuracy 0.872038\n",
      "step 1220, cost 22.0068\n",
      "step 1220, change in cost 0.0147457\n",
      "step 1230, training accuracy 0.872038\n",
      "step 1230, cost 21.9923\n",
      "step 1230, change in cost 0.0144501\n",
      "step 1240, training accuracy 0.872038\n",
      "step 1240, cost 21.9782\n",
      "step 1240, change in cost 0.0141506\n",
      "step 1250, training accuracy 0.872038\n",
      "step 1250, cost 21.9643\n",
      "step 1250, change in cost 0.0138702\n",
      "step 1260, training accuracy 0.872038\n",
      "step 1260, cost 21.9507\n",
      "step 1260, change in cost 0.0135956\n",
      "step 1270, training accuracy 0.872038\n",
      "step 1270, cost 21.9374\n",
      "step 1270, change in cost 0.0133266\n",
      "step 1280, training accuracy 0.872038\n",
      "step 1280, cost 21.9243\n",
      "step 1280, change in cost 0.0130653\n",
      "step 1290, training accuracy 0.872038\n",
      "step 1290, cost 21.9115\n",
      "step 1290, change in cost 0.0128098\n",
      "step 1300, training accuracy 0.872038\n",
      "step 1300, cost 21.8989\n",
      "step 1300, change in cost 0.0125675\n",
      "step 1310, training accuracy 0.872038\n",
      "step 1310, cost 21.8866\n",
      "step 1310, change in cost 0.0123215\n",
      "step 1320, training accuracy 0.867299\n",
      "step 1320, cost 21.8745\n",
      "step 1320, change in cost 0.0120907\n",
      "step 1330, training accuracy 0.867299\n",
      "step 1330, cost 21.8627\n",
      "step 1330, change in cost 0.0118637\n",
      "step 1340, training accuracy 0.867299\n",
      "step 1340, cost 21.851\n",
      "step 1340, change in cost 0.0116386\n",
      "step 1350, training accuracy 0.867299\n",
      "step 1350, cost 21.8396\n",
      "step 1350, change in cost 0.0114231\n",
      "step 1360, training accuracy 0.867299\n",
      "step 1360, cost 21.8284\n",
      "step 1360, change in cost 0.0112133\n",
      "step 1370, training accuracy 0.867299\n",
      "step 1370, cost 21.8174\n",
      "step 1370, change in cost 0.0110092\n",
      "step 1380, training accuracy 0.867299\n",
      "step 1380, cost 21.8066\n",
      "step 1380, change in cost 0.0108051\n",
      "step 1390, training accuracy 0.867299\n",
      "step 1390, cost 21.796\n",
      "step 1390, change in cost 0.0106087\n",
      "step 1400, training accuracy 0.867299\n",
      "step 1400, cost 21.7855\n",
      "step 1400, change in cost 0.0104237\n",
      "step 1410, training accuracy 0.862559\n",
      "step 1410, cost 21.7753\n",
      "step 1410, change in cost 0.0102348\n",
      "step 1420, training accuracy 0.862559\n",
      "step 1420, cost 21.7652\n",
      "step 1420, change in cost 0.0100517\n",
      "step 1430, training accuracy 0.862559\n",
      "step 1430, cost 21.7554\n",
      "step 1430, change in cost 0.00987816\n",
      "step 1440, training accuracy 0.862559\n",
      "step 1440, cost 21.7457\n",
      "step 1440, change in cost 0.0097084\n",
      "step 1450, training accuracy 0.862559\n",
      "step 1450, cost 21.7361\n",
      "step 1450, change in cost 0.00953484\n",
      "step 1460, training accuracy 0.862559\n",
      "step 1460, cost 21.7267\n",
      "step 1460, change in cost 0.00937271\n",
      "step 1470, training accuracy 0.862559\n",
      "step 1470, cost 21.7175\n",
      "step 1470, change in cost 0.00921249\n",
      "step 1480, training accuracy 0.862559\n",
      "step 1480, cost 21.7085\n",
      "step 1480, change in cost 0.00905418\n",
      "step 1490, training accuracy 0.862559\n",
      "step 1490, cost 21.6996\n",
      "step 1490, change in cost 0.00889969\n",
      "step 1500, training accuracy 0.862559\n",
      "step 1500, cost 21.6908\n",
      "step 1500, change in cost 0.00875092\n",
      "step 1510, training accuracy 0.862559\n",
      "step 1510, cost 21.6822\n",
      "step 1510, change in cost 0.00860214\n",
      "step 1520, training accuracy 0.862559\n",
      "step 1520, cost 21.6738\n",
      "step 1520, change in cost 0.00846291\n",
      "step 1530, training accuracy 0.862559\n",
      "step 1530, cost 21.6655\n",
      "step 1530, change in cost 0.00831604\n",
      "step 1540, training accuracy 0.862559\n",
      "step 1540, cost 21.6573\n",
      "step 1540, change in cost 0.00818634\n",
      "step 1550, training accuracy 0.862559\n",
      "step 1550, cost 21.6492\n",
      "step 1550, change in cost 0.00805283\n",
      "step 1560, training accuracy 0.862559\n",
      "step 1560, cost 21.6413\n",
      "step 1560, change in cost 0.00792122\n",
      "step 1570, training accuracy 0.862559\n",
      "step 1570, cost 21.6335\n",
      "step 1570, change in cost 0.0077858\n",
      "step 1580, training accuracy 0.862559\n",
      "step 1580, cost 21.6258\n",
      "step 1580, change in cost 0.00766945\n",
      "step 1590, training accuracy 0.862559\n",
      "step 1590, cost 21.6183\n",
      "step 1590, change in cost 0.00753975\n",
      "step 1600, training accuracy 0.862559\n",
      "step 1600, cost 21.6109\n",
      "step 1600, change in cost 0.00742722\n",
      "step 1610, training accuracy 0.862559\n",
      "step 1610, cost 21.6036\n",
      "step 1610, change in cost 0.00730705\n",
      "step 1620, training accuracy 0.862559\n",
      "step 1620, cost 21.5964\n",
      "step 1620, change in cost 0.0071888\n",
      "step 1630, training accuracy 0.862559\n",
      "step 1630, cost 21.5893\n",
      "step 1630, change in cost 0.00707817\n",
      "step 1640, training accuracy 0.862559\n",
      "step 1640, cost 21.5823\n",
      "step 1640, change in cost 0.00696564\n",
      "step 1650, training accuracy 0.862559\n",
      "step 1650, cost 21.5755\n",
      "step 1650, change in cost 0.00685883\n",
      "step 1660, training accuracy 0.862559\n",
      "step 1660, cost 21.5687\n",
      "step 1660, change in cost 0.00675392\n",
      "step 1670, training accuracy 0.862559\n",
      "step 1670, cost 21.5621\n",
      "step 1670, change in cost 0.00665092\n",
      "step 1680, training accuracy 0.862559\n",
      "step 1680, cost 21.5555\n",
      "step 1680, change in cost 0.00654411\n",
      "step 1690, training accuracy 0.867299\n",
      "step 1690, cost 21.5491\n",
      "step 1690, change in cost 0.00644875\n",
      "step 1700, training accuracy 0.867299\n",
      "step 1700, cost 21.5427\n",
      "step 1700, change in cost 0.00635338\n",
      "step 1710, training accuracy 0.867299\n",
      "step 1710, cost 21.5365\n",
      "step 1710, change in cost 0.0062542\n",
      "step 1720, training accuracy 0.867299\n",
      "step 1720, cost 21.5303\n",
      "step 1720, change in cost 0.00615692\n",
      "step 1730, training accuracy 0.867299\n",
      "step 1730, cost 21.5242\n",
      "step 1730, change in cost 0.00606537\n",
      "step 1740, training accuracy 0.867299\n",
      "step 1740, cost 21.5183\n",
      "step 1740, change in cost 0.00597763\n",
      "step 1750, training accuracy 0.867299\n",
      "step 1750, cost 21.5124\n",
      "step 1750, change in cost 0.00588989\n",
      "step 1760, training accuracy 0.867299\n",
      "step 1760, cost 21.5066\n",
      "step 1760, change in cost 0.00580215\n",
      "step 1770, training accuracy 0.867299\n",
      "step 1770, cost 21.5009\n",
      "step 1770, change in cost 0.00571823\n",
      "step 1780, training accuracy 0.867299\n",
      "step 1780, cost 21.4952\n",
      "step 1780, change in cost 0.00563431\n",
      "step 1790, training accuracy 0.867299\n",
      "step 1790, cost 21.4897\n",
      "step 1790, change in cost 0.00554848\n",
      "step 1800, training accuracy 0.867299\n",
      "step 1800, cost 21.4842\n",
      "step 1800, change in cost 0.00546837\n",
      "step 1810, training accuracy 0.867299\n",
      "step 1810, cost 21.4788\n",
      "step 1810, change in cost 0.00539398\n",
      "step 1820, training accuracy 0.867299\n",
      "step 1820, cost 21.4735\n",
      "step 1820, change in cost 0.00531769\n",
      "step 1830, training accuracy 0.867299\n",
      "step 1830, cost 21.4683\n",
      "step 1830, change in cost 0.00522995\n",
      "step 1840, training accuracy 0.867299\n",
      "step 1840, cost 21.4631\n",
      "step 1840, change in cost 0.0051651\n",
      "step 1850, training accuracy 0.867299\n",
      "step 1850, cost 21.458\n",
      "step 1850, change in cost 0.00508881\n",
      "step 1860, training accuracy 0.867299\n",
      "step 1860, cost 21.453\n",
      "step 1860, change in cost 0.00501251\n",
      "step 1870, training accuracy 0.867299\n",
      "step 1870, cost 21.448\n",
      "step 1870, change in cost 0.00494957\n",
      "step 1880, training accuracy 0.867299\n",
      "step 1880, cost 21.4432\n",
      "step 1880, change in cost 0.00487328\n",
      "step 1890, training accuracy 0.867299\n",
      "step 1890, cost 21.4384\n",
      "step 1890, change in cost 0.00481033\n",
      "step 1900, training accuracy 0.867299\n",
      "step 1900, cost 21.4336\n",
      "step 1900, change in cost 0.00473785\n",
      "step 1910, training accuracy 0.867299\n",
      "step 1910, cost 21.4289\n",
      "step 1910, change in cost 0.00467682\n",
      "step 1920, training accuracy 0.862559\n",
      "step 1920, cost 21.4243\n",
      "step 1920, change in cost 0.00461006\n",
      "step 1930, training accuracy 0.862559\n",
      "step 1930, cost 21.4198\n",
      "step 1930, change in cost 0.00454903\n",
      "step 1940, training accuracy 0.862559\n",
      "step 1940, cost 21.4153\n",
      "step 1940, change in cost 0.00447845\n",
      "step 1950, training accuracy 0.862559\n",
      "step 1950, cost 21.4109\n",
      "step 1950, change in cost 0.00442886\n",
      "step 1960, training accuracy 0.862559\n",
      "step 1960, cost 21.4065\n",
      "step 1960, change in cost 0.0043602\n",
      "step 1970, training accuracy 0.862559\n",
      "step 1970, cost 21.4022\n",
      "step 1970, change in cost 0.00430298\n",
      "step 1980, training accuracy 0.862559\n",
      "step 1980, cost 21.398\n",
      "step 1980, change in cost 0.00424576\n",
      "step 1990, training accuracy 0.862559\n",
      "step 1990, cost 21.3938\n",
      "step 1990, change in cost 0.00418472\n",
      "step 2000, training accuracy 0.862559\n",
      "step 2000, cost 21.3897\n",
      "step 2000, change in cost 0.00413132\n",
      "step 2010, training accuracy 0.862559\n",
      "step 2010, cost 21.3856\n",
      "step 2010, change in cost 0.00407028\n",
      "step 2020, training accuracy 0.862559\n",
      "step 2020, cost 21.3816\n",
      "step 2020, change in cost 0.00402451\n",
      "step 2030, training accuracy 0.862559\n",
      "step 2030, cost 21.3776\n",
      "step 2030, change in cost 0.00396729\n",
      "step 2040, training accuracy 0.862559\n",
      "step 2040, cost 21.3737\n",
      "step 2040, change in cost 0.00391388\n",
      "step 2050, training accuracy 0.862559\n",
      "step 2050, cost 21.3698\n",
      "step 2050, change in cost 0.00386238\n",
      "step 2060, training accuracy 0.862559\n",
      "step 2060, cost 21.366\n",
      "step 2060, change in cost 0.00381088\n",
      "step 2070, training accuracy 0.862559\n",
      "step 2070, cost 21.3622\n",
      "step 2070, change in cost 0.00376701\n",
      "step 2080, training accuracy 0.862559\n",
      "step 2080, cost 21.3585\n",
      "step 2080, change in cost 0.00370789\n",
      "step 2090, training accuracy 0.862559\n",
      "step 2090, cost 21.3549\n",
      "step 2090, change in cost 0.00366592\n",
      "step 2100, training accuracy 0.862559\n",
      "step 2100, cost 21.3513\n",
      "step 2100, change in cost 0.00361252\n",
      "step 2110, training accuracy 0.862559\n",
      "step 2110, cost 21.3477\n",
      "step 2110, change in cost 0.00357437\n",
      "step 2120, training accuracy 0.862559\n",
      "step 2120, cost 21.3442\n",
      "step 2120, change in cost 0.00352287\n",
      "step 2130, training accuracy 0.862559\n",
      "step 2130, cost 21.3407\n",
      "step 2130, change in cost 0.003479\n",
      "step 2140, training accuracy 0.862559\n",
      "step 2140, cost 21.3372\n",
      "step 2140, change in cost 0.00343513\n",
      "step 2150, training accuracy 0.862559\n",
      "step 2150, cost 21.3339\n",
      "step 2150, change in cost 0.00338936\n",
      "step 2160, training accuracy 0.862559\n",
      "step 2160, cost 21.3305\n",
      "step 2160, change in cost 0.00334549\n",
      "step 2170, training accuracy 0.867299\n",
      "step 2170, cost 21.3272\n",
      "step 2170, change in cost 0.00330544\n",
      "step 2180, training accuracy 0.867299\n",
      "step 2180, cost 21.3239\n",
      "step 2180, change in cost 0.00326157\n",
      "step 2190, training accuracy 0.867299\n",
      "step 2190, cost 21.3207\n",
      "step 2190, change in cost 0.00322342\n",
      "step 2200, training accuracy 0.867299\n",
      "step 2200, cost 21.3175\n",
      "step 2200, change in cost 0.00318146\n",
      "step 2210, training accuracy 0.867299\n",
      "step 2210, cost 21.3144\n",
      "step 2210, change in cost 0.0031395\n",
      "step 2220, training accuracy 0.867299\n",
      "step 2220, cost 21.3113\n",
      "step 2220, change in cost 0.00310135\n",
      "step 2230, training accuracy 0.867299\n",
      "step 2230, cost 21.3082\n",
      "step 2230, change in cost 0.00306892\n",
      "step 2240, training accuracy 0.867299\n",
      "step 2240, cost 21.3052\n",
      "step 2240, change in cost 0.00302505\n",
      "step 2250, training accuracy 0.867299\n",
      "step 2250, cost 21.3022\n",
      "step 2250, change in cost 0.00298882\n",
      "step 2260, training accuracy 0.867299\n",
      "step 2260, cost 21.2993\n",
      "step 2260, change in cost 0.00295258\n",
      "step 2270, training accuracy 0.867299\n",
      "step 2270, cost 21.2963\n",
      "step 2270, change in cost 0.00291634\n",
      "step 2280, training accuracy 0.867299\n",
      "step 2280, cost 21.2935\n",
      "step 2280, change in cost 0.00287819\n",
      "step 2290, training accuracy 0.867299\n",
      "step 2290, cost 21.2906\n",
      "step 2290, change in cost 0.00284576\n",
      "step 2300, training accuracy 0.867299\n",
      "step 2300, cost 21.2878\n",
      "step 2300, change in cost 0.00281143\n",
      "step 2310, training accuracy 0.867299\n",
      "step 2310, cost 21.285\n",
      "step 2310, change in cost 0.00277901\n",
      "step 2320, training accuracy 0.867299\n",
      "step 2320, cost 21.2823\n",
      "step 2320, change in cost 0.00274467\n",
      "step 2330, training accuracy 0.867299\n",
      "step 2330, cost 21.2796\n",
      "step 2330, change in cost 0.00271034\n",
      "step 2340, training accuracy 0.867299\n",
      "step 2340, cost 21.2769\n",
      "step 2340, change in cost 0.00267982\n",
      "step 2350, training accuracy 0.867299\n",
      "step 2350, cost 21.2743\n",
      "step 2350, change in cost 0.00264359\n",
      "step 2360, training accuracy 0.867299\n",
      "step 2360, cost 21.2716\n",
      "step 2360, change in cost 0.00261688\n",
      "step 2370, training accuracy 0.867299\n",
      "step 2370, cost 21.2691\n",
      "step 2370, change in cost 0.00258255\n",
      "step 2380, training accuracy 0.867299\n",
      "step 2380, cost 21.2665\n",
      "step 2380, change in cost 0.00255394\n",
      "step 2390, training accuracy 0.867299\n",
      "step 2390, cost 21.264\n",
      "step 2390, change in cost 0.00252724\n",
      "step 2400, training accuracy 0.867299\n",
      "step 2400, cost 21.2615\n",
      "step 2400, change in cost 0.00249481\n",
      "step 2410, training accuracy 0.867299\n",
      "step 2410, cost 21.259\n",
      "step 2410, change in cost 0.00246429\n",
      "step 2420, training accuracy 0.867299\n",
      "step 2420, cost 21.2566\n",
      "step 2420, change in cost 0.00243759\n",
      "step 2430, training accuracy 0.867299\n",
      "step 2430, cost 21.2542\n",
      "step 2430, change in cost 0.00240707\n",
      "step 2440, training accuracy 0.867299\n",
      "step 2440, cost 21.2518\n",
      "step 2440, change in cost 0.00238037\n",
      "step 2450, training accuracy 0.867299\n",
      "step 2450, cost 21.2494\n",
      "step 2450, change in cost 0.00235367\n",
      "step 2460, training accuracy 0.867299\n",
      "step 2460, cost 21.2471\n",
      "step 2460, change in cost 0.00233078\n",
      "step 2470, training accuracy 0.867299\n",
      "step 2470, cost 21.2448\n",
      "step 2470, change in cost 0.00229836\n",
      "step 2480, training accuracy 0.867299\n",
      "step 2480, cost 21.2425\n",
      "step 2480, change in cost 0.00227928\n",
      "step 2490, training accuracy 0.867299\n",
      "step 2490, cost 21.2403\n",
      "step 2490, change in cost 0.00224304\n",
      "step 2500, training accuracy 0.867299\n",
      "step 2500, cost 21.2381\n",
      "step 2500, change in cost 0.00222397\n",
      "step 2510, training accuracy 0.867299\n",
      "step 2510, cost 21.2359\n",
      "step 2510, change in cost 0.00219917\n",
      "step 2520, training accuracy 0.867299\n",
      "step 2520, cost 21.2337\n",
      "step 2520, change in cost 0.00217247\n",
      "step 2530, training accuracy 0.867299\n",
      "step 2530, cost 21.2315\n",
      "step 2530, change in cost 0.00215149\n",
      "step 2540, training accuracy 0.867299\n",
      "step 2540, cost 21.2294\n",
      "step 2540, change in cost 0.00212097\n",
      "step 2550, training accuracy 0.867299\n",
      "step 2550, cost 21.2273\n",
      "step 2550, change in cost 0.00210381\n",
      "step 2560, training accuracy 0.867299\n",
      "step 2560, cost 21.2252\n",
      "step 2560, change in cost 0.00208092\n",
      "step 2570, training accuracy 0.867299\n",
      "step 2570, cost 21.2232\n",
      "step 2570, change in cost 0.00205612\n",
      "step 2580, training accuracy 0.867299\n",
      "step 2580, cost 21.2211\n",
      "step 2580, change in cost 0.00203323\n",
      "step 2590, training accuracy 0.867299\n",
      "step 2590, cost 21.2191\n",
      "step 2590, change in cost 0.00201035\n",
      "step 2600, training accuracy 0.867299\n",
      "step 2600, cost 21.2171\n",
      "step 2600, change in cost 0.00198936\n",
      "step 2610, training accuracy 0.867299\n",
      "step 2610, cost 21.2152\n",
      "step 2610, change in cost 0.00197029\n",
      "step 2620, training accuracy 0.867299\n",
      "step 2620, cost 21.2132\n",
      "step 2620, change in cost 0.00194931\n",
      "step 2630, training accuracy 0.867299\n",
      "step 2630, cost 21.2113\n",
      "step 2630, change in cost 0.0019207\n",
      "step 2640, training accuracy 0.867299\n",
      "step 2640, cost 21.2094\n",
      "step 2640, change in cost 0.00190926\n",
      "step 2650, training accuracy 0.867299\n",
      "step 2650, cost 21.2075\n",
      "step 2650, change in cost 0.00188446\n",
      "step 2660, training accuracy 0.867299\n",
      "step 2660, cost 21.2056\n",
      "step 2660, change in cost 0.00186157\n",
      "step 2670, training accuracy 0.867299\n",
      "step 2670, cost 21.2038\n",
      "step 2670, change in cost 0.00184822\n",
      "step 2680, training accuracy 0.867299\n",
      "step 2680, cost 21.202\n",
      "step 2680, change in cost 0.00182152\n",
      "step 2690, training accuracy 0.867299\n",
      "step 2690, cost 21.2002\n",
      "step 2690, change in cost 0.00181198\n",
      "step 2700, training accuracy 0.867299\n",
      "step 2700, cost 21.1984\n",
      "step 2700, change in cost 0.00178528\n",
      "step 2710, training accuracy 0.867299\n",
      "step 2710, cost 21.1966\n",
      "step 2710, change in cost 0.00176811\n",
      "step 2720, training accuracy 0.867299\n",
      "step 2720, cost 21.1949\n",
      "step 2720, change in cost 0.00174904\n",
      "step 2730, training accuracy 0.867299\n",
      "step 2730, cost 21.1931\n",
      "step 2730, change in cost 0.00173569\n",
      "step 2740, training accuracy 0.867299\n",
      "step 2740, cost 21.1914\n",
      "step 2740, change in cost 0.0017128\n",
      "step 2750, training accuracy 0.867299\n",
      "step 2750, cost 21.1897\n",
      "step 2750, change in cost 0.00169945\n",
      "step 2760, training accuracy 0.867299\n",
      "step 2760, cost 21.188\n",
      "step 2760, change in cost 0.00167656\n",
      "step 2770, training accuracy 0.867299\n",
      "step 2770, cost 21.1864\n",
      "step 2770, change in cost 0.00166321\n",
      "step 2780, training accuracy 0.867299\n",
      "step 2780, cost 21.1847\n",
      "step 2780, change in cost 0.00164795\n",
      "step 2790, training accuracy 0.867299\n",
      "step 2790, cost 21.1831\n",
      "step 2790, change in cost 0.00162506\n",
      "step 2800, training accuracy 0.867299\n",
      "step 2800, cost 21.1815\n",
      "step 2800, change in cost 0.00161362\n",
      "step 2810, training accuracy 0.867299\n",
      "step 2810, cost 21.1799\n",
      "step 2810, change in cost 0.00159454\n",
      "step 2820, training accuracy 0.867299\n",
      "step 2820, cost 21.1783\n",
      "step 2820, change in cost 0.00157928\n",
      "step 2830, training accuracy 0.867299\n",
      "step 2830, cost 21.1767\n",
      "step 2830, change in cost 0.00156784\n",
      "step 2840, training accuracy 0.867299\n",
      "step 2840, cost 21.1752\n",
      "step 2840, change in cost 0.00154877\n",
      "step 2850, training accuracy 0.867299\n",
      "step 2850, cost 21.1737\n",
      "step 2850, change in cost 0.00153351\n",
      "step 2860, training accuracy 0.867299\n",
      "step 2860, cost 21.1721\n",
      "step 2860, change in cost 0.00151443\n",
      "step 2870, training accuracy 0.867299\n",
      "step 2870, cost 21.1706\n",
      "step 2870, change in cost 0.0015049\n",
      "step 2880, training accuracy 0.867299\n",
      "step 2880, cost 21.1692\n",
      "step 2880, change in cost 0.00148964\n",
      "step 2890, training accuracy 0.867299\n",
      "step 2890, cost 21.1677\n",
      "step 2890, change in cost 0.00147438\n",
      "step 2900, training accuracy 0.867299\n",
      "step 2900, cost 21.1662\n",
      "step 2900, change in cost 0.00145531\n",
      "step 2910, training accuracy 0.867299\n",
      "step 2910, cost 21.1648\n",
      "step 2910, change in cost 0.00144958\n",
      "step 2920, training accuracy 0.867299\n",
      "step 2920, cost 21.1633\n",
      "step 2920, change in cost 0.00143051\n",
      "step 2930, training accuracy 0.867299\n",
      "step 2930, cost 21.1619\n",
      "step 2930, change in cost 0.00141525\n",
      "step 2940, training accuracy 0.867299\n",
      "step 2940, cost 21.1605\n",
      "step 2940, change in cost 0.0014019\n",
      "step 2950, training accuracy 0.867299\n",
      "step 2950, cost 21.1591\n",
      "step 2950, change in cost 0.00139427\n",
      "step 2960, training accuracy 0.867299\n",
      "step 2960, cost 21.1578\n",
      "step 2960, change in cost 0.00137711\n",
      "step 2970, training accuracy 0.867299\n",
      "step 2970, cost 21.1564\n",
      "step 2970, change in cost 0.00135994\n",
      "step 2980, training accuracy 0.867299\n",
      "step 2980, cost 21.155\n",
      "step 2980, change in cost 0.0013504\n",
      "step 2990, training accuracy 0.867299\n",
      "step 2990, cost 21.1537\n",
      "step 2990, change in cost 0.00133896\n",
      "step 3000, training accuracy 0.867299\n",
      "step 3000, cost 21.1524\n",
      "step 3000, change in cost 0.00132942\n",
      "step 3010, training accuracy 0.867299\n",
      "step 3010, cost 21.1511\n",
      "step 3010, change in cost 0.00130844\n",
      "step 3020, training accuracy 0.867299\n",
      "step 3020, cost 21.1498\n",
      "step 3020, change in cost 0.00130081\n",
      "step 3030, training accuracy 0.867299\n",
      "step 3030, cost 21.1485\n",
      "step 3030, change in cost 0.00128555\n",
      "step 3040, training accuracy 0.867299\n",
      "step 3040, cost 21.1472\n",
      "step 3040, change in cost 0.00128174\n",
      "step 3050, training accuracy 0.867299\n",
      "step 3050, cost 21.1459\n",
      "step 3050, change in cost 0.00125885\n",
      "step 3060, training accuracy 0.867299\n",
      "step 3060, cost 21.1447\n",
      "step 3060, change in cost 0.00125885\n",
      "step 3070, training accuracy 0.867299\n",
      "step 3070, cost 21.1434\n",
      "step 3070, change in cost 0.00123787\n",
      "step 3080, training accuracy 0.867299\n",
      "step 3080, cost 21.1422\n",
      "step 3080, change in cost 0.00123024\n",
      "step 3090, training accuracy 0.867299\n",
      "step 3090, cost 21.141\n",
      "step 3090, change in cost 0.0012188\n",
      "step 3100, training accuracy 0.867299\n",
      "step 3100, cost 21.1398\n",
      "step 3100, change in cost 0.00120926\n",
      "step 3110, training accuracy 0.867299\n",
      "step 3110, cost 21.1386\n",
      "step 3110, change in cost 0.00119591\n",
      "step 3120, training accuracy 0.867299\n",
      "step 3120, cost 21.1374\n",
      "step 3120, change in cost 0.00119019\n",
      "step 3130, training accuracy 0.867299\n",
      "step 3130, cost 21.1362\n",
      "step 3130, change in cost 0.00117302\n",
      "step 3140, training accuracy 0.867299\n",
      "step 3140, cost 21.1351\n",
      "step 3140, change in cost 0.00116158\n",
      "step 3150, training accuracy 0.867299\n",
      "step 3150, cost 21.1339\n",
      "step 3150, change in cost 0.00115585\n",
      "step 3160, training accuracy 0.867299\n",
      "step 3160, cost 21.1328\n",
      "step 3160, change in cost 0.00114441\n",
      "step 3170, training accuracy 0.867299\n",
      "step 3170, cost 21.1316\n",
      "step 3170, change in cost 0.00113106\n",
      "step 3180, training accuracy 0.867299\n",
      "step 3180, cost 21.1305\n",
      "step 3180, change in cost 0.00112343\n",
      "step 3190, training accuracy 0.867299\n",
      "step 3190, cost 21.1294\n",
      "step 3190, change in cost 0.00111389\n",
      "step 3200, training accuracy 0.867299\n",
      "step 3200, cost 21.1283\n",
      "step 3200, change in cost 0.00110626\n",
      "step 3210, training accuracy 0.867299\n",
      "step 3210, cost 21.1272\n",
      "step 3210, change in cost 0.001091\n",
      "step 3220, training accuracy 0.867299\n",
      "step 3220, cost 21.1261\n",
      "step 3220, change in cost 0.00108719\n",
      "step 3230, training accuracy 0.867299\n",
      "step 3230, cost 21.125\n",
      "step 3230, change in cost 0.00107384\n",
      "step 3240, training accuracy 0.867299\n",
      "step 3240, cost 21.124\n",
      "step 3240, change in cost 0.0010643\n",
      "step 3250, training accuracy 0.867299\n",
      "step 3250, cost 21.1229\n",
      "step 3250, change in cost 0.00105476\n",
      "step 3260, training accuracy 0.867299\n",
      "step 3260, cost 21.1219\n",
      "step 3260, change in cost 0.00104904\n",
      "step 3270, training accuracy 0.867299\n",
      "step 3270, cost 21.1208\n",
      "step 3270, change in cost 0.00103951\n",
      "step 3280, training accuracy 0.867299\n",
      "step 3280, cost 21.1198\n",
      "step 3280, change in cost 0.00102997\n",
      "step 3290, training accuracy 0.867299\n",
      "step 3290, cost 21.1188\n",
      "step 3290, change in cost 0.00101662\n",
      "step 3300, training accuracy 0.867299\n",
      "step 3300, cost 21.1178\n",
      "step 3300, change in cost 0.00101471\n",
      "step 3310, training accuracy 0.867299\n",
      "step 3310, cost 21.1168\n",
      "step 3310, change in cost 0.000999451\n",
      "step 3320, training accuracy 0.867299\n",
      "step 3320, cost 21.1158\n",
      "step 3320, change in cost 0.000995636\n",
      "step 3330, training accuracy 0.867299\n",
      "step 3330, cost 21.1148\n",
      "step 3330, change in cost 0.000991821\n",
      "step 3340, training accuracy 0.872038\n",
      "step 3340, cost 21.1138\n",
      "step 3340, change in cost 0.000972748\n",
      "step 3350, training accuracy 0.872038\n",
      "step 3350, cost 21.1128\n",
      "step 3350, change in cost 0.00097084\n",
      "step 3360, training accuracy 0.872038\n",
      "step 3360, cost 21.1119\n",
      "step 3360, change in cost 0.000959396\n",
      "step 3370, training accuracy 0.872038\n",
      "step 3370, cost 21.1109\n",
      "step 3370, change in cost 0.000953674\n",
      "step 3380, training accuracy 0.872038\n",
      "step 3380, cost 21.11\n",
      "step 3380, change in cost 0.000944138\n",
      "step 3390, training accuracy 0.872038\n",
      "step 3390, cost 21.109\n",
      "step 3390, change in cost 0.000940323\n",
      "step 3400, training accuracy 0.872038\n",
      "step 3400, cost 21.1081\n",
      "step 3400, change in cost 0.000926971\n",
      "step 3410, training accuracy 0.872038\n",
      "step 3410, cost 21.1072\n",
      "step 3410, change in cost 0.000923157\n",
      "step 3420, training accuracy 0.872038\n",
      "step 3420, cost 21.1063\n",
      "step 3420, change in cost 0.00091362\n",
      "step 3430, training accuracy 0.872038\n",
      "step 3430, cost 21.1054\n",
      "step 3430, change in cost 0.000909805\n",
      "step 3440, training accuracy 0.872038\n",
      "step 3440, cost 21.1045\n",
      "step 3440, change in cost 0.000898361\n",
      "step 3450, training accuracy 0.872038\n",
      "step 3450, cost 21.1036\n",
      "step 3450, change in cost 0.000888824\n",
      "step 3460, training accuracy 0.872038\n",
      "step 3460, cost 21.1027\n",
      "step 3460, change in cost 0.000890732\n",
      "step 3470, training accuracy 0.872038\n",
      "step 3470, cost 21.1018\n",
      "step 3470, change in cost 0.000871658\n",
      "step 3480, training accuracy 0.872038\n",
      "step 3480, cost 21.1009\n",
      "step 3480, change in cost 0.000875473\n",
      "step 3490, training accuracy 0.872038\n",
      "step 3490, cost 21.1001\n",
      "step 3490, change in cost 0.000858307\n",
      "step 3500, training accuracy 0.872038\n",
      "step 3500, cost 21.0992\n",
      "step 3500, change in cost 0.0008564\n",
      "step 3510, training accuracy 0.872038\n",
      "step 3510, cost 21.0984\n",
      "step 3510, change in cost 0.000850677\n",
      "step 3520, training accuracy 0.872038\n",
      "step 3520, cost 21.0975\n",
      "step 3520, change in cost 0.000841141\n",
      "step 3530, training accuracy 0.872038\n",
      "step 3530, cost 21.0967\n",
      "step 3530, change in cost 0.000835419\n",
      "step 3540, training accuracy 0.872038\n",
      "step 3540, cost 21.0959\n",
      "step 3540, change in cost 0.000829697\n",
      "step 3550, training accuracy 0.872038\n",
      "step 3550, cost 21.095\n",
      "step 3550, change in cost 0.000825882\n",
      "step 3560, training accuracy 0.872038\n",
      "step 3560, cost 21.0942\n",
      "step 3560, change in cost 0.000814438\n",
      "step 3570, training accuracy 0.872038\n",
      "step 3570, cost 21.0934\n",
      "step 3570, change in cost 0.000810623\n",
      "step 3580, training accuracy 0.872038\n",
      "step 3580, cost 21.0926\n",
      "step 3580, change in cost 0.000804901\n",
      "step 3590, training accuracy 0.872038\n",
      "step 3590, cost 21.0918\n",
      "step 3590, change in cost 0.000797272\n",
      "step 3600, training accuracy 0.872038\n",
      "step 3600, cost 21.091\n",
      "step 3600, change in cost 0.000789642\n",
      "step 3610, training accuracy 0.872038\n",
      "step 3610, cost 21.0902\n",
      "step 3610, change in cost 0.000782013\n",
      "step 3620, training accuracy 0.872038\n",
      "step 3620, cost 21.0895\n",
      "step 3620, change in cost 0.000780106\n",
      "step 3630, training accuracy 0.872038\n",
      "step 3630, cost 21.0887\n",
      "step 3630, change in cost 0.000772476\n",
      "step 3640, training accuracy 0.872038\n",
      "step 3640, cost 21.0879\n",
      "step 3640, change in cost 0.000766754\n",
      "step 3650, training accuracy 0.872038\n",
      "step 3650, cost 21.0872\n",
      "step 3650, change in cost 0.000759125\n",
      "step 3660, training accuracy 0.872038\n",
      "step 3660, cost 21.0864\n",
      "step 3660, change in cost 0.00075531\n",
      "step 3670, training accuracy 0.872038\n",
      "step 3670, cost 21.0857\n",
      "step 3670, change in cost 0.000747681\n",
      "step 3680, training accuracy 0.872038\n",
      "step 3680, cost 21.0849\n",
      "step 3680, change in cost 0.000747681\n",
      "step 3690, training accuracy 0.872038\n",
      "step 3690, cost 21.0842\n",
      "step 3690, change in cost 0.000732422\n",
      "step 3700, training accuracy 0.872038\n",
      "step 3700, cost 21.0835\n",
      "step 3700, change in cost 0.000732422\n",
      "step 3710, training accuracy 0.872038\n",
      "step 3710, cost 21.0827\n",
      "step 3710, change in cost 0.000728607\n",
      "step 3720, training accuracy 0.872038\n",
      "step 3720, cost 21.082\n",
      "step 3720, change in cost 0.000717163\n",
      "step 3730, training accuracy 0.872038\n",
      "step 3730, cost 21.0813\n",
      "step 3730, change in cost 0.000717163\n",
      "step 3740, training accuracy 0.872038\n",
      "step 3740, cost 21.0806\n",
      "step 3740, change in cost 0.000709534\n",
      "step 3750, training accuracy 0.872038\n",
      "step 3750, cost 21.0799\n",
      "step 3750, change in cost 0.000705719\n",
      "step 3760, training accuracy 0.872038\n",
      "step 3760, cost 21.0792\n",
      "step 3760, change in cost 0.000694275\n",
      "step 3770, training accuracy 0.872038\n",
      "step 3770, cost 21.0785\n",
      "step 3770, change in cost 0.00069809\n",
      "step 3780, training accuracy 0.872038\n",
      "step 3780, cost 21.0778\n",
      "step 3780, change in cost 0.000688553\n",
      "step 3790, training accuracy 0.872038\n",
      "step 3790, cost 21.0771\n",
      "step 3790, change in cost 0.000684738\n",
      "step 3800, training accuracy 0.872038\n",
      "step 3800, cost 21.0764\n",
      "step 3800, change in cost 0.000679016\n",
      "step 3810, training accuracy 0.872038\n",
      "step 3810, cost 21.0758\n",
      "step 3810, change in cost 0.000669479\n",
      "step 3820, training accuracy 0.872038\n",
      "step 3820, cost 21.0751\n",
      "step 3820, change in cost 0.000669479\n",
      "step 3830, training accuracy 0.872038\n",
      "step 3830, cost 21.0744\n",
      "step 3830, change in cost 0.000663757\n",
      "step 3840, training accuracy 0.872038\n",
      "step 3840, cost 21.0738\n",
      "step 3840, change in cost 0.00066185\n",
      "step 3850, training accuracy 0.872038\n",
      "step 3850, cost 21.0731\n",
      "step 3850, change in cost 0.000648499\n",
      "step 3860, training accuracy 0.872038\n",
      "step 3860, cost 21.0725\n",
      "step 3860, change in cost 0.000652313\n",
      "step 3870, training accuracy 0.872038\n",
      "step 3870, cost 21.0718\n",
      "step 3870, change in cost 0.000646591\n",
      "step 3880, training accuracy 0.872038\n",
      "step 3880, cost 21.0712\n",
      "step 3880, change in cost 0.000637054\n",
      "step 3890, training accuracy 0.872038\n",
      "step 3890, cost 21.0705\n",
      "step 3890, change in cost 0.000637054\n",
      "step 3900, training accuracy 0.872038\n",
      "step 3900, cost 21.0699\n",
      "step 3900, change in cost 0.000629425\n",
      "step 3910, training accuracy 0.872038\n",
      "step 3910, cost 21.0693\n",
      "step 3910, change in cost 0.00062561\n",
      "step 3920, training accuracy 0.872038\n",
      "step 3920, cost 21.0687\n",
      "step 3920, change in cost 0.000619888\n",
      "step 3930, training accuracy 0.872038\n",
      "step 3930, cost 21.0681\n",
      "step 3930, change in cost 0.000616074\n",
      "step 3940, training accuracy 0.872038\n",
      "step 3940, cost 21.0674\n",
      "step 3940, change in cost 0.000614166\n",
      "step 3950, training accuracy 0.872038\n",
      "step 3950, cost 21.0668\n",
      "step 3950, change in cost 0.000608444\n",
      "step 3960, training accuracy 0.872038\n",
      "step 3960, cost 21.0662\n",
      "step 3960, change in cost 0.000600815\n",
      "step 3970, training accuracy 0.872038\n",
      "step 3970, cost 21.0656\n",
      "step 3970, change in cost 0.000602722\n",
      "step 3980, training accuracy 0.872038\n",
      "step 3980, cost 21.065\n",
      "step 3980, change in cost 0.000591278\n",
      "step 3990, training accuracy 0.872038\n",
      "step 3990, cost 21.0644\n",
      "step 3990, change in cost 0.000589371\n",
      "step 4000, training accuracy 0.872038\n",
      "step 4000, cost 21.0639\n",
      "step 4000, change in cost 0.000589371\n",
      "step 4010, training accuracy 0.872038\n",
      "step 4010, cost 21.0633\n",
      "step 4010, change in cost 0.000579834\n",
      "step 4020, training accuracy 0.872038\n",
      "step 4020, cost 21.0627\n",
      "step 4020, change in cost 0.000585556\n",
      "step 4030, training accuracy 0.872038\n",
      "step 4030, cost 21.0621\n",
      "step 4030, change in cost 0.000570297\n",
      "step 4040, training accuracy 0.872038\n",
      "step 4040, cost 21.0616\n",
      "step 4040, change in cost 0.00056839\n",
      "step 4050, training accuracy 0.872038\n",
      "step 4050, cost 21.061\n",
      "step 4050, change in cost 0.000566483\n",
      "step 4060, training accuracy 0.872038\n",
      "step 4060, cost 21.0604\n",
      "step 4060, change in cost 0.000562668\n",
      "step 4070, training accuracy 0.872038\n",
      "step 4070, cost 21.0599\n",
      "step 4070, change in cost 0.00056076\n",
      "step 4080, training accuracy 0.872038\n",
      "step 4080, cost 21.0593\n",
      "step 4080, change in cost 0.000551224\n",
      "step 4090, training accuracy 0.872038\n",
      "step 4090, cost 21.0588\n",
      "step 4090, change in cost 0.000551224\n",
      "step 4100, training accuracy 0.872038\n",
      "step 4100, cost 21.0582\n",
      "step 4100, change in cost 0.000545502\n",
      "step 4110, training accuracy 0.872038\n",
      "step 4110, cost 21.0577\n",
      "step 4110, change in cost 0.000541687\n",
      "step 4120, training accuracy 0.872038\n",
      "step 4120, cost 21.0571\n",
      "step 4120, change in cost 0.000543594\n",
      "step 4130, training accuracy 0.872038\n",
      "step 4130, cost 21.0566\n",
      "step 4130, change in cost 0.000535965\n",
      "step 4140, training accuracy 0.872038\n",
      "step 4140, cost 21.0561\n",
      "step 4140, change in cost 0.000530243\n",
      "step 4150, training accuracy 0.872038\n",
      "step 4150, cost 21.0555\n",
      "step 4150, change in cost 0.000530243\n",
      "step 4160, training accuracy 0.872038\n",
      "step 4160, cost 21.055\n",
      "step 4160, change in cost 0.000526428\n",
      "step 4170, training accuracy 0.872038\n",
      "step 4170, cost 21.0545\n",
      "step 4170, change in cost 0.000516891\n",
      "step 4180, training accuracy 0.872038\n",
      "step 4180, cost 21.054\n",
      "step 4180, change in cost 0.000516891\n",
      "step 4190, training accuracy 0.872038\n",
      "step 4190, cost 21.0535\n",
      "step 4190, change in cost 0.000513077\n",
      "step 4200, training accuracy 0.872038\n",
      "step 4200, cost 21.0529\n",
      "step 4200, change in cost 0.000511169\n",
      "step 4210, training accuracy 0.872038\n",
      "step 4210, cost 21.0524\n",
      "step 4210, change in cost 0.000511169\n",
      "step 4220, training accuracy 0.872038\n",
      "step 4220, cost 21.0519\n",
      "step 4220, change in cost 0.000497818\n",
      "step 4230, training accuracy 0.872038\n",
      "step 4230, cost 21.0514\n",
      "step 4230, change in cost 0.00050354\n",
      "step 4240, training accuracy 0.872038\n",
      "step 4240, cost 21.0509\n",
      "step 4240, change in cost 0.000497818\n",
      "step 4250, training accuracy 0.872038\n",
      "step 4250, cost 21.0504\n",
      "step 4250, change in cost 0.000490189\n",
      "step 4260, training accuracy 0.872038\n",
      "step 4260, cost 21.0499\n",
      "step 4260, change in cost 0.000495911\n",
      "step 4270, training accuracy 0.872038\n",
      "step 4270, cost 21.0495\n",
      "step 4270, change in cost 0.000484467\n",
      "step 4280, training accuracy 0.872038\n",
      "step 4280, cost 21.049\n",
      "step 4280, change in cost 0.000484467\n",
      "step 4290, training accuracy 0.872038\n",
      "step 4290, cost 21.0485\n",
      "step 4290, change in cost 0.000476837\n",
      "step 4300, training accuracy 0.872038\n",
      "step 4300, cost 21.048\n",
      "step 4300, change in cost 0.000480652\n",
      "step 4310, training accuracy 0.872038\n",
      "step 4310, cost 21.0476\n",
      "step 4310, change in cost 0.000473022\n",
      "step 4320, training accuracy 0.872038\n",
      "step 4320, cost 21.0471\n",
      "step 4320, change in cost 0.000469208\n",
      "step 4330, training accuracy 0.872038\n",
      "step 4330, cost 21.0466\n",
      "step 4330, change in cost 0.000471115\n",
      "step 4340, training accuracy 0.872038\n",
      "step 4340, cost 21.0461\n",
      "step 4340, change in cost 0.0004673\n",
      "step 4350, training accuracy 0.872038\n",
      "step 4350, cost 21.0457\n",
      "step 4350, change in cost 0.000461578\n",
      "step 4360, training accuracy 0.872038\n",
      "step 4360, cost 21.0452\n",
      "step 4360, change in cost 0.000457764\n",
      "step 4370, training accuracy 0.872038\n",
      "step 4370, cost 21.0448\n",
      "step 4370, change in cost 0.000455856\n",
      "step 4380, training accuracy 0.872038\n",
      "step 4380, cost 21.0443\n",
      "step 4380, change in cost 0.000452042\n",
      "step 4390, training accuracy 0.872038\n",
      "step 4390, cost 21.0439\n",
      "step 4390, change in cost 0.000450134\n",
      "step 4400, training accuracy 0.872038\n",
      "step 4400, cost 21.0434\n",
      "step 4400, change in cost 0.000442505\n",
      "step 4410, training accuracy 0.872038\n",
      "step 4410, cost 21.043\n",
      "step 4410, change in cost 0.000448227\n",
      "step 4420, training accuracy 0.872038\n",
      "step 4420, cost 21.0425\n",
      "step 4420, change in cost 0.000440598\n",
      "step 4430, training accuracy 0.872038\n",
      "step 4430, cost 21.0421\n",
      "step 4430, change in cost 0.000442505\n",
      "step 4440, training accuracy 0.872038\n",
      "step 4440, cost 21.0417\n",
      "step 4440, change in cost 0.000431061\n",
      "step 4450, training accuracy 0.872038\n",
      "step 4450, cost 21.0412\n",
      "step 4450, change in cost 0.000432968\n",
      "step 4460, training accuracy 0.872038\n",
      "step 4460, cost 21.0408\n",
      "step 4460, change in cost 0.000431061\n",
      "step 4470, training accuracy 0.872038\n",
      "step 4470, cost 21.0404\n",
      "step 4470, change in cost 0.000429153\n",
      "step 4480, training accuracy 0.872038\n",
      "step 4480, cost 21.0399\n",
      "step 4480, change in cost 0.000423431\n",
      "step 4490, training accuracy 0.872038\n",
      "step 4490, cost 21.0395\n",
      "step 4490, change in cost 0.000421524\n",
      "step 4500, training accuracy 0.872038\n",
      "step 4500, cost 21.0391\n",
      "step 4500, change in cost 0.000421524\n",
      "step 4510, training accuracy 0.872038\n",
      "step 4510, cost 21.0387\n",
      "step 4510, change in cost 0.000417709\n",
      "step 4520, training accuracy 0.872038\n",
      "step 4520, cost 21.0383\n",
      "step 4520, change in cost 0.00041008\n",
      "step 4530, training accuracy 0.872038\n",
      "step 4530, cost 21.0379\n",
      "step 4530, change in cost 0.000413895\n",
      "step 4540, training accuracy 0.872038\n",
      "step 4540, cost 21.0375\n",
      "step 4540, change in cost 0.000408173\n",
      "step 4550, training accuracy 0.872038\n",
      "step 4550, cost 21.037\n",
      "step 4550, change in cost 0.000402451\n",
      "step 4560, training accuracy 0.872038\n",
      "step 4560, cost 21.0366\n",
      "step 4560, change in cost 0.000408173\n",
      "step 4570, training accuracy 0.872038\n",
      "step 4570, cost 21.0362\n",
      "step 4570, change in cost 0.000398636\n",
      "step 4580, training accuracy 0.872038\n",
      "step 4580, cost 21.0358\n",
      "step 4580, change in cost 0.000400543\n",
      "step 4590, training accuracy 0.872038\n",
      "step 4590, cost 21.0354\n",
      "step 4590, change in cost 0.000398636\n",
      "step 4600, training accuracy 0.872038\n",
      "step 4600, cost 21.0351\n",
      "step 4600, change in cost 0.000391006\n",
      "step 4610, training accuracy 0.872038\n",
      "step 4610, cost 21.0347\n",
      "step 4610, change in cost 0.000391006\n",
      "step 4620, training accuracy 0.872038\n",
      "step 4620, cost 21.0343\n",
      "step 4620, change in cost 0.000389099\n",
      "step 4630, training accuracy 0.872038\n",
      "step 4630, cost 21.0339\n",
      "step 4630, change in cost 0.000385284\n",
      "step 4640, training accuracy 0.872038\n",
      "step 4640, cost 21.0335\n",
      "step 4640, change in cost 0.000385284\n",
      "step 4650, training accuracy 0.872038\n",
      "step 4650, cost 21.0331\n",
      "step 4650, change in cost 0.000379562\n",
      "step 4660, training accuracy 0.872038\n",
      "step 4660, cost 21.0327\n",
      "step 4660, change in cost 0.000379562\n",
      "step 4670, training accuracy 0.872038\n",
      "step 4670, cost 21.0324\n",
      "step 4670, change in cost 0.000377655\n",
      "step 4680, training accuracy 0.872038\n",
      "step 4680, cost 21.032\n",
      "step 4680, change in cost 0.00037384\n",
      "step 4690, training accuracy 0.872038\n",
      "step 4690, cost 21.0316\n",
      "step 4690, change in cost 0.00037384\n",
      "step 4700, training accuracy 0.872038\n",
      "step 4700, cost 21.0312\n",
      "step 4700, change in cost 0.000366211\n",
      "step 4710, training accuracy 0.872038\n",
      "step 4710, cost 21.0309\n",
      "step 4710, change in cost 0.000370026\n",
      "step 4720, training accuracy 0.872038\n",
      "step 4720, cost 21.0305\n",
      "step 4720, change in cost 0.000366211\n",
      "step 4730, training accuracy 0.872038\n",
      "step 4730, cost 21.0301\n",
      "step 4730, change in cost 0.000366211\n",
      "step 4740, training accuracy 0.872038\n",
      "step 4740, cost 21.0298\n",
      "step 4740, change in cost 0.000356674\n",
      "step 4750, training accuracy 0.872038\n",
      "step 4750, cost 21.0294\n",
      "step 4750, change in cost 0.000360489\n",
      "step 4760, training accuracy 0.872038\n",
      "step 4760, cost 21.0291\n",
      "step 4760, change in cost 0.000356674\n",
      "step 4770, training accuracy 0.872038\n",
      "step 4770, cost 21.0287\n",
      "step 4770, change in cost 0.000352859\n",
      "step 4780, training accuracy 0.872038\n",
      "step 4780, cost 21.0284\n",
      "step 4780, change in cost 0.000358582\n",
      "step 4790, training accuracy 0.872038\n",
      "step 4790, cost 21.028\n",
      "step 4790, change in cost 0.000347137\n",
      "step 4800, training accuracy 0.872038\n",
      "step 4800, cost 21.0277\n",
      "step 4800, change in cost 0.000347137\n",
      "step 4810, training accuracy 0.872038\n",
      "step 4810, cost 21.0273\n",
      "step 4810, change in cost 0.000349045\n",
      "step 4820, training accuracy 0.872038\n",
      "step 4820, cost 21.027\n",
      "step 4820, change in cost 0.000341415\n",
      "step 4830, training accuracy 0.872038\n",
      "step 4830, cost 21.0266\n",
      "step 4830, change in cost 0.000347137\n",
      "step 4840, training accuracy 0.872038\n",
      "step 4840, cost 21.0263\n",
      "step 4840, change in cost 0.000335693\n",
      "step 4850, training accuracy 0.872038\n",
      "step 4850, cost 21.026\n",
      "step 4850, change in cost 0.000339508\n",
      "step 4860, training accuracy 0.872038\n",
      "step 4860, cost 21.0256\n",
      "step 4860, change in cost 0.000335693\n",
      "step 4870, training accuracy 0.872038\n",
      "step 4870, cost 21.0253\n",
      "step 4870, change in cost 0.000329971\n",
      "step 4880, training accuracy 0.872038\n",
      "step 4880, cost 21.025\n",
      "step 4880, change in cost 0.000337601\n",
      "step 4890, training accuracy 0.872038\n",
      "step 4890, cost 21.0246\n",
      "step 4890, change in cost 0.000328064\n",
      "step 4900, training accuracy 0.872038\n",
      "step 4900, cost 21.0243\n",
      "step 4900, change in cost 0.000326157\n",
      "step 4910, training accuracy 0.872038\n",
      "step 4910, cost 21.024\n",
      "step 4910, change in cost 0.000329971\n",
      "step 4920, training accuracy 0.872038\n",
      "step 4920, cost 21.0236\n",
      "step 4920, change in cost 0.000320435\n",
      "step 4930, training accuracy 0.872038\n",
      "step 4930, cost 21.0233\n",
      "step 4930, change in cost 0.000324249\n",
      "step 4940, training accuracy 0.872038\n",
      "step 4940, cost 21.023\n",
      "step 4940, change in cost 0.000320435\n",
      "step 4950, training accuracy 0.872038\n",
      "step 4950, cost 21.0227\n",
      "step 4950, change in cost 0.000320435\n",
      "step 4960, training accuracy 0.872038\n",
      "step 4960, cost 21.0224\n",
      "step 4960, change in cost 0.000320435\n",
      "step 4970, training accuracy 0.872038\n",
      "step 4970, cost 21.0221\n",
      "step 4970, change in cost 0.000310898\n",
      "step 4980, training accuracy 0.872038\n",
      "step 4980, cost 21.0217\n",
      "step 4980, change in cost 0.000310898\n",
      "step 4990, training accuracy 0.872038\n",
      "step 4990, cost 21.0214\n",
      "step 4990, change in cost 0.000314713\n",
      "step 5000, training accuracy 0.872038\n",
      "step 5000, cost 21.0211\n",
      "step 5000, change in cost 0.000307083\n",
      "step 5010, training accuracy 0.872038\n",
      "step 5010, cost 21.0208\n",
      "step 5010, change in cost 0.00030899\n",
      "step 5020, training accuracy 0.872038\n",
      "step 5020, cost 21.0205\n",
      "step 5020, change in cost 0.000303268\n",
      "step 5030, training accuracy 0.872038\n",
      "step 5030, cost 21.0202\n",
      "step 5030, change in cost 0.000310898\n",
      "step 5040, training accuracy 0.872038\n",
      "step 5040, cost 21.0199\n",
      "step 5040, change in cost 0.000297546\n",
      "step 5050, training accuracy 0.872038\n",
      "step 5050, cost 21.0196\n",
      "step 5050, change in cost 0.000301361\n",
      "step 5060, training accuracy 0.872038\n",
      "step 5060, cost 21.0193\n",
      "step 5060, change in cost 0.000299454\n",
      "step 5070, training accuracy 0.872038\n",
      "step 5070, cost 21.019\n",
      "step 5070, change in cost 0.000299454\n",
      "step 5080, training accuracy 0.872038\n",
      "step 5080, cost 21.0187\n",
      "step 5080, change in cost 0.000295639\n",
      "step 5090, training accuracy 0.872038\n",
      "step 5090, cost 21.0184\n",
      "step 5090, change in cost 0.000291824\n",
      "step 5100, training accuracy 0.872038\n",
      "step 5100, cost 21.0181\n",
      "step 5100, change in cost 0.000293732\n",
      "step 5110, training accuracy 0.872038\n",
      "step 5110, cost 21.0178\n",
      "step 5110, change in cost 0.000289917\n",
      "step 5120, training accuracy 0.872038\n",
      "step 5120, cost 21.0175\n",
      "step 5120, change in cost 0.000293732\n",
      "step 5130, training accuracy 0.872038\n",
      "step 5130, cost 21.0172\n",
      "step 5130, change in cost 0.000284195\n",
      "step 5140, training accuracy 0.872038\n",
      "step 5140, cost 21.017\n",
      "step 5140, change in cost 0.000282288\n",
      "step 5150, training accuracy 0.872038\n",
      "step 5150, cost 21.0167\n",
      "step 5150, change in cost 0.000284195\n",
      "step 5160, training accuracy 0.876777\n",
      "step 5160, cost 21.0164\n",
      "step 5160, change in cost 0.00028801\n",
      "step 5170, training accuracy 0.876777\n",
      "step 5170, cost 21.0161\n",
      "step 5170, change in cost 0.00028038\n",
      "step 5180, training accuracy 0.876777\n",
      "step 5180, cost 21.0158\n",
      "step 5180, change in cost 0.000274658\n",
      "step 5190, training accuracy 0.876777\n",
      "step 5190, cost 21.0156\n",
      "step 5190, change in cost 0.000278473\n",
      "step 5200, training accuracy 0.876777\n",
      "step 5200, cost 21.0153\n",
      "step 5200, change in cost 0.000278473\n",
      "step 5210, training accuracy 0.876777\n",
      "step 5210, cost 21.015\n",
      "step 5210, change in cost 0.000270844\n",
      "step 5220, training accuracy 0.876777\n",
      "step 5220, cost 21.0147\n",
      "step 5220, change in cost 0.000274658\n",
      "step 5230, training accuracy 0.876777\n",
      "step 5230, cost 21.0145\n",
      "step 5230, change in cost 0.000274658\n",
      "step 5240, training accuracy 0.876777\n",
      "step 5240, cost 21.0142\n",
      "step 5240, change in cost 0.000268936\n",
      "step 5250, training accuracy 0.876777\n",
      "step 5250, cost 21.0139\n",
      "step 5250, change in cost 0.000268936\n",
      "step 5260, training accuracy 0.876777\n",
      "step 5260, cost 21.0137\n",
      "step 5260, change in cost 0.000268936\n",
      "step 5270, training accuracy 0.876777\n",
      "step 5270, cost 21.0134\n",
      "step 5270, change in cost 0.000261307\n",
      "step 5280, training accuracy 0.876777\n",
      "step 5280, cost 21.0131\n",
      "step 5280, change in cost 0.000267029\n",
      "step 5290, training accuracy 0.876777\n",
      "step 5290, cost 21.0129\n",
      "step 5290, change in cost 0.000261307\n",
      "step 5300, training accuracy 0.876777\n",
      "step 5300, cost 21.0126\n",
      "step 5300, change in cost 0.000263214\n",
      "step 5310, training accuracy 0.876777\n",
      "step 5310, cost 21.0123\n",
      "step 5310, change in cost 0.000261307\n",
      "step 5320, training accuracy 0.876777\n",
      "step 5320, cost 21.0121\n",
      "step 5320, change in cost 0.000255585\n",
      "step 5330, training accuracy 0.876777\n",
      "step 5330, cost 21.0118\n",
      "step 5330, change in cost 0.000257492\n",
      "step 5340, training accuracy 0.876777\n",
      "step 5340, cost 21.0116\n",
      "step 5340, change in cost 0.000255585\n",
      "step 5350, training accuracy 0.876777\n",
      "step 5350, cost 21.0113\n",
      "step 5350, change in cost 0.000255585\n",
      "step 5360, training accuracy 0.876777\n",
      "step 5360, cost 21.0111\n",
      "step 5360, change in cost 0.000253677\n",
      "step 5370, training accuracy 0.876777\n",
      "step 5370, cost 21.0108\n",
      "step 5370, change in cost 0.00025177\n",
      "step 5380, training accuracy 0.876777\n",
      "step 5380, cost 21.0106\n",
      "step 5380, change in cost 0.000247955\n",
      "step 5390, training accuracy 0.876777\n",
      "step 5390, cost 21.0103\n",
      "step 5390, change in cost 0.000249863\n",
      "step 5400, training accuracy 0.876777\n",
      "step 5400, cost 21.0101\n",
      "step 5400, change in cost 0.000246048\n",
      "step 5410, training accuracy 0.876777\n",
      "step 5410, cost 21.0098\n",
      "step 5410, change in cost 0.000247955\n",
      "step 5420, training accuracy 0.876777\n",
      "step 5420, cost 21.0096\n",
      "step 5420, change in cost 0.000244141\n",
      "step 5430, training accuracy 0.876777\n",
      "step 5430, cost 21.0093\n",
      "step 5430, change in cost 0.000242233\n",
      "step 5440, training accuracy 0.876777\n",
      "step 5440, cost 21.0091\n",
      "step 5440, change in cost 0.000242233\n",
      "step 5450, training accuracy 0.876777\n",
      "step 5450, cost 21.0088\n",
      "step 5450, change in cost 0.000244141\n",
      "step 5460, training accuracy 0.876777\n",
      "step 5460, cost 21.0086\n",
      "step 5460, change in cost 0.000238419\n",
      "step 5470, training accuracy 0.876777\n",
      "step 5470, cost 21.0084\n",
      "step 5470, change in cost 0.000236511\n",
      "step 5480, training accuracy 0.876777\n",
      "step 5480, cost 21.0081\n",
      "step 5480, change in cost 0.000238419\n",
      "step 5490, training accuracy 0.876777\n",
      "step 5490, cost 21.0079\n",
      "step 5490, change in cost 0.000234604\n",
      "step 5500, training accuracy 0.876777\n",
      "step 5500, cost 21.0077\n",
      "step 5500, change in cost 0.000238419\n",
      "step 5510, training accuracy 0.876777\n",
      "step 5510, cost 21.0074\n",
      "step 5510, change in cost 0.000228882\n",
      "step 5520, training accuracy 0.876777\n",
      "step 5520, cost 21.0072\n",
      "step 5520, change in cost 0.000234604\n",
      "step 5530, training accuracy 0.876777\n",
      "step 5530, cost 21.007\n",
      "step 5530, change in cost 0.000232697\n",
      "step 5540, training accuracy 0.876777\n",
      "step 5540, cost 21.0067\n",
      "step 5540, change in cost 0.000228882\n",
      "step 5550, training accuracy 0.876777\n",
      "step 5550, cost 21.0065\n",
      "step 5550, change in cost 0.000226974\n",
      "step 5560, training accuracy 0.876777\n",
      "step 5560, cost 21.0063\n",
      "step 5560, change in cost 0.000228882\n",
      "step 5570, training accuracy 0.876777\n",
      "step 5570, cost 21.0061\n",
      "step 5570, change in cost 0.000225067\n",
      "step 5580, training accuracy 0.876777\n",
      "step 5580, cost 21.0058\n",
      "step 5580, change in cost 0.000225067\n",
      "step 5590, training accuracy 0.876777\n",
      "step 5590, cost 21.0056\n",
      "step 5590, change in cost 0.000221252\n",
      "step 5600, training accuracy 0.876777\n",
      "step 5600, cost 21.0054\n",
      "step 5600, change in cost 0.000225067\n",
      "step 5610, training accuracy 0.876777\n",
      "step 5610, cost 21.0052\n",
      "step 5610, change in cost 0.000221252\n",
      "step 5620, training accuracy 0.876777\n",
      "step 5620, cost 21.0049\n",
      "step 5620, change in cost 0.000221252\n",
      "step 5630, training accuracy 0.876777\n",
      "step 5630, cost 21.0047\n",
      "step 5630, change in cost 0.000217438\n",
      "step 5640, training accuracy 0.876777\n",
      "step 5640, cost 21.0045\n",
      "step 5640, change in cost 0.000217438\n",
      "step 5650, training accuracy 0.876777\n",
      "step 5650, cost 21.0043\n",
      "step 5650, change in cost 0.000217438\n",
      "step 5660, training accuracy 0.876777\n",
      "step 5660, cost 21.0041\n",
      "step 5660, change in cost 0.000213623\n",
      "step 5670, training accuracy 0.876777\n",
      "step 5670, cost 21.0039\n",
      "step 5670, change in cost 0.000217438\n",
      "step 5680, training accuracy 0.876777\n",
      "step 5680, cost 21.0036\n",
      "step 5680, change in cost 0.000213623\n",
      "step 5690, training accuracy 0.876777\n",
      "step 5690, cost 21.0034\n",
      "step 5690, change in cost 0.00021553\n",
      "step 5700, training accuracy 0.876777\n",
      "step 5700, cost 21.0032\n",
      "step 5700, change in cost 0.000205994\n",
      "step 5710, training accuracy 0.876777\n",
      "step 5710, cost 21.003\n",
      "step 5710, change in cost 0.000211716\n",
      "step 5720, training accuracy 0.876777\n",
      "step 5720, cost 21.0028\n",
      "step 5720, change in cost 0.000207901\n",
      "step 5730, training accuracy 0.876777\n",
      "step 5730, cost 21.0026\n",
      "step 5730, change in cost 0.000211716\n",
      "step 5740, training accuracy 0.876777\n",
      "step 5740, cost 21.0024\n",
      "step 5740, change in cost 0.000204086\n",
      "step 5750, training accuracy 0.876777\n",
      "step 5750, cost 21.0022\n",
      "step 5750, change in cost 0.000207901\n",
      "step 5760, training accuracy 0.876777\n",
      "step 5760, cost 21.002\n",
      "step 5760, change in cost 0.000204086\n",
      "step 5770, training accuracy 0.876777\n",
      "step 5770, cost 21.0018\n",
      "step 5770, change in cost 0.000204086\n",
      "step 5780, training accuracy 0.876777\n",
      "step 5780, cost 21.0016\n",
      "step 5780, change in cost 0.000202179\n",
      "step 5790, training accuracy 0.876777\n",
      "step 5790, cost 21.0014\n",
      "step 5790, change in cost 0.000204086\n",
      "step 5800, training accuracy 0.876777\n",
      "step 5800, cost 21.0012\n",
      "step 5800, change in cost 0.000200272\n",
      "step 5810, training accuracy 0.876777\n",
      "step 5810, cost 21.001\n",
      "step 5810, change in cost 0.000202179\n",
      "step 5820, training accuracy 0.876777\n",
      "step 5820, cost 21.0008\n",
      "step 5820, change in cost 0.000196457\n",
      "step 5830, training accuracy 0.876777\n",
      "step 5830, cost 21.0006\n",
      "step 5830, change in cost 0.000196457\n",
      "step 5840, training accuracy 0.876777\n",
      "step 5840, cost 21.0004\n",
      "step 5840, change in cost 0.000196457\n",
      "step 5850, training accuracy 0.876777\n",
      "step 5850, cost 21.0002\n",
      "step 5850, change in cost 0.000196457\n",
      "step 5860, training accuracy 0.876777\n",
      "step 5860, cost 21\n",
      "step 5860, change in cost 0.00019455\n",
      "step 5870, training accuracy 0.876777\n",
      "step 5870, cost 20.9998\n",
      "step 5870, change in cost 0.000190735\n",
      "step 5880, training accuracy 0.876777\n",
      "step 5880, cost 20.9996\n",
      "step 5880, change in cost 0.000196457\n",
      "step 5890, training accuracy 0.876777\n",
      "step 5890, cost 20.9994\n",
      "step 5890, change in cost 0.000190735\n",
      "step 5900, training accuracy 0.876777\n",
      "step 5900, cost 20.9992\n",
      "step 5900, change in cost 0.000192642\n",
      "step 5910, training accuracy 0.876777\n",
      "step 5910, cost 20.999\n",
      "step 5910, change in cost 0.00018692\n",
      "step 5920, training accuracy 0.876777\n",
      "step 5920, cost 20.9988\n",
      "step 5920, change in cost 0.00018692\n",
      "step 5930, training accuracy 0.876777\n",
      "step 5930, cost 20.9986\n",
      "step 5930, change in cost 0.000188828\n",
      "step 5940, training accuracy 0.876777\n",
      "step 5940, cost 20.9985\n",
      "step 5940, change in cost 0.000192642\n",
      "step 5950, training accuracy 0.876777\n",
      "step 5950, cost 20.9983\n",
      "step 5950, change in cost 0.000183105\n",
      "step 5960, training accuracy 0.876777\n",
      "step 5960, cost 20.9981\n",
      "step 5960, change in cost 0.000183105\n",
      "step 5970, training accuracy 0.876777\n",
      "step 5970, cost 20.9979\n",
      "step 5970, change in cost 0.000183105\n",
      "step 5980, training accuracy 0.876777\n",
      "step 5980, cost 20.9977\n",
      "step 5980, change in cost 0.000185013\n",
      "step 5990, training accuracy 0.876777\n",
      "step 5990, cost 20.9975\n",
      "step 5990, change in cost 0.000181198\n",
      "step 6000, training accuracy 0.876777\n",
      "step 6000, cost 20.9974\n",
      "step 6000, change in cost 0.000185013\n",
      "step 6010, training accuracy 0.876777\n",
      "step 6010, cost 20.9972\n",
      "step 6010, change in cost 0.000181198\n",
      "step 6020, training accuracy 0.876777\n",
      "step 6020, cost 20.997\n",
      "step 6020, change in cost 0.000177383\n",
      "step 6030, training accuracy 0.876777\n",
      "step 6030, cost 20.9968\n",
      "step 6030, change in cost 0.000181198\n",
      "step 6040, training accuracy 0.876777\n",
      "step 6040, cost 20.9966\n",
      "step 6040, change in cost 0.000175476\n",
      "step 6050, training accuracy 0.876777\n",
      "step 6050, cost 20.9965\n",
      "step 6050, change in cost 0.000177383\n",
      "step 6060, training accuracy 0.876777\n",
      "step 6060, cost 20.9963\n",
      "step 6060, change in cost 0.000179291\n",
      "step 6070, training accuracy 0.876777\n",
      "step 6070, cost 20.9961\n",
      "step 6070, change in cost 0.000171661\n",
      "step 6080, training accuracy 0.876777\n",
      "step 6080, cost 20.9959\n",
      "step 6080, change in cost 0.000173569\n",
      "step 6090, training accuracy 0.876777\n",
      "step 6090, cost 20.9958\n",
      "step 6090, change in cost 0.000173569\n",
      "step 6100, training accuracy 0.876777\n",
      "step 6100, cost 20.9956\n",
      "step 6100, change in cost 0.000173569\n",
      "step 6110, training accuracy 0.876777\n",
      "step 6110, cost 20.9954\n",
      "step 6110, change in cost 0.000171661\n",
      "step 6120, training accuracy 0.876777\n",
      "step 6120, cost 20.9952\n",
      "step 6120, change in cost 0.000171661\n",
      "step 6130, training accuracy 0.876777\n",
      "step 6130, cost 20.9951\n",
      "step 6130, change in cost 0.000171661\n",
      "step 6140, training accuracy 0.876777\n",
      "step 6140, cost 20.9949\n",
      "step 6140, change in cost 0.000167847\n",
      "step 6150, training accuracy 0.876777\n",
      "step 6150, cost 20.9947\n",
      "step 6150, change in cost 0.000171661\n",
      "step 6160, training accuracy 0.876777\n",
      "step 6160, cost 20.9946\n",
      "step 6160, change in cost 0.000167847\n",
      "step 6170, training accuracy 0.876777\n",
      "step 6170, cost 20.9944\n",
      "step 6170, change in cost 0.000165939\n",
      "step 6180, training accuracy 0.876777\n",
      "step 6180, cost 20.9942\n",
      "step 6180, change in cost 0.000165939\n",
      "step 6190, training accuracy 0.876777\n",
      "step 6190, cost 20.9941\n",
      "step 6190, change in cost 0.000165939\n",
      "step 6200, training accuracy 0.876777\n",
      "step 6200, cost 20.9939\n",
      "step 6200, change in cost 0.000167847\n",
      "step 6210, training accuracy 0.876777\n",
      "step 6210, cost 20.9937\n",
      "step 6210, change in cost 0.000162125\n",
      "step 6220, training accuracy 0.876777\n",
      "step 6220, cost 20.9936\n",
      "step 6220, change in cost 0.000162125\n",
      "step 6230, training accuracy 0.876777\n",
      "step 6230, cost 20.9934\n",
      "step 6230, change in cost 0.000162125\n",
      "step 6240, training accuracy 0.876777\n",
      "step 6240, cost 20.9933\n",
      "step 6240, change in cost 0.000164032\n",
      "step 6250, training accuracy 0.876777\n",
      "step 6250, cost 20.9931\n",
      "step 6250, change in cost 0.00015831\n",
      "step 6260, training accuracy 0.876777\n",
      "step 6260, cost 20.9929\n",
      "step 6260, change in cost 0.000160217\n",
      "step 6270, training accuracy 0.876777\n",
      "step 6270, cost 20.9928\n",
      "step 6270, change in cost 0.00015831\n",
      "step 6280, training accuracy 0.876777\n",
      "step 6280, cost 20.9926\n",
      "step 6280, change in cost 0.000162125\n",
      "step 6290, training accuracy 0.876777\n",
      "step 6290, cost 20.9925\n",
      "step 6290, change in cost 0.000154495\n",
      "step 6300, training accuracy 0.876777\n",
      "step 6300, cost 20.9923\n",
      "step 6300, change in cost 0.00015831\n",
      "step 6310, training accuracy 0.876777\n",
      "step 6310, cost 20.9921\n",
      "step 6310, change in cost 0.000160217\n",
      "step 6320, training accuracy 0.876777\n",
      "step 6320, cost 20.992\n",
      "step 6320, change in cost 0.000154495\n",
      "step 6330, training accuracy 0.876777\n",
      "step 6330, cost 20.9918\n",
      "step 6330, change in cost 0.000152588\n",
      "step 6340, training accuracy 0.876777\n",
      "step 6340, cost 20.9917\n",
      "step 6340, change in cost 0.000152588\n",
      "step 6350, training accuracy 0.876777\n",
      "step 6350, cost 20.9915\n",
      "step 6350, change in cost 0.000156403\n",
      "step 6360, training accuracy 0.876777\n",
      "step 6360, cost 20.9914\n",
      "step 6360, change in cost 0.000146866\n",
      "step 6370, training accuracy 0.876777\n",
      "step 6370, cost 20.9912\n",
      "step 6370, change in cost 0.00015831\n",
      "step 6380, training accuracy 0.876777\n",
      "step 6380, cost 20.9911\n",
      "step 6380, change in cost 0.000146866\n",
      "step 6390, training accuracy 0.876777\n",
      "step 6390, cost 20.9909\n",
      "step 6390, change in cost 0.000156403\n",
      "step 6400, training accuracy 0.876777\n",
      "step 6400, cost 20.9908\n",
      "step 6400, change in cost 0.000144958\n",
      "step 6410, training accuracy 0.876777\n",
      "step 6410, cost 20.9906\n",
      "step 6410, change in cost 0.000150681\n",
      "step 6420, training accuracy 0.876777\n",
      "step 6420, cost 20.9905\n",
      "step 6420, change in cost 0.000146866\n",
      "step 6430, training accuracy 0.876777\n",
      "step 6430, cost 20.9903\n",
      "step 6430, change in cost 0.000150681\n",
      "step 6440, training accuracy 0.876777\n",
      "step 6440, cost 20.9902\n",
      "step 6440, change in cost 0.000144958\n",
      "step 6450, training accuracy 0.876777\n",
      "step 6450, cost 20.99\n",
      "step 6450, change in cost 0.000152588\n",
      "step 6460, training accuracy 0.876777\n",
      "step 6460, cost 20.9899\n",
      "step 6460, change in cost 0.000141144\n",
      "step 6470, training accuracy 0.876777\n",
      "step 6470, cost 20.9897\n",
      "step 6470, change in cost 0.000144958\n",
      "step 6480, training accuracy 0.876777\n",
      "step 6480, cost 20.9896\n",
      "step 6480, change in cost 0.000143051\n",
      "step 6490, training accuracy 0.876777\n",
      "step 6490, cost 20.9895\n",
      "step 6490, change in cost 0.000143051\n",
      "step 6500, training accuracy 0.876777\n",
      "step 6500, cost 20.9893\n",
      "step 6500, change in cost 0.000143051\n",
      "step 6510, training accuracy 0.876777\n",
      "step 6510, cost 20.9892\n",
      "step 6510, change in cost 0.000144958\n",
      "step 6520, training accuracy 0.876777\n",
      "step 6520, cost 20.989\n",
      "step 6520, change in cost 0.000144958\n",
      "step 6530, training accuracy 0.876777\n",
      "step 6530, cost 20.9889\n",
      "step 6530, change in cost 0.000135422\n",
      "step 6540, training accuracy 0.876777\n",
      "step 6540, cost 20.9887\n",
      "step 6540, change in cost 0.000141144\n",
      "step 6550, training accuracy 0.876777\n",
      "step 6550, cost 20.9886\n",
      "step 6550, change in cost 0.000141144\n",
      "step 6560, training accuracy 0.876777\n",
      "step 6560, cost 20.9885\n",
      "step 6560, change in cost 0.000139236\n",
      "step 6570, training accuracy 0.876777\n",
      "step 6570, cost 20.9883\n",
      "step 6570, change in cost 0.000139236\n",
      "step 6580, training accuracy 0.876777\n",
      "step 6580, cost 20.9882\n",
      "step 6580, change in cost 0.000135422\n",
      "step 6590, training accuracy 0.876777\n",
      "step 6590, cost 20.988\n",
      "step 6590, change in cost 0.000139236\n",
      "step 6600, training accuracy 0.876777\n",
      "step 6600, cost 20.9879\n",
      "step 6600, change in cost 0.000137329\n",
      "step 6610, training accuracy 0.876777\n",
      "step 6610, cost 20.9878\n",
      "step 6610, change in cost 0.000135422\n",
      "step 6620, training accuracy 0.876777\n",
      "step 6620, cost 20.9876\n",
      "step 6620, change in cost 0.000135422\n",
      "step 6630, training accuracy 0.876777\n",
      "step 6630, cost 20.9875\n",
      "step 6630, change in cost 0.000133514\n",
      "step 6640, training accuracy 0.876777\n",
      "step 6640, cost 20.9874\n",
      "step 6640, change in cost 0.000133514\n",
      "step 6650, training accuracy 0.876777\n",
      "step 6650, cost 20.9872\n",
      "step 6650, change in cost 0.000135422\n",
      "step 6660, training accuracy 0.876777\n",
      "step 6660, cost 20.9871\n",
      "step 6660, change in cost 0.000131607\n",
      "step 6670, training accuracy 0.876777\n",
      "step 6670, cost 20.987\n",
      "step 6670, change in cost 0.000137329\n",
      "step 6680, training accuracy 0.876777\n",
      "step 6680, cost 20.9868\n",
      "step 6680, change in cost 0.0001297\n",
      "step 6690, training accuracy 0.876777\n",
      "step 6690, cost 20.9867\n",
      "step 6690, change in cost 0.000133514\n",
      "step 6700, training accuracy 0.876777\n",
      "step 6700, cost 20.9866\n",
      "step 6700, change in cost 0.000125885\n",
      "step 6710, training accuracy 0.876777\n",
      "step 6710, cost 20.9864\n",
      "step 6710, change in cost 0.000131607\n",
      "step 6720, training accuracy 0.876777\n",
      "step 6720, cost 20.9863\n",
      "step 6720, change in cost 0.000131607\n",
      "step 6730, training accuracy 0.876777\n",
      "step 6730, cost 20.9862\n",
      "step 6730, change in cost 0.000127792\n",
      "step 6740, training accuracy 0.876777\n",
      "step 6740, cost 20.9861\n",
      "step 6740, change in cost 0.000127792\n",
      "step 6750, training accuracy 0.876777\n",
      "step 6750, cost 20.9859\n",
      "step 6750, change in cost 0.000127792\n",
      "step 6760, training accuracy 0.876777\n",
      "step 6760, cost 20.9858\n",
      "step 6760, change in cost 0.000127792\n",
      "step 6770, training accuracy 0.876777\n",
      "step 6770, cost 20.9857\n",
      "step 6770, change in cost 0.00012207\n",
      "step 6780, training accuracy 0.876777\n",
      "step 6780, cost 20.9856\n",
      "step 6780, change in cost 0.0001297\n",
      "step 6790, training accuracy 0.876777\n",
      "step 6790, cost 20.9854\n",
      "step 6790, change in cost 0.000123978\n",
      "step 6800, training accuracy 0.876777\n",
      "step 6800, cost 20.9853\n",
      "step 6800, change in cost 0.000123978\n",
      "step 6810, training accuracy 0.876777\n",
      "step 6810, cost 20.9852\n",
      "step 6810, change in cost 0.000125885\n",
      "step 6820, training accuracy 0.876777\n",
      "step 6820, cost 20.9851\n",
      "step 6820, change in cost 0.000123978\n",
      "step 6830, training accuracy 0.876777\n",
      "step 6830, cost 20.9849\n",
      "step 6830, change in cost 0.00012207\n",
      "step 6840, training accuracy 0.876777\n",
      "step 6840, cost 20.9848\n",
      "step 6840, change in cost 0.00012207\n",
      "step 6850, training accuracy 0.876777\n",
      "step 6850, cost 20.9847\n",
      "step 6850, change in cost 0.000127792\n",
      "step 6860, training accuracy 0.876777\n",
      "step 6860, cost 20.9846\n",
      "step 6860, change in cost 0.000118256\n",
      "step 6870, training accuracy 0.876777\n",
      "step 6870, cost 20.9844\n",
      "step 6870, change in cost 0.00012207\n",
      "step 6880, training accuracy 0.876777\n",
      "step 6880, cost 20.9843\n",
      "step 6880, change in cost 0.000118256\n",
      "step 6890, training accuracy 0.876777\n",
      "step 6890, cost 20.9842\n",
      "step 6890, change in cost 0.00012207\n",
      "step 6900, training accuracy 0.876777\n",
      "step 6900, cost 20.9841\n",
      "step 6900, change in cost 0.000116348\n",
      "step 6910, training accuracy 0.876777\n",
      "step 6910, cost 20.984\n",
      "step 6910, change in cost 0.000120163\n",
      "step 6920, training accuracy 0.876777\n",
      "step 6920, cost 20.9838\n",
      "step 6920, change in cost 0.00012207\n",
      "step 6930, training accuracy 0.876777\n",
      "step 6930, cost 20.9837\n",
      "step 6930, change in cost 0.000118256\n",
      "step 6940, training accuracy 0.876777\n",
      "step 6940, cost 20.9836\n",
      "step 6940, change in cost 0.000116348\n",
      "step 6950, training accuracy 0.876777\n",
      "step 6950, cost 20.9835\n",
      "step 6950, change in cost 0.000116348\n",
      "step 6960, training accuracy 0.876777\n",
      "step 6960, cost 20.9834\n",
      "step 6960, change in cost 0.000114441\n",
      "step 6970, training accuracy 0.876777\n",
      "step 6970, cost 20.9833\n",
      "step 6970, change in cost 0.000118256\n",
      "step 6980, training accuracy 0.876777\n",
      "step 6980, cost 20.9831\n",
      "step 6980, change in cost 0.000114441\n",
      "step 6990, training accuracy 0.876777\n",
      "step 6990, cost 20.983\n",
      "step 6990, change in cost 0.000116348\n",
      "step 7000, training accuracy 0.876777\n",
      "step 7000, cost 20.9829\n",
      "step 7000, change in cost 0.000112534\n",
      "step 7010, training accuracy 0.876777\n",
      "step 7010, cost 20.9828\n",
      "step 7010, change in cost 0.000114441\n",
      "step 7020, training accuracy 0.876777\n",
      "step 7020, cost 20.9827\n",
      "step 7020, change in cost 0.000112534\n",
      "step 7030, training accuracy 0.876777\n",
      "step 7030, cost 20.9826\n",
      "step 7030, change in cost 0.000112534\n",
      "step 7040, training accuracy 0.876777\n",
      "step 7040, cost 20.9825\n",
      "step 7040, change in cost 0.000112534\n",
      "step 7050, training accuracy 0.876777\n",
      "step 7050, cost 20.9824\n",
      "step 7050, change in cost 0.000112534\n",
      "step 7060, training accuracy 0.876777\n",
      "step 7060, cost 20.9822\n",
      "step 7060, change in cost 0.000114441\n",
      "step 7070, training accuracy 0.876777\n",
      "step 7070, cost 20.9821\n",
      "step 7070, change in cost 0.000108719\n",
      "step 7080, training accuracy 0.876777\n",
      "step 7080, cost 20.982\n",
      "step 7080, change in cost 0.000112534\n",
      "step 7090, training accuracy 0.876777\n",
      "step 7090, cost 20.9819\n",
      "step 7090, change in cost 0.000108719\n",
      "step 7100, training accuracy 0.876777\n",
      "step 7100, cost 20.9818\n",
      "step 7100, change in cost 0.000108719\n",
      "step 7110, training accuracy 0.876777\n",
      "step 7110, cost 20.9817\n",
      "step 7110, change in cost 0.000110626\n",
      "step 7120, training accuracy 0.876777\n",
      "step 7120, cost 20.9816\n",
      "step 7120, change in cost 0.000106812\n",
      "step 7130, training accuracy 0.876777\n",
      "step 7130, cost 20.9815\n",
      "step 7130, change in cost 0.000110626\n",
      "step 7140, training accuracy 0.876777\n",
      "step 7140, cost 20.9814\n",
      "step 7140, change in cost 0.000106812\n",
      "step 7150, training accuracy 0.876777\n",
      "step 7150, cost 20.9813\n",
      "step 7150, change in cost 0.000106812\n",
      "step 7160, training accuracy 0.876777\n",
      "step 7160, cost 20.9812\n",
      "step 7160, change in cost 0.000106812\n",
      "step 7170, training accuracy 0.876777\n",
      "step 7170, cost 20.981\n",
      "step 7170, change in cost 0.000106812\n",
      "step 7180, training accuracy 0.876777\n",
      "step 7180, cost 20.9809\n",
      "step 7180, change in cost 0.000108719\n",
      "step 7190, training accuracy 0.876777\n",
      "step 7190, cost 20.9808\n",
      "step 7190, change in cost 0.000101089\n",
      "step 7200, training accuracy 0.876777\n",
      "step 7200, cost 20.9807\n",
      "step 7200, change in cost 0.000110626\n",
      "step 7210, training accuracy 0.876777\n",
      "step 7210, cost 20.9806\n",
      "step 7210, change in cost 9.91821e-05\n",
      "change in cost 9.91821e-05; convergence.\n",
      "final accuracy on test set: 0.867925\n"
     ]
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAbAAAAEaCAYAAABwyQKiAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nDckNmJmZNSQ3YGZm1pD+P5aBXGtODfvLAAAAAElFTkSuQmCC\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7fbc6bf07890>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "###########################\n",
    "### GRAPH LIVE UPDATING ###\n",
    "###########################\n",
    "\n",
    "epoch_values=[]\n",
    "accuracy_values=[]\n",
    "cost_values=[]\n",
    "# Turn on interactive plotting\n",
    "plt.ion()\n",
    "# Create the main, super plot\n",
    "fig = plt.figure()\n",
    "# Create two subplots on their own axes and give titles\n",
    "ax1 = plt.subplot(\"211\")\n",
    "ax1.set_title(\"TRAINING ACCURACY\", fontsize=18)\n",
    "ax2 = plt.subplot(\"212\")\n",
    "ax2.set_title(\"TRAINING COST\", fontsize=18)\n",
    "plt.tight_layout()\n",
    "\n",
    "#####################\n",
    "### RUN THE GRAPH ###\n",
    "#####################\n",
    "\n",
    "# Create a tensorflow session\n",
    "sess = tf.Session()\n",
    "\n",
    "# Initialize all tensorflow variables\n",
    "sess.run(init_OP)\n",
    "\n",
    "## Ops for vizualization\n",
    "# argmax(activation_OP, 1) gives the label our model thought was most likely\n",
    "# argmax(yGold, 1) is the correct label\n",
    "correct_predictions_OP = tf.equal(tf.argmax(activation_OP,1),tf.argmax(yGold,1))\n",
    "# False is 0 and True is 1, what was our average?\n",
    "accuracy_OP = tf.reduce_mean(tf.cast(correct_predictions_OP, \"float\"))\n",
    "# Summary op for regression output\n",
    "activation_summary_OP = tf.histogram_summary(\"output\", activation_OP)\n",
    "# Summary op for accuracy\n",
    "accuracy_summary_OP = tf.scalar_summary(\"accuracy\", accuracy_OP)\n",
    "# Summary op for cost\n",
    "cost_summary_OP = tf.scalar_summary(\"cost\", cost_OP)\n",
    "# Summary ops to check how variables (W, b) are updating after each iteration\n",
    "weightSummary = tf.histogram_summary(\"weights\", weights.eval(session=sess))\n",
    "biasSummary = tf.histogram_summary(\"biases\", bias.eval(session=sess))\n",
    "# Merge all summaries\n",
    "all_summary_OPS = tf.merge_all_summaries()\n",
    "# Summary writer\n",
    "writer = tf.train.SummaryWriter(\"summary_logs\", sess.graph_def)\n",
    "\n",
    "\n",
    "\n",
    "# Initialize reporting variables\n",
    "cost = 0\n",
    "diff = 1\n",
    "\n",
    "# Training epochs\n",
    "for i in range(numEpochs):\n",
    "    if i > 1 and diff < .0001:\n",
    "        print(\"change in cost %g; convergence.\"%diff)\n",
    "        break\n",
    "    else:\n",
    "        # Run training step\n",
    "        step = sess.run(training_OP, feed_dict={X: trainX, yGold: trainY})\n",
    "        # Report occasional stats\n",
    "        if i % 10 == 0:\n",
    "            # Add epoch to epoch_values\n",
    "            epoch_values.append(i)\n",
    "            # Generate accuracy stats on test data\n",
    "            summary_results, train_accuracy, newCost = sess.run(\n",
    "                [all_summary_OPS, accuracy_OP,cost_OP], \n",
    "                feed_dict={X: trainX, yGold: trainY}\n",
    "            )\n",
    "            # Add accuracy to live graphing variable\n",
    "            accuracy_values.append(train_accuracy)\n",
    "            # Add cost to live graphing variable\n",
    "            cost_values.append(newCost)\n",
    "            # Write summary stats to writer\n",
    "            writer.add_summary(summary_results, i)\n",
    "            # Re-assign values for variables\n",
    "            diff = abs(newCost - cost)\n",
    "            cost = newCost\n",
    "\n",
    "            #generate print statements\n",
    "            print(\"step %d, training accuracy %g\"%(i, train_accuracy))\n",
    "            print(\"step %d, cost %g\"%(i, newCost))\n",
    "            print(\"step %d, change in cost %g\"%(i, diff))\n",
    "\n",
    "            # Plot progress to our two subplots\n",
    "            accuracyLine, = ax1.plot(epoch_values, accuracy_values)\n",
    "            costLine, = ax2.plot(epoch_values, cost_values)\n",
    "            fig.canvas.draw()\n",
    "            #time.sleep(1)\n",
    "\n",
    "\n",
    "# How well do we perform on held-out test data?\n",
    "print(\"final accuracy on test set: %s\" %str(sess.run(accuracy_OP, \n",
    "                                                     feed_dict={X: testX, \n",
    "                                                                yGold: testY})))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 86,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "final accuracy on test set: 0.867925\n"
     ]
    }
   ],
   "source": [
    "print(\"final accuracy on test set: %s\" %str(sess.run(accuracy_OP, \n",
    "                                                     feed_dict={X: testX, \n",
    "                                                                yGold: testY})))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<tensorflow.python.ops.variables.Variable at 0x7fbc6bfdb410>"
      ]
     },
     "execution_count": 12,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "weights"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {
    "collapsed": false,
    "scrolled": true
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([[-6.29393435,  6.39006042]], dtype=float32)"
      ]
     },
     "execution_count": 13,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "sess.run(bias)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 33,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([[ 0.00194525,  0.98826116]], dtype=float32)"
      ]
     },
     "execution_count": 33,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "#get probability value for a given input.\n",
    "x1=np.array([[0.0,0.0,0.2,0.1]])\n",
    "y1=np.array([[1,0]])\n",
    "sess.run(activation_OP,feed_dict={X:x1})"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(1, 4)"
      ]
     },
     "execution_count": 17,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "x1.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(53, 4)"
      ]
     },
     "execution_count": 18,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "testX.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(211, 4)"
      ]
     },
     "execution_count": 19,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "trainX.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(211, 2)"
      ]
     },
     "execution_count": 20,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "trainY.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([ 0.09746567], dtype=float32)"
      ]
     },
     "execution_count": 21,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "sess.run(tf.nn.softmax_cross_entropy_with_logits(add_bias_OP,yGold),feed_dict={X:x1,yGold:y1})"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0.99995458"
      ]
     },
     "execution_count": 22,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "sess.run(tf.nn.sigmoid(10.0))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0.060847588"
      ]
     },
     "execution_count": 23,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "sess.run(cost_OP,feed_dict={X:x1,yGold:y1})"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 40,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0.53249544"
      ]
     },
     "execution_count": 40,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "sess.run(cost_OP1,feed_dict={X:testX,yGold:testY})"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[0.49289098,\n",
       " 0.49289098,\n",
       " 0.49289098,\n",
       " 0.49289098,\n",
       " 0.49289098,\n",
       " 0.49289098,\n",
       " 0.49289098,\n",
       " 0.49289098,\n",
       " 0.49289098,\n",
       " 0.49289098,\n",
       " 0.49289098,\n",
       " 0.49289098,\n",
       " 0.5687204,\n",
       " 0.63033175,\n",
       " 0.72037917,\n",
       " 0.74407583,\n",
       " 0.79146922,\n",
       " 0.83412319,\n",
       " 0.84834123,\n",
       " 0.85308057,\n",
       " 0.85781991,\n",
       " 0.85781991,\n",
       " 0.85781991]"
      ]
     },
     "execution_count": 11,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "accuracy_values"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[5.4447966,\n",
       " 5.0179462,\n",
       " 4.7105241,\n",
       " 4.3523402,\n",
       " 4.004334,\n",
       " 3.7014742,\n",
       " 3.4462745,\n",
       " 3.2194164,\n",
       " 2.9931815,\n",
       " 2.7364919,\n",
       " 2.3938942,\n",
       " 1.7644708,\n",
       " 0.74838257,\n",
       " 0.67568296,\n",
       " 0.60561264,\n",
       " 0.54509306,\n",
       " 0.49430948,\n",
       " 0.45201296,\n",
       " 0.43114838,\n",
       " 0.42073059,\n",
       " 0.41593388,\n",
       " 0.41424325,\n",
       " 0.41416937]"
      ]
     },
     "execution_count": 12,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "cost_values"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[0,\n",
       " 10,\n",
       " 20,\n",
       " 30,\n",
       " 40,\n",
       " 50,\n",
       " 60,\n",
       " 70,\n",
       " 80,\n",
       " 90,\n",
       " 100,\n",
       " 110,\n",
       " 120,\n",
       " 130,\n",
       " 140,\n",
       " 150,\n",
       " 160,\n",
       " 170,\n",
       " 180,\n",
       " 190,\n",
       " 200,\n",
       " 210,\n",
       " 220]"
      ]
     },
     "execution_count": 13,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "epoch_values"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": [
    "%matplotlib inline\n",
    "plt.figure()\n",
    "plt.plot(epoch_values,cost_values)\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 78,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "from sklearn.metrics import roc_curve,auc,accuracy_score"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 46,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": [
    "score=sess.run(activation_OP,feed_dict={X:testX})\n",
    "score_pred = [b[1] for b in score]\n",
    "actual = [b[1] for b in testY]\n",
    "fpr,tpr,thresholds = roc_curve(actual,score_pred)\n",
    "roc_auc = auc(fpr,tpr)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 33,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[<matplotlib.lines.Line2D at 0x7fbc5e27aa50>]"
      ]
     },
     "execution_count": 33,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAXkAAAEACAYAAABWLgY0AAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\n/wNRVwUsHeBYwAAAAABJRU5ErkJggg==\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7fbc5e41f150>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.plot(fpr,tpr)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 36,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0.92307692307692313"
      ]
     },
     "execution_count": 36,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "roc_auc"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 40,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<matplotlib.legend.Legend at 0x7fbc5d3bc990>"
      ]
     },
     "execution_count": 40,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAYgAAAEZCAYAAACNebLAAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz\nAAALEgAACxIB0t1+/AAAIABJREFUeJzt3Xd4FFX3wPHvSeiQhKaUQAABRaoiIhY0KCgovlYUUH8i\nKgqoiAVR8QU7qIgFQUDUF3sXRRQViUgRURGQItVQDCUESIAQUs7vjxniJmySJWQzm+R8nmef7PQz\nk909M/feuSOqijHGGJNbmNcBGGOMCU2WIIwxxvhlCcIYY4xfliCMMcb4ZQnCGGOMX5YgjDHG+GUJ\nogQTkb4i8o3XcXhNRBqKSLKISDFus5GIZIlIqfgOicifInJuIZYrtZ9BETlPRDZ7HYeXxO6DKBoi\n8jdwPJAB7ANmAYNV9YCXcZVGIrIRuFlVf/AwhkbABqC8qmZ5FYcbSxbQTFU3BHk7jYCNQDmv97k4\niMh5wFuqGuN1LF4pFWc/IUKBS1Q1EjgFOBV40NuQCkdEwsvitr1SBPtc6LO8o9y2uNsqtis14y1L\nEEVLAFR1B84VxCnZE0QqiMhzIhIvIgkiMkFEKvpMv0xElojIXhFZKyIXuuMjReQ1EflHRDaLyOOH\ni1JE5EYR+cl9P0FEns0RjMjnInK3+76eiHwsIjtEZL2I3Okz30gR+UhE3hKRPcCNR+yYE8c0d/mN\nIvKwz7QbRWSeiLwsIntEZKWInJ9r2fz2YZ6IPC8iicBIETlBRGaLSKK7vbdFJNKdfxoQA3zpFivd\nl7u4R0TmiMhj7nqTReQbEanpE8//icjfIrJTREa4+5Mdb679riQiY935d4vIXJ//mwDXu//THSLy\nkM9yp4vIAneZre6xKeczPUtEBonIGmCNO+4FEdnkfgYWi8g5PvOHichDIrLO3afFItJARH5041jm\nju/lzt/T/Tztdo9DG591bRSRYSKyFNgnIuG+x8CNfbEbR4KIPOcu+qP7d4+7rTN8P4Pusq1E5FsR\n2eUuOzyP45r7+zDx8HF1Y/vZ5/85UESWi0gFd/hDd5ndIhInIi191vuGiLwiIjNFJEVEfhKROiIy\nTkSS3M9mu1zHYriIrHBjnnp4O35izvM7VGqpqr2K4IVz6X2++74BsAx43mf6OOBzIAqoCkwHnnSn\ndQT2+CxfDzjRff8ZMAGoBNQGfgZudafdCMx133cG4n22Vx04ANTB+QH5FXgYCAcaA+uAbu68I4E0\n4FJ3uKKf/ZvmxlIFaAT8BdzkE0c6cJe7/mvc/ake4D6kA4NwTlgqAk2BC4ByQC0gLtex3Ah08Rlu\nBGQCYe7wHGCtu56K7vBT7rSWQApwprv+Z919Pz+P/+srwA9AXfc4dgLKu9vMAiYBFYC2wEHgJHe5\n9u7/VXAS2grgLp/1ZuGcREQdPt5AX/f/FgYMBRKACu60+4GlOEVJAG2AGj7rauKz7lOB7UAHd/s3\nuMesvM/x+x2o77Nt38/vAuA6930VoGOu4yw+2/L9DFYD/gHudo9JVeD0PI5rft8Hcf/n/wWaAUlA\nW59l+7lxlQeeB5b4THsD2IFzclYBmI1TFHidu97HgR9yfZaWuceiOjAPeMyddh6wySemPL9DpfXl\neQCl5eV+0JLdVxbwHRDpM31fri/xmcAG9/2rwFg/6zwe50enos+43oc/4L5fTnf4b+Ac9/0twPfu\n+zOAv3Otezgw1X0/EojLZ9/CcH5ET/IZNyBXHFtyLbPI/VIGsg9/57Vtd57LgN9yHevzfYb9JYiH\nfKYPBGa67x8B3vGZVpk8EoT7o3AAaO1n2uFt1su1z9fksQ9DgE98hrOA8wrY7ySgjft+NdAzj/my\ngBN8hicAj+aaZzXQ2ef43ejn83s4QcS5n4laeexzmM843wTR2/f/VMC+5fl98NnWLmAlMCyf9VR3\n9z/CHX4DmOQz/Q5ghc9wayAp137f6jPcA1jrvvdNEPl+h0rrK/uS1xSJy1R1joh0Bt7FOVtOFpHj\ncM54fpN/G9qE8W9ZbkPgKz/ra4RzlpTgLifua1Me2/8A6INzFtQXeMsdHwNEi0iSOyzu9uf6LJtf\na43aOGfbvtuNB6J9hrfmWiYe56wskH3IsW0ROR54EeeqqBrOGVsSR2ebz/sD7npwY8renqqmisiu\nPNZRG+cKJL/K3+3+tiMizXHObjvgJKFywG+5lt3iOyAi9wH9ca4gASLcGMD5jARaCd0I+D+fIhDB\n+R/Uz2vbudyMc6a9WkQ24JxR+/t85tYQWF/QTAF8H1DVeBGZg/ODPcFn2TDgKeBqnGOj7qs2zpUh\n5PyfpPoZrkZOvsfi8Oc2t0C+Q6WO1UEUrcN1ED8B/wPGuuMTcX48WqlqTfdVXVWj3OmbcYpDctuM\nc/Zdy12mhrtc2zy2/x5wtYjE4JzxfOKzng0+266hqlGqeqnPsprPfiXiFAM18hnXiJxJIZqcYnCK\nGwLZh9zbfgrnrLCVqlYHridnxWh+sRYkAacIEAARqYxTjOVPohu7v/9NQSYCq4Cm7j48zJGVu9n7\n4dY33A9c7R6jGjhXo4eXyesz4s9mnOIa3/93NVX9wN+2c1PV9araV1WPA54BPnaPU0HHPdAYC/o+\nICKX4FxVzAae81m2L3ApztVOdZyinsMnHYXV0Od9I5zPbW6BfIdKHUsQwfMC0E1E2qhzPToFeME9\ne0JEosWtiAamAjeJSBdx1BeRk1R1G/AtME5EItxpJ0ge7dVV9Q+cy/LXgG9UNdmd9AuQ4lb+VXIr\nJVuJSIdAdkSdJo0fAk+KSDVxmjsO5d8rFIDjReROESnnVpS2wCnWOap9cEXgFEGkiEg0zg+nr23A\nCbnGBfoD8TFwqYh0EpHywKi8ZnT/b68Dz7sVlGE+yxW0zQggWVUPiEgLnGKu/ETgJOFdbgXuf91x\nh70GPC4izQBEpI2I1HCn5T4eU4DbRaSjO29VEblYRKoWEAPu/NeJyOErl704iSEL2On+zSsJzADq\nishd7j5UOxyDr4K+D+62p+BcTfUDeopID3fxCJwiwd3u/jzN0Z8w5P6/DXa3XxN4CHjfzzLH9B0q\nqSxBFJ0cH1JVTcS5ivivO2o4TqXWz+K0FPoWONGddzFwE05S2YtTBny47fX/4VS2rcQpZvkIp8I0\nL+/iVPC+4xNLFtATp+JuI04l3hQg8ij27y6cs74NOJfVb6vqGz7TFwHNcc4OHweuUtXdhdyHR4HT\ncCq6v+TfK6HDRgOPuK1S7jm8mz7T8zs7XgnciVMc9w/OWfoOnB8df+4DlgOLcZLvaP793uTejuZa\n7joRScapyM79o5N72Vnuaw3O/+gAOYvensdJ0t+KyF6chFHZnfYoMM09Hler6m/ArcB4t0hkDTlb\npvk7Pr7jugMr3NjHAdeqapqqpgJPAvPdbeX48VfVfUA34D84SWsNEOtnWwAPkMf3Aed4faaqs1Q1\nCac+bYqbEKfhFE9uBf7EqVA/Wrn3/113++twGjc8ecQCRfMdKnHsRjlzzETkRpwb1476TlyvuWeh\ne3BaB8V7HY8pXhICN12GMruCMGWOOPcIVHaTw1hgmSUHY45kCcKURZfhFC9twSlP7+1tOMZDVoSS\nDytiMsYY45ddQRhjjPGrxNwoJyJ2qWOMMYWgqoW6T6REXUF4fdt5qLxGjhzpeQyh8rJjYcfCjkX+\nr2NRohKEMcaY4mMJwhhjjF+WIEqg2NhYr0MIGXYs/mXH4l92LIpGiWnmKiJaUmI1xphQISJoKFZS\nu09n2i4iy/KZ5yVxnqD2h4icktd8xhhjilewi5jeAC7Ka6LbQ2NTVW0O3Ibz4BxjjDEhIKgJQlXn\nAbvzmeUynN4ZUdVFQJSI1AlmTMYYYwLj9Y1y0eTs0nirO267/9mNMeYofHoJbJzpdRSeUIXP/jz5\nmNbhdYI4KqNGjcp+Hxsbay0VjDH5K4PJIW4dxLkPfp2z/sAxrSvorZjcp499qX4ekykirwJz1H0U\nooisxnmQ+xFXENaKyRhz1Ma6jXfuLZu/Hamp6VSpUiE0WzG58nte7Bc4TxtDRDoBe/wlB2OMMXnL\nzMzyO75y5fJ+xwcq2M1c38V5JOCJIrJJRG4SkdtEZACAqs4ENorIOpzHDA4KZjzGGFPa/PLLVk47\nbTLffLOuyNdtN8oZY0qvUlzElJKSxiOPzOGllxahCl26NOaHH248Yr5juVGuRFVSG2OMga++WsPA\ngV+xeXMy4eHCPfecyciR5xX5dixBGGNMCXLwYAaDBs1k8+ZkTjutHlOmXMqpp9YLyrYsQZiSoQy3\nZzfGV6VK5Xj11UtYtSqRu+46g3LlgleVbHUQpmQYW6giVGOgycVw5VdeR+EZq4MwZUcprGw0xp9D\nhzJ59dVfGTDgNCpV8uan2hKEMcaEmIULN3PrrV+yYsVOEhMP8NhjXTyJwxKEMcaEiOTkNB56aDYT\nJixGFZo1q0mXLo09i8cShDHGhIBt2/bRocNktm5NoVy5MIYNO4sRI8495ruhj4UlCGOMCQF16lSl\nfft6REdHMmXKpbRt6/2TDyxBGGNMCBARpk27goiICoSHF0c3eQWzBFFS2H0AxpQae/ceJCqq0hHj\nq1c/cpyXQiNNmYJZcnDasxtTgqWlZTBqVByNGr3AunVJXodTILuCKGnsPgBjSqSffopnwIAZrF6d\nCMDMmWu5664zPI4qf5YgjDEmiPbsOcgDD3zH5Mm/A3DSSbWYPPlSzj23kceRFcwShDHGBNGuXQeY\nNm0Z5cuH8eCD5/Dgg509uzP6aJWMKI0xpoRq2rQmr7/+H9q2rUOrVsd7Hc5RsQRhjDFB1qdPG69D\nKBRrxWSMMUVg+fLtPPjg95SmXqftCuJY2f0JxpRpBw9m8MQTcxkzZj4ZGVl06FCfq65q6XVYRcIS\nxLEqzuRg9wEYE1Li4v5mwIAvWbvWuadh4MAOdO16gsdRFR1LEEXF7k8wpkz5/PPVXHHFBwC0bHkc\nkyf35OyzYzyOqmhZgjDGmELo3r0Z7drV4corT+aBB86mYsXS93Na+vbIGGOKQaVK5fj11wFBfSa0\n10rvnhljTBHIzMxi/Xr//SaV5uQAliCMMSZPS5du48wzp9Kly/9ISUnzOpxiZwnCGGNySU1NZ/jw\n7znttMksXvwPqrBhw26vwyp2Vgfhy+5pMKbMmzs3nv79p7N+/W5E4M47O/Lkk+cTEVHR69CKnSUI\nX4VNDnZ/gjGlRmpqOuvX76Z16+OZMuVSOnVq4HVInpGSclu4iGjQYx0rzl+7p8GYMu3zz1dzySXN\nKV8+3OtQjpmIoKpSmGXtCsIYY3K5/PIWXocQEqyS2hhT5mRkZDF27AIef/xHr0MJaXYFYYwpU37/\nPYFbbvmCJUu2Ua5cGP/3f+1o1Ki612GFJLuCMMaUCfv3H+K++77l9NOnsGTJNmJiopg+vbclh3wE\nPUGISHcRWS0ia0TkAT/TI0XkCxH5Q0SWi0i/YMdkjCl77rvvW8aOXQjA0KGdWLFiEBdf3NzjqEJb\nUIuYRCQMGA9cAPwDLBaR6aq62me2wcAKVf2PiNQG/hKRt1U1I5ixGWPKlhEjzmXVqkSee+5COnSo\n73U4JUKw6yA6AmtVNR5ARN4HLgN8E4QCEe77CGCXJQdjTFGLjo4kLq6f12GUKMEuYooGNvsMb3HH\n+RoPtBSRf4ClwJAgx2SMKcXWrUvKs3M9c3RCoZL6ImCJqtYHTgVeEZFqHsdkjClh0tMzGT16Hm3a\nTKR//y/IyrIbXo9VsIuYtgK+j1hq4I7zdRPwNICqrheRjUAL4NfcKxs1alT2+9jYWGJjY4s2WmNM\nibR48VZuvfVLli7dDkCjRlEcPJhBlSrlPY6s+MXFxREXF1ck6wpqVxsiEg78hVNJnQD8AvRR1VU+\n87wC7FDVR0WkDk5iaKeqSbnWZV1tGGOO8NBDsxkzZj5ZWUqTJtV59dWeXHhhU6/DChkh29WGqmaK\nyB3AtzjFWVNVdZWI3OZM1snAE8CbIrLMXWxY7uRgjDF5qVWrMiJw//1nMWpUbJm8aggW66zPl11B\nGFPiZGRksWrVTtq0qeN1KCHpWK4gQqGS2hhjCqSq+DtJLFcuzJJDkFiCMMaEvL/+SqRLl//x3nt/\neh1KmWKd9RljQtahQ5k888x8nnhiLmlpmSQmHqB379aEhRWqxMQcJUsQxpiQtHDhZm699UtWrNgJ\nwE03ncJzz11oyaEYWYIwxoScrCzltttmsGLFTpo1q8mkST05//wmXodV5liCMMaEnLAw4dVXezJj\nxhoeeeRcKle2pqtesGauvqyZqzGmlAnZG+U89eklsHGm11EYY/KRlaX8739/cNVVLYmMrOh1OCaX\n0tvMtbDJocnFRRuHMcavVat2ct55b9K//xc89NBsr8MxfpTeK4jDrLjImJCSlpbB00/P46mnfiI9\nPYs6dapy3nmNvA7L+BFQghCRCkCMqq4LcjzGmFJs375DdOw4hVWrEgG49db2jBnTlRo1KnscmfGn\nwAQhIpcAzwMVgCYicgowUlWvCHZwxpjSpVq1CnTsGE1mpjJ5ck/OO6+x1yGZfBTYiklEfsPprnuO\nqp7qjluuqm2KIT7fOI6uFZO1SDImJCUnp1GhQjiVKpX+Eu5QEOzO+tJVdU+ucfara4zJ1759h/yO\nj4ysaMmhhAgkQawSkWuAMBFpIiLjgJ+DHJcxpoTKzMxi/PhfiIkZx5IlCV6HY45BIAniDuA0IAv4\nFEgDhgQzKGNMybR8+XbOOecN7rzza3bvPshHH630OiRzDAK5zrtIVR8AHjg8QkSuxEkWxhjDwYMZ\nPPHEXMaMmU9GRhb160cwfnwPrrjiZK9DM8cgkCuIEX7GPVzUgRhjSq49ew7yyiuLycjIYuDADqxc\nOciSQymQ5xWEiFwEdAeiReR5n0mROMVNxhgDQN261XjttUupW7caZ58d43U4pojkV8S0A/gTOAis\n8BmfAgwPZlDGmJLnqqtaeh2CKWJ5JghVXQIsEZF3VPVgMcZkjAlRmzbtZeLExTz55AX24J4yIJBK\n6mgReRJoCVQ6PFJVTwxaVMaYkHK46erDD//A/v3pNGlSgwEDTvM6LBNkgSSIN4EngOeAHsBN2I1y\nxpQZS5du45ZbvuTXX/8BoFevllx6qZ0flgWBtGKqoqqzAFR1vaqOwEkUxphSbv78TZx22mR+/fUf\nGjSI5IsvevPhh72oVy/C69BMMQjkCiJNRMKA9SJyO7AVsE+HMWVAp04N6NChPh07RvPkk+cTEWEP\n9SlLAkkQQ4GqwF3Ak0AU0D+YQRljQkN4eBhz595EhQrhXodiPFBgglDVRe7bFOAGABGJDmZQxpji\nparEx++lcePqR0yz5FB25VsHISKni8jlIlLbHW4lItOARfktZ4wpOTZu3E2PHu/QocNkdu7c73U4\nJoTkmSBE5GngHeA64BsRGQXMAZYC1oTBmBIuIyOL555bQKtWE5g1az1ZWcrKlTu9DsuEkPyKmC4D\n2qlqqojUBDYDbVR1Q/GEZowJlmXLtnPTTdP5/XenO+4+fVozbtxF1KlTzePITCjJL0EcVNVUAFVN\nEpE1lhyMKR1SU9NZsiSBmJgoJk68hIsvbu51SCYE5ZcgThCRw116C87zqLO7+FbVK4MamTEmaM44\nowGffHIN3bo1pVq1Cl6HY0JUns+kFpEL8ltQVWcHJaI82DOpjTHm6B3LM6nz66yvWBOAMaZoqSrT\npi1l1apERo/u6nU4pgQKpKuNYyIi3UVktYisEZEH8pgnVkSWiMifIjIn2DEZU9qtW5dE165v0a/f\ndMaMmW/PhjaFEsid1IXmdtExHrgA+AdYLCLTVXW1zzxRwCvAhaq69fA9F8aYo5eensnYsQt59NEf\nOXgwg1q1KjNu3EWcckpdr0MzJVDACUJEKqpq2lGuvyOwVlXj3XW8j9N8drXPPH2BT1R1K4CqJh7l\nNowxrieemMtjj80F4IYb2jJ27IUcd1xVj6MyJVWBRUwi0lFElgNr3eF2IvJygOuPxrl/4rAt7jhf\nJwI1RWSOiCwWkRsCXLcxJpchQzpx5pkNmDXreqZNu8KSgzkmgVxBvAT0BD4HUNWlItKliGNoD5yP\n0yngQhFZqKrrcs84atSo7PexsbHExsYWYRjGlHw1a1Zm/vz+iNjT3sqquLg44uLiimRdgSSIMFWN\nz/WBywxw/VsB3yeYN3DH+doCJLqPNT0oInOBdkC+CcKYsmz79n3s3n2QFi2OrLKz5FC25T55fvTR\nRwu9rkBaMW0WkY6Aiki4iNwNrAlw/YuBZiLSSEQqAL2BL3LNMx04x113FeAMYFWA6zemTFFVpk79\nnRYtXqF3749JTw/0XM2YoxfIFcRAnGKmGGA78L07rkCqmikidwDf4iSjqaq6SkRucybrZFVdLSKz\ngGU4VyaTVXVlIfbFmFJtzZpdDBjwJT/+GA84D/NJTk6jVq0qHkdmSqs876TOnkGkpqomFVM8+cVh\nd1KbMuullxYxbNh3pKVlctxxVXjxxe707t3aipNMgYJyJ7WPxSLyF/AB8KmqphRmQ8aYwouIqEBa\nWib9+5/Cs89eSM2alb0OyZQBBV5BAIjIWTj1B/8B/gDeV9X3gxxb7hjsCsKUWarKr7/+w+mn28Mc\nzdE5liuIgLraUNUFqnoXTnPUZJwHCRljgsDfiZCIWHIwxS6QG+Wqich1IvIl8AuwEzgr6JEZU8b8\n808KV131IRMmLPY6FGOAwOog/gS+BJ5R1Z+CHI8xZU5WljJ58m888MD3JCen8fPPW7jllvZUrBjU\nrtKMKVAgn8ATVDUr6JEYUwatWrWTAQNmMG/eJgB69jyRV1652JKDCQl5fgpFZKyq3gt8IiJHFIra\nE+WMOXaDBs1k3rxN1KlTlZdf7sHVV7e0pqsmZOR3mvKB+3d8cQRiTFn08ss9ePnlRYwe3ZUaNazp\nqgktgdwod4eqji9oXLBZM1djjDl6wW7m2t/PuJsLszFjyiJV5dNPV7F9+z6vQzHmqORXB3Etzs1x\nTUTkU59JEcCeYAdmTGmwZUsygwfP5Isv/qJPn9a8++5VXodkTMDyq4P4BdiF00X3Kz7jU4AlwQzK\nmJIuMzOLiRN/5aGHZpOScojIyIp07hyDqloltCkx8kwQqroR2IjTe6sxJkAZGVnExr7J/PnOwxSv\nuKIFL7/cg+joSI8jM+bo5FfE9KOqniciuwHfml7B6aq7ZtCjM6YEKlcujI4do9m4cQ/jx/fgiitO\n9jokYwolz1ZMIhKmqlkiEu5vuqoW65NKrBWTKUn27z9ERkYWUVGVvA7FlHFBacXkc/d0QyDcTQhn\nArfhPDvamDIvNTXd7/iqVStYcjAlXiDNXD/HedxoU+ANoDnwblCjMibEqSoffPAnJ5zwEnFxf3sd\njjFBEUiCyFLVdOBK4GVVHQpYv8OmzNq0aS+XXvoevXt/wrZt+/jf/5Z6HZIxQRFIj2AZItILuAG4\n3B1XPnghGROaMjOzGD/+Fx5++Af2708nKqoizz7bjZtvbu91aMYERSAJoj8wCKe77w0i0gR4L7hh\nGRN69u07xJgx89m/P51evVry4ovdqVcvwuuwjAmaQB85Wg5o5g6uU9WMoEblPwZrxWQ8N3PmWjIz\ns7j00pO8DsWYgBxLK6YCryBEpDPwFrAV5x6IuiJyg6rOL8wGjSnJLr64udchGFNsAqmkHgdcrKpn\nq+pZwCXAi8ENyxjvJCYe4OGHZ5OeXqy3+hgTcgKpg6igqisPD6jqKhGpEMSYjPGEqvLOO8sZOnQW\niYkHqFatAg8+2NnrsIzxTCAJ4ncReRV42x2+Duusz5QyGzbsZuDAr/j22/UAdOnSmKuvbultUMZ4\nLJAEcTtwFzDMHf4JeDloERlTzFat2slpp00mNTWDGjUqMXbshfTrd4r1umrKvHwThIi0AZoCn6nq\nM8UTkjHFq0WL2pxzTgy1alXhhRcuok6dal6HZExIyK8314dwnhz3O3C6iDymqq8XW2TGFBMR4Ysv\n+lCpUiAX1MaUHfm1YroOaKuqvYDTgYHFE5IxwbNlS7Lf8ZYcjDlSfgkiTVX3A6jqzgLmNSak7dy5\nn+uv/5QWLcYTH29PzDUmEPmdNp3g8yxqAZr6PptaVa8MamTGFAFVZdq0pdxzz7ckJaVSuXI5fvst\ngUaNqnsdmjEhL78Ekfvp6uODGYgxRW3jxt3ceuuXzJ69EYBu3U5g4sRLaNrUHoZoTCDyeyb17OIM\nxJiidvBgBnPnxlOrVmXGjbuI669va01XjTkKQa9XEJHuIrJaRNaIyAP5zHe6iKSLiBVdmSJx8snH\n8eGHvVi9+g5uuKGdJQdjjlJQm26ISBhO0dQFwD/AYhGZrqqr/cw3GpgVzHhM2XP55S28DsGYEivg\nBCEiFVU17SjX3xFYq6rx7jreBy4DVuea707gY5zmtHkba2eA5khffbWG777bwAsvdPc6FGNKlQKL\nmESko4gsB9a6w+1EJNCuNqKBzT7DW8j1uFIRqQ9crqoTcVpLFZ0mFxfp6kxo2bZtH9de+zE9e77H\niy8u4vvvN3gdkjGlSiBXEC8BPYHPAVR1qYh0KcIYXgB86ybyThL28B+D03R16tQl3H//d+zZc5Aq\nVcrz+ONdiI1t7HVoxpQqgSSIMFWNz1XBF2hH+VuBGJ/hBu44Xx2A98XZQG2gh4ikq+oXuVc2atSo\n7PexsbHExsYGGIYpTSZO/JXBg2cC0L17MyZOvITGje2+BmMA4uLiiIuLK5J1FfjIURH5BBgDvIpT\nR3AncLbbBUdBy4YDf+FUUicAvwB9VHVVHvO/AXypqp/6mXZ0jxw1pdaBA+lccME07rqrI717t7bW\nScbkI6iPHMXpg+klnCuB7cD3BNgvk6pmisgdwLc49R1T3QcO3eZM1sm5Fwk4clNmValSngUL+lti\nMCbICryCCBV2BVH2JCensXnzXlq1Ot7rUIwpsYJ6BSEiU/BzZq+qAwqzQWMCMX36agYPnknFiuVY\nvnwgVaqU9zokY8qcQIqYvvd5Xwm4gpxNV40pMv/8k8Kdd37Np5861VQdO0aza9cBqlSJ8jgyY8qe\noy5icu9CHK6QAAAbQElEQVR6nqeqZwUnpDy3a0VMpdy77y5n4MCvSE5Oo1q1Cjz11PkMGnQ64eHW\n07wxhRXsSurcmgB1CrMxY/JTtWp5kpPT6NnzRCZMuJiGDe2qwRgvBdLMdTf/1kGEAUnAcFX9MMix\n5Y7DriDKgHnzNnH22Q2thZIxReRYriDyTRDuzWsN+ffmtiyvfqUtQZQuqmpJwJhicCwJIt/CXfcX\neaaqZrov+4U2x2TPnoPcfvsMnnrqJ69DMcYUIJA6iD9E5FRVXRL0aEyppap8+ukq7rzzaxIS9lGt\nWgUGD+5I9eqVvA7NGJOHPBOEiJRT1QzgVJznOKwH9uN0pqeq2r6YYjQl3JYtydxxx0ymT/8LgLPO\nasjkyT0tORgT4vKsgxCR31W1vYg09TddVdcHNbIj47ESrhKqV6+P+PjjlUREVGDMmK7cdlsHwsKs\n/sGY4hCsZq4CxZ8ITOnz7LPdKFcujOee60Z0dKTX4RhjApTfFcQW4Pm8FlTVPKcFg11BGGPM0QtW\nK6ZwoBoQkcfLmBzi4v5m48bdXodhjCkiBdZBFHM8ebIriNCVlJTKsGHfMXXqEi66qClff32d3eNg\nTIgIah2EMXlRVT78cAV33fUNO3bsp0KFcM4+uyFZWUp4uH18jCnp8ruCqKmqScUcT57sCiK0qCq9\nen3EJ584va527hzD5MmX0qJFbY8jM8b4CkodRCglBxN6RIRTTqlLVFRFJk/uSVxcP0sOxpQy9kQ5\nU2iHDmWya9cB6tWzNgvGhKqgddYXSixBeCctLYOKFQvTM7wxxmtB66zPmO+/30CrVhP47LNVXodi\njClmliCMX4mJB+jX73O6dXuL9et3M2nSb16HZIwpZlZuYHJQVd55ZzlDh84iMfEAFSuGM3Lkedx3\nX7E+YdYYEwKsDsLkkJaWQevWE1m3LokuXRozaVJPmjev5XVYxphCskpqU6QOd5nRr98pdke0MSWc\nJQhjjDF+WSsmc9T27z/EY4/9yP79h7wOxRgToqySugyaNWsdt9/+FX//vYfk5DSee+5Cr0MyxoQg\nSxBlyI4d+xk6dBbvvrscgHbt6nDtta08jsoYE6osQZQRCQkptG49kaSkVCpVKsejj8YydGgnypcP\n9zo0Y0yIsgRRRtSrF0HXrieQlJTKq69eQtOmNb0OyRgT4qwVUxly4EA6lSuXs6arxpQh1orJ5LBt\n2z6/46tUKW/JwRgTMEsQpUhKShp33/0NTZq8yIoVO7wOxxhTwlkdRCnx1VdrGDjwKzZvTiY8XJg/\nfzOtWh3vdVjGmBIs6FcQItJdRFaLyBoRecDP9L4istR9zRORNsGOqTTZsWM/1177MT17vsfmzcmc\ndlo9Fi++lQEDTvM6NGNMCRfUKwgRCQPGAxcA/wCLRWS6qq72mW0DcK6q7hWR7sAUoFMw4ypN0tIy\nmDlzLVWqlOfxx7tw111nUK6clRwaY45dsIuYOgJrVTUeQETeBy4DshOEqv7sM//PQHSQYypVGjaM\n4p13rqRt2zo0blzd63CMMaVIsBNENLDZZ3gLTtLIyy3A10GNqBT6z39O8joEY0wpFDKV1CLSBbgJ\nOCeveUaNGpX9PjY2ltjY2KDHFSoWLtzM228vY/z4i62pqjEmT3FxccTFxRXJuoJ6o5yIdAJGqWp3\nd3g4oKo6Jtd8bYFPgO6quj6PdZXJG+WSk9N46KHZTJiwGFV4772r6N27tddhGWNKiGO5US7YVxCL\ngWYi0ghIAHoDfXxnEJEYnORwQ17JoayaPn01gwfPZOvWFMqVC+P++8/issusOMkYUzyCmiBUNVNE\n7gC+xWlSO1VVV4nIbc5knQw8AtQEJohTdpKuqvnVU5QJn366iquu+hCAjh2jmTLlUtq2reNxVMaY\nssT6YgpR6emZdOnyP669thWDBp1OeLg1XTXGHD175GgppapWIW2MOSbWWV8JlpaWwZ9/+u83yZKD\nMcZLliA8NG/eJk45ZRJdu05j9+5Ur8MxxpgcLEF4YM+eg9x++ww6d36D1asTiYqqREKC/y66jTHG\nKyFzo1xZMWvWOm66aToJCfsoXz6M4cPP4aGHOlOpkv0rjDGhxX6VilnlyuVJSNjHmWc2YMqUS61L\nbmNMyLJWTB6YM2cj553XmLAwq4Q2xgSXNXM1xhjjlzVzDTEHD2YwYsQPDBv2ndehGGNModkVRBGL\ni/ubAQO+ZO3aJMLDhQ0bhhATE+V1WGVC48aNiY+P9zoMYzzRqFEj/v777yPGh3JnfWVGUlIqw4Z9\nx9SpSwBo2fI4Jk/uacmhGMXHx1MSTiKMCYZg3FhrCaKIjBjxA1OnLqFChXBGjOjMsGFnU7GiHV5j\nTMllRUxFZMeO/QwY8CWjR3elRYvaXodTJrmX0l6HYYwn8vr8WysmY7AEYcq2YCQIa8V0lJYu3cbK\nlTu9DsMYY4LOEkSAUlPTGT78e047bTL9+08nMzPL65CMKdFWrlzJ6aef7nUYJcKOHTto2bIl6enp\nxbpdSxAB+P77DbRpM5ExY+aTlaV07BjNoUOZXodlSpjGjRtTpUoVIiMjqV+/PjfddBMHDhzIMc+C\nBQu44IILiIyMpEaNGlx22WWsWrUqxzwpKSncfffdNGrUiMjISJo3b84999xDUlJSce7OMfvvf//L\nsGHDvA7jmBw6dIj+/fsTFRVF/fr1GTduXL7zP/nkkzRq1Ijq1avTt29f9u37t5PO+++/nxNPPJGo\nqChatmzJW2+9lT3t+OOP5/zzz2fSpElB2xe/VLVEvJxQi9+dd85UGKUwSlu3nqALF272JA5TMK8+\nI4Fq3Lix/vDDD6qqun37dm3Xrp2OGDEie/qCBQu0WrVq+vLLL+u+fft09+7dOmLECK1Ro4Zu3LhR\nVVUPHTqkHTp00AsvvFBXr16tqqo7d+7UJ598Ur/++uugxZ6RkVGk60tISNBatWppWlpaSMRTWMOH\nD9dzzz1X9+7dq6tWrdK6devqrFmz/M775ptv6sknn6xbt27V/fv362WXXaY33nhj9vRRo0bpmjVr\nVFV10aJFWqNGDV24cGH29Pnz52vr1q3zjCWvz787vnC/u4VdsLhfXn35X3nlF61Y8XF98sm5euhQ\naHwojX8lIUHMnj07e3jYsGHas2fP7OHOnTvrHXfcccRyPXr0yP4hmTJlitatW1cPHDgQ8Hb//PNP\n7datm9asWVPr1q2rTz/9tKqq9uvXTx955JHs+eLi4rRBgwY54h0zZoy2bdtWK1WqpGPGjNGrr746\nx7rvuusuHTJkiKqq7t27V2+++WatV6+eNmjQQEeMGKFZWVl+Y5o2bZp269Ytx7jRo0dr06ZNNSIi\nQlu1aqWfffZZ9rQ333xTzz77bB06dKjWqlUrO+6pU6fqySefrDVr1tTu3btrfHx89jJDhgzRhg0b\namRkpHbo0EF/+umngI9ZoOrXr6/ff/999vB///tf7dOnj995r776an322WezhxcsWKCVK1fW1NRU\nv/P/5z//0eeffz57OCMjQ6tUqaKbNm3yO38wEoQ11C/A7bd3oEePZjRpUsPrUMyxGluENxLde2yt\npbZs2cLXX39N165dAUhNTWXBggU8/vjjR8x7zTXX8PDDDwMwe/ZsunfvTuXKlQPazr59++jWrRvD\nhg1jxowZpKens3Llyjznz32z1fvvv8/XX39NrVq12L59O4899hj79++natWqZGVl8dFHHzF9+nQA\nbrzxRurVq8eGDRvYt28fPXv2JCYmhltvvfWI7SxfvpyTTjopx7hmzZoxf/586tSpw0cffcT111/P\n+vXrqVOnDgCLFi2ib9++7Nixg/T0dKZPn87o0aOZMWMGzZo1Y/To0fTp04f58+cD0LFjR0aNGkVk\nZCQvvvgivXr1Ij4+ngoVKhwRz5gxYxg9enSOlkCH34uI3+K7PXv2kJCQQNu2bbPHtWvXjs8//zzP\n4+srKyuLtLQ01q5dS5s2bXJMS01NZfHixQwePDh7XHh4OM2aNWPp0qU0bNgwoG0cK6uDcGVkZJGV\ndeSXPixMLDmYInP55ZcTGRlJTEwMderUYdSoUQAkJSWRlZVFvXr1jlimXr16JCYmArBr1y6/8+Rl\nxowZ1KtXj7vvvpsKFSpQtWrVo6oYHjJkCPXr16dixYrExMTQvn17PvvsM8BJVofXt337dr7++mvG\njRtHpUqVqF27NnfffTfvvfee3/Xu2bOHiIiIHOOuuuqq7GTQq1cvmjdvzi+//JI9PTo6mkGDBhEW\nFkbFihWZNGkSDz74ICeeeCJhYWEMHz6cP/74g82bNwPQt29fqlevTlhYGEOHDiUtLY2//vrLbzwP\nPPAAu3fvJikpid27d+d4n1fdzr59+xARoqL+7S0hMjKSlJQUv/N3796d1157jfj4ePbu3cszzzwD\ncEQ9FMDtt9/OqaeeyoUXXphjfEREBHv27PG7/mCwKwjg998TuOWWLxg8+HRuvrm91+GYYDnGs/6i\nMH36dLp06cJPP/1E3759SUxMzK6QDgsLIyEhgRNPPDHHMgkJCdSu7dx8WatWLRISEgLe3ubNm2na\ntGmh423QoEGO4T59+vDee+9x/fXX895779G3b18ANm3aRHp6enbyOlxEERMT43e9NWrUOOKHdNq0\naYwbNy67P6H9+/dnJ0bgiLPm+Ph4hgwZwr333pu9TRFh69atNGzYkOeee47XX389+3ilpKTkWN+x\nqlatGgDJycnZ/5+9e/cekfgO69+/P1u2bCE2NpbMzEzuvfdeZsyYccQxvv/++1m5ciVz5sw5Yh0p\nKSlUr169yPahIGX6CmL//kPcd9+3nH76FJYs2cb48Yv9XkUYU1QOF1907tyZG2+8MfvHrUqVKpx5\n5pl89NFHRyzz4YcfZhdFde3alVmzZpGaGtgzzBs2bMj69ev9TqtatWqOs1d/iSd3kVOvXr2Ii4tj\n69atfPbZZ9kJomHDhlSqVIldu3Zln3nv2bOHZcuW+d1227ZtWbNmTfbwpk2bGDBgABMmTMg+g2/V\nqlWOG79yxxITE8OkSZNISkrK3ua+ffvo1KkT8+bN49lnn+Xjjz/OXl9kZGSeN1I+/fTTREREEBkZ\nmeN1eJw/1atXp169eixdujR73NKlS2nVqpXf+UWEkSNHsnHjRjZt2sTJJ59MdHQ00dHR2fOMHDmS\nWbNm8d1332UnoMMyMzNZt24d7dq187v+oChs5UVxvyjiCshvvlmrjRu/oDBKw8Ie1Xvu+UZTUgrX\nosKEhqL+jBS13JXUO3fu1KpVq+qyZctUVXXevHnZrZhSUlI0KSlJH374Ya1Ro4auW7dOVVXT0tK0\nY8eO2qNHD129erVmZWVpYmKiPvXUU35bMaWkpGj9+vX1xRdf1LS0NE1JSdFFixapqlPhffLJJ2tS\nUpImJCRop06dtGHDhnnGe1iPHj20W7du2r59+xzjL7/8ch0yZIgmJydrVlaWrl+/Xn/88Ue/x2L7\n9u1au3bt7FZMK1eu1MqVK+uaNWs0MzNTX3/9dS1XrpxOnTpVVZ1K6s6dO+dYx2effaatW7fWFStW\nqKrqnj179KOPPlJV1ZkzZ2p0dLRu27ZN09LS9NFHH9Vy5cr53Z9jMXz4cI2NjdXdu3frypUrtW7d\nuvrtt9/6nTcpKUnXr1+vqqorVqzQ1q1b62uvvZY9/amnntLmzZvr9u3b/S6/YMECbdWqVZ6x5PX5\nx1oxHZ3MzCxt336Swig95ZRXdfHirUW2buOdUE8QTZo0OeIHatCgQTlaBs2fP19jY2O1WrVqGhUV\npT179tSVK1fmWCY5OVmHDh2qDRs21IiICG3WrJnee++9mpSU5He7K1as0AsuuEBr1Kih9erV0zFj\nxqiq6sGDB/Xaa6/VyMhIbdeunb7wwgs5EoS/eFVV33rrLQ0LC9OxY8ceEdfAgQO1QYMGWr16dW3f\nvr1+8MEHeR6Pa665Jsf0ESNGaM2aNfW4447Te++9V2NjY/NNEKqqb7/9trZp00ajoqI0JiZGb775\nZlVVzczM1P79+2tkZKTWr19fn3322Tz351ikpaVlb6du3br6wgsv5JherVo1nTdvnqqqrlmzRk86\n6SStWrWqNm7c+Ih5RUQrVaqkERERWq1aNY2IiMhucaaqOnjwYH355ZfzjCUYCaLM9sX022//MHv2\nRoYO7UT58uFFtl7jHeuLqWRZtWoV/fr1Y9GiRV6HEvJ27txJbGwsS5Ys8dsKC6yzviJNEKb0sQRh\nyjLrrO8opadn8vzzC0lKCqxCzxhjzL9KbTPXX37Zyq23fsmyZdv5888dvP76ZV6HZIwxJUqpSxAp\nKWk88sgcXnppEarQpEl1evdu7XVYxhhT4pSqBJGSkkabNhOJj99LeLhwzz1nMmpULFWqlPc6NGOM\nKXFKVYKIiKhIjx7NWLz4H6ZMuZRTTw28SwJjjDE5lbpWTAcOpFOhQjjlypXq+nfjR+PGjYmPj/c6\nDGM80ahRo+xuSnyFdDNXEekOvIDTYmqqqo7xM89LQA9gP9BPVf/wM0+OBJGYeIDatasELW5jjCkN\nQraZq4iEAeOBi4BWQB8RaZFrnh5AU1VtDtwGvJrfOg8dyuSJJ+YSEzOOhQs3Byny0BYXF+d1CCHD\njsW/7Fj8y45F0Qh2OUxHYK2qxqtqOvA+kLu96WXANABVXQREiUgdfytbuHAz7dtP4pFH5pCamsH3\n328IZuwhyz78/7Jj8S87Fv+yY1E0gl1JHQ34nuZvwUka+c2z1R23PffKzj77dVShWbOaTJrUk/PP\nb1LU8RpjjHGVqFZM4eFh3H//WTzyyLlUrmxNV40xJpiCWkktIp2AUara3R0ejtOz4BifeV4F5qjq\nB+7wauA8Vd2ea10lo7mVMcaEmMJWUgf7CmIx0ExEGgEJQG+gT655vgAGAx+4CWVP7uQAhd9BY4wx\nhRPUBKGqmSJyB/At/zZzXSUitzmTdbKqzhSRi0VkHU4z15uCGZMxxpjAlJgb5YwxxhSvkLvdWES6\ni8hqEVkjIg/kMc9LIrJWRP4QkVOKO8biUtCxEJG+IrLUfc0TkTZexFkcAvlcuPOdLiLpInJlccZX\nnAL8jsSKyBIR+VNE5hR3jMUlgO9IpIh84f5WLBeRfh6EGXQiMlVEtouI/4eAU8jfzcI+ii4YL5yE\ntQ5oBJQH/gBa5JqnB/CV+/4M4Gev4/bwWHQCotz33cvysfCZbzYwA7jS67g9/FxEASuAaHe4ttdx\ne3gsHgSePnwcgF1AOa9jD8KxOAc4BViWx/RC/W6G2hVEkd5YV8IVeCxU9WdV3esO/oxz/0hpFMjn\nAuBO4GNgR3EGV8wCORZ9gU9UdSuAqiYWc4zFJZBjoUCE+z4C2KWqGcUYY7FQ1XnA7nxmKdTvZqgl\nCH831uX+0cvrxrrSJpBj4esW4OugRuSdAo+FiNQHLlfViUBpbvEWyOfiRKCmiMwRkcUickOxRVe8\nAjkW44GWIvIPsBQYUkyxhZpC/W6WqBvljH8i0gWn9dc5XsfioRcA3zLo0pwkClIOaA+cD1QFForI\nQlVd521YnrgIWKKq54tIU+A7EWmrqvu8DqwkCLUEsRWI8Rlu4I7LPU/DAuYpDQI5FohIW2Ay0F1V\n87vELMkCORYdgPdFRHDKmnuISLqqflFMMRaXQI7FFiBRVQ8CB0VkLtAOp7y+NAnkWNwEPA2gqutF\nZCPQAvi1WCIMHYX63Qy1IqbsG+tEpALOjXW5v+BfAP8H2Xdq+72xrhQo8FiISAzwCXCDqq73IMbi\nUuCxUNUT3FcTnHqIQaUwOUBg35HpwDkiEi4iVXAqJVcVc5zFIZBjEQ90BXDL3E8ESmsvn0LeV86F\n+t0MqSsItRvrsgVyLIBHgJrABPfMOV1Vc3eGWOIFeCxyLFLsQRaTAL8jq0VkFrAMyAQmq+pKD8MO\nigA/F08Ab/o0/xymqkkehRw0IvIuEAvUEpFNwEigAsf4u2k3yhljjPEr1IqYjDHGhAhLEMYYY/yy\nBGGMMcYvSxDGGGP8sgRhjDHGL0sQxhhj/LIEYUKGiGSKyO9uN9W/uzcC5jVvIxFZXgTbnON2F/2H\niPwkIs0LsY7bROR69/2NIlLXZ9pkEWlRxHEucu+gL2iZISJS6Vi3bcouSxAmlOxX1faqeqr7d1MB\n8xfVTTx9VPUUnN4unzvahVV1kqq+7Q72w6cTNFUdoKqriyTKf+OcSGBx3g1UKaJtmzLIEoQJJUd0\nE+BeKcwVkV/dVyc/87R0z6p/d8+wm7rjr/MZP9G92zy/7c4FDi97gbvcUhF5TUTKu+NHuw/h+UNE\nnnHHjRSRe0XkKpw+od52l63knvm3d68ynvGJ+UYReamQcS4E6vusa4KI/CLOA3FGuuPudOeZIyKz\n3XEXisgC9zh+4HbDYUyeLEGYUFLZp4jpE3fcdqCrqnbA6WvnZT/L3Q68oKrtcX6gt7jFOtcCZ7nj\ns4DrCtj+f4DlIlIReAPopartcB5GM1BEauJ0Kd7aPZN/wmdZVdVPcDqB6+teAR30mf4JcIXP8LU4\nnQsWJs7uwOc+ww+5Xay0A2JFpLWqvozTGVusql4gIrWAh4EL3GP5G3BvAdsxZVxI9cVkyrwD7o+k\nrwrAeHEekZgJ+KsjWAg8LCINgU9VdZ2IXIDT5fVi94y8Ek6y8ecdEUkF/sZ56NBJwAafDhD/BwwC\nXgFSReQ14CucJ9f5c8QVgKomish6EemI06vqSaq6QEQGH2WcFXG68PZ9ZGRvEbkV5/tcF2gJ/EnO\nzts6uePnu9spj3PcjMmTJQgT6oYC21S1rYiEA6m5Z1DV90TkZ6An8JXbWZsA/1PVhwPYRl9VXXJ4\nwD3b9vcjn+n+wF8A9ALucN8H6gOcq4XVwGeHN3e0cbpFVeOBq0SkMc6VwGmqmiwib+AkmdwE+FZV\nC7o6MSabFTGZUOKv7D0KSHDf/x8QfsRCIk1UdaNbrPIF0Bbn2dRXi8hx7jw18mkVlXu7fwGNROQE\nd/gG4Ee3zL66qn4D3ONuJ7cUIDKP7XyG8+jH3jiPx6SQcf4XOENETnS3tQ9IEac76x4+8yf7xPIz\ncLZP/UyVwrTYMmWLJQgTSvy1SpoA9BORJTh9+e/3M881bsXxEqAVME1VVwEjgG9FZClOl9B1/Sx7\nxDZVNQ2nO+SP3WUzgVdxfmxnuOPm4lzd5PYm8OrhSmrf9avqHpznMsSo6q/uuKOO063bGAvcr6rL\ngD/c9b4NzPNZZgrwjYjMdp9LfRPwnrudBThFacbkybr7NsYY45ddQRhjjPHLEoQxxhi/LEEYY4zx\nyxKEMcYYvyxBGGOM8csShDHGGL8sQRhjjPHLEoQxxhi//h+UjDLoj88p9QAAAABJRU5ErkJggg==\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7fbc5e27aa10>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.plot(fpr,tpr,color='darkorange',\n",
    "         lw=2, label='ROC curve (area = %0.2f)' % roc_auc)\n",
    "plt.plot([0, 1], [0, 1], color='navy', lw=2, linestyle='--')\n",
    "plt.xlim([0.0, 1.0])\n",
    "plt.ylim([0.0, 1.05])\n",
    "plt.xlabel('False Positive Rate')\n",
    "plt.ylabel('True Positive Rate')\n",
    "plt.title('Receiver operating characteristic example')\n",
    "plt.legend(loc=\"lower right\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 41,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "15"
      ]
     },
     "execution_count": 41,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "len(tpr)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 42,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([ 0.03703704,  0.40740741,  0.40740741,  0.74074074,  0.74074074,\n",
       "        0.85185185,  0.85185185,  0.88888889,  0.88888889,  0.92592593,\n",
       "        0.92592593,  0.96296296,  0.96296296,  1.        ,  1.        ])"
      ]
     },
     "execution_count": 42,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "tpr"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 43,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([ 0.        ,  0.        ,  0.03846154,  0.03846154,  0.07692308,\n",
       "        0.07692308,  0.11538462,  0.11538462,  0.19230769,  0.19230769,\n",
       "        0.38461538,  0.38461538,  0.80769231,  0.80769231,  1.        ])"
      ]
     },
     "execution_count": 43,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "fpr"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 44,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "15"
      ]
     },
     "execution_count": 44,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "len(fpr)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 47,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([  9.90558624e-01,   9.11436081e-01,   8.86400938e-01,\n",
       "         7.69172072e-01,   7.62400866e-01,   5.97232044e-01,\n",
       "         5.89095771e-01,   4.91873413e-01,   4.15975571e-01,\n",
       "         3.84383053e-01,   2.35399887e-01,   1.31808177e-01,\n",
       "         5.21235634e-03,   2.64048972e-03,   3.51882009e-05], dtype=float32)"
      ]
     },
     "execution_count": 47,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "thresholds"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 53,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0.38438305"
      ]
     },
     "execution_count": 53,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "thresholds[9]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 52,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(0.92592592592592593, 0.19230769230769232)"
      ]
     },
     "execution_count": 52,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "tpr[9],fpr[9]"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "score=sess.run(activation_OP,feed_dict={X:trainX})\n",
    "score_pred = [b[1] for b in score]\n",
    "actual = [b[1] for b in trainY]\n",
    "fpr,tpr,thresholds = roc_curve(actual,score_pred)\n",
    "roc_auc = auc(fpr,tpr)# lets try ROC on training data"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 54,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<matplotlib.legend.Legend at 0x7fbc5d2aaad0>"
      ]
     },
     "execution_count": 54,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAYgAAAEZCAYAAACNebLAAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7fbc5e35e110>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "score=sess.run(activation_OP,feed_dict={X:trainX})\n",
    "score_pred = [b[1] for b in score]\n",
    "actual = [b[1] for b in trainY]\n",
    "fpr,tpr,thresholds = roc_curve(actual,score_pred)\n",
    "roc_auc = auc(fpr,tpr)# lets try ROC on training data\n",
    "\n",
    "plt.plot(fpr,tpr,color='darkorange',\n",
    "         lw=2, label='ROC curve (area = %0.2f)' % roc_auc)\n",
    "plt.plot([0, 1], [0, 1], color='navy', lw=2, linestyle='--')\n",
    "plt.xlim([0.0, 1.0])\n",
    "plt.ylim([0.0, 1.05])\n",
    "plt.xlabel('False Positive Rate')\n",
    "plt.ylabel('True Positive Rate')\n",
    "plt.title('Receiver operating characteristic example')\n",
    "plt.legend(loc=\"lower right\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 91,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": [
    "def pred(x,threshold):\n",
    "    if x >= threshold :\n",
    "        return 1;\n",
    "    else:\n",
    "        return 0;"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 92,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": [
    "pred = [pred(i,0.39) for i in score_pred]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 93,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0.86255924170616116"
      ]
     },
     "execution_count": 93,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "accuracy_score(actual,pred)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 80,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([  9.90789533e-01,   9.57410991e-01,   9.56822932e-01,\n",
       "         9.18812990e-01,   9.18606281e-01,   9.04708445e-01,\n",
       "         9.04326439e-01,   8.91598701e-01,   8.87227356e-01,\n",
       "         8.58119071e-01,   8.57847989e-01,   8.22800636e-01,\n",
       "         8.22089911e-01,   8.12487364e-01,   8.11531723e-01,\n",
       "         7.55397260e-01,   7.47315526e-01,   7.36158192e-01,\n",
       "         7.27279544e-01,   7.06188619e-01,   6.89285755e-01,\n",
       "         6.68463409e-01,   6.58181131e-01,   6.55458391e-01,\n",
       "         6.51911497e-01,   6.29335999e-01,   5.99199831e-01,\n",
       "         5.90839982e-01,   5.58441043e-01,   5.04830062e-01,\n",
       "         4.49685037e-01,   4.23222423e-01,   3.78207654e-01,\n",
       "         3.73239547e-01,   2.02710629e-01,   1.66968077e-01,\n",
       "         1.28060877e-01,   1.17892981e-01,   7.65089616e-02,\n",
       "         7.18089193e-02,   5.42636886e-02,   4.96369600e-02,\n",
       "         4.76536974e-02,   4.66688350e-02,   2.76231120e-04,\n",
       "         2.62755377e-04,   1.60164582e-05], dtype=float32)"
      ]
     },
     "execution_count": 80,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "thresholds"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python [Root]",
   "language": "python",
   "name": "Python [Root]"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 2
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython2",
   "version": "2.7.12"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 0
}