train_start.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "from __future__ import division, print_function, absolute_import"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/home/shared/anaconda3/lib/python3.6/site-packages/h5py/__init__.py:36: FutureWarning: Conversion of the second argument of issubdtype from `float` to `np.floating` is deprecated. In future, it will be treated as `np.float64 == np.dtype(float).type`.\n",
      "  from ._conv import register_converters as _register_converters\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "WARNING:tensorflow:From /home/shared/anaconda3/lib/python3.6/site-packages/tensorflow/contrib/learn/python/learn/datasets/base.py:198: retry (from tensorflow.contrib.learn.python.learn.datasets.base) is deprecated and will be removed in a future version.\n",
      "Instructions for updating:\n",
      "Use the retry module or similar alternatives.\n",
      "WARNING:tensorflow:From /home/coder_lzx_dev/Bearing_detect/autoencoder.py:10: read_data_sets (from tensorflow.contrib.learn.python.learn.datasets.mnist) is deprecated and will be removed in a future version.\n",
      "Instructions for updating:\n",
      "Please use alternatives such as official/mnist/dataset.py from tensorflow/models.\n",
      "WARNING:tensorflow:From /home/shared/anaconda3/lib/python3.6/site-packages/tensorflow/contrib/learn/python/learn/datasets/mnist.py:260: maybe_download (from tensorflow.contrib.learn.python.learn.datasets.base) is deprecated and will be removed in a future version.\n",
      "Instructions for updating:\n",
      "Please write your own downloading logic.\n",
      "WARNING:tensorflow:From /home/shared/anaconda3/lib/python3.6/site-packages/tensorflow/contrib/learn/python/learn/datasets/mnist.py:262: extract_images (from tensorflow.contrib.learn.python.learn.datasets.mnist) is deprecated and will be removed in a future version.\n",
      "Instructions for updating:\n",
      "Please use tf.data to implement this functionality.\n",
      "Extracting /tmp/data/train-images-idx3-ubyte.gz\n",
      "WARNING:tensorflow:From /home/shared/anaconda3/lib/python3.6/site-packages/tensorflow/contrib/learn/python/learn/datasets/mnist.py:267: extract_labels (from tensorflow.contrib.learn.python.learn.datasets.mnist) is deprecated and will be removed in a future version.\n",
      "Instructions for updating:\n",
      "Please use tf.data to implement this functionality.\n",
      "Extracting /tmp/data/train-labels-idx1-ubyte.gz\n",
      "WARNING:tensorflow:From /home/shared/anaconda3/lib/python3.6/site-packages/tensorflow/contrib/learn/python/learn/datasets/mnist.py:110: dense_to_one_hot (from tensorflow.contrib.learn.python.learn.datasets.mnist) is deprecated and will be removed in a future version.\n",
      "Instructions for updating:\n",
      "Please use tf.one_hot on tensors.\n",
      "Extracting /tmp/data/t10k-images-idx3-ubyte.gz\n",
      "Extracting /tmp/data/t10k-labels-idx1-ubyte.gz\n",
      "WARNING:tensorflow:From /home/shared/anaconda3/lib/python3.6/site-packages/tensorflow/contrib/learn/python/learn/datasets/mnist.py:290: DataSet.__init__ (from tensorflow.contrib.learn.python.learn.datasets.mnist) is deprecated and will be removed in a future version.\n",
      "Instructions for updating:\n",
      "Please use alternatives such as official/mnist/dataset.py from tensorflow/models.\n"
     ]
    }
   ],
   "source": [
    "import tensorflow as tf\n",
    "import numpy as np\n",
    "import matplotlib.pyplot as plt\n",
    "import autoencoder as ae\n",
    "#import classification as cf\n",
    "import autoencoder_depart as aed\n",
    "import depart_classification as dcf\n",
    "from input import Bearing_dataset\n",
    "\n",
    "%matplotlib inline\n",
    "plt.rcParams['figure.figsize'] = (10.0, 8.0) # set default size of plots\n",
    "plt.rcParams['image.interpolation'] = 'nearest'\n",
    "plt.rcParams['image.cmap'] = 'gray'"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "learning_rate = 0.01\n",
    "X = tf.placeholder(dtype = tf.float32, shape = [None, ae.num_input], name = 'layer_1_input')\n",
    "#layer_2_X = tf.placeholder(dtype = tf.float32, shape = [None, ae.num_hidden_1], name = 'layer_2_input')\n",
    "#layer_3_X = tf.placeholder(dtype = tf.float32, shape = [None, ae.num_hidden_2], name = 'layer_3_input')\n",
    "is_training = tf.placeholder(dtype = tf.bool, name = \"train\")\n",
    "layer_1_para = aed.layer_para(X, aed.scope_name[0], is_training, 2.6e-3)\n",
    "layer_2_X = aed.Encoder(X, aed.scope_name[0], is_training)\n",
    "layer_2_para = aed.layer_para(layer_2_X, aed.scope_name[1], is_training, 8e-3)\n",
    "layer_3_X = aed.Encoder(layer_2_X, aed.scope_name[1], is_training)\n",
    "layer_3_para = aed.layer_para(layer_3_X, aed.scope_name[2], is_training, 6e-3)\n",
    "layer_4_X = aed.Encoder(layer_3_X, aed.scope_name[2], is_training)\n",
    "layer_4_para = aed.layer_para(layer_4_X, aed.scope_name[3], is_training, 6e-3)\n",
    "\n",
    "def run_autoencoder(session, loss_val, Xd, predict,\n",
    "                    drop_prob = 0.4,\n",
    "                    num_steps = 40, batch_size=256, print_every = 100,\n",
    "                    training = None, plot_losses = False):\n",
    "    extra_update_ops = tf.get_collection(tf.GraphKeys.UPDATE_OPS)\n",
    "\n",
    "    train_indices = np.arange(Xd.shape[0])\n",
    "    np.random.shuffle(train_indices)\n",
    "\n",
    "    variables = [loss_val, training, extra_update_ops]\n",
    "   \n",
    "    iter_cnt = 0\n",
    "    correct = 0\n",
    "    losses = []\n",
    "    for i in range(num_steps):\n",
    "        start_idx = (i * batch_size)%Xd.shape[0]\n",
    "        idx = train_indices[start_idx: start_idx + batch_size]\n",
    "\n",
    "        feed_dict = {X: Xd[idx, :], is_training: training is not None}\n",
    "        actual_batch_size = Xd[idx].shape[0]\n",
    "        loss,  _, _ = session.run(variables, feed_dict = feed_dict)\n",
    "        losses.append(loss * actual_batch_size)\n",
    "        #correct += np.sum(corr)\n",
    "\n",
    "        if (iter_cnt%print_every) == 0:\n",
    "            print(\"Iteration{0}: with minibatch training loss = {1:.3g}\".format(iter_cnt, loss))\n",
    "        iter_cnt += 1\n",
    "    total_loss = np.sum(losses)/Xd.shape[0]\n",
    "    print(\"Overall loss = {0:.3g} \".format(total_loss))\n",
    "    if plot_losses:\n",
    "        plt.plot(losses)\n",
    "        plt.grid(True)\n",
    "        plt.title('Loss')\n",
    "        plt.xlabel(\"minibatch number\")\n",
    "        plt.ylabel('minibatch loss')\n",
    "        plt.show()\n",
    "    return total_loss\n",
    "\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [],
   "source": [
    "train_data = ae.mnist.train.images\n",
    "train_labels = np.asarray(ae.mnist.train.labels, dtype=np.int32)\n",
    "eval_data = ae.mnist.test.images\n",
    "eval_labels = np.asarray(ae.mnist.test.labels, dtype=np.int32)\n",
    "init = tf.global_variables_initializer()\n",
    "saver = tf.train.Saver()\n",
    "b_train_data = Bearing_dataset.train_data\n",
    "b_train_labels = Bearing_dataset.train_label\n",
    "b_train_labels = tf.one_hot(b_train_labels, 10)\n",
    "b_eval_labels = Bearing_dataset.eval_label\n",
    "b_eval_labels = tf.one_hot(b_eval_labels, 10)\n",
    "b_eval_data = Bearing_dataset.eval_data\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Tensor(\"one_hot:0\", shape=(7000, 10), dtype=float32)\n",
      "(7000, 1200)\n"
     ]
    }
   ],
   "source": [
    "print(b_train_labels)\n",
    "print(b_train_data.shape)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [],
   "source": [
    "writer = tf.summary.FileWriter(\"/tmp/Bearing_detect/\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Pretrain layer_1:\n",
      "Iteration0: with minibatch training loss = 1.48\n",
      "Iteration50: with minibatch training loss = 0.882\n",
      "Iteration100: with minibatch training loss = 0.711\n",
      "Iteration150: with minibatch training loss = 0.656\n",
      "Iteration200: with minibatch training loss = 0.726\n",
      "Iteration250: with minibatch training loss = 0.675\n",
      "Iteration300: with minibatch training loss = 0.624\n",
      "Iteration350: with minibatch training loss = 0.64\n",
      "Iteration400: with minibatch training loss = 0.648\n",
      "Iteration450: with minibatch training loss = 0.574\n",
      "Iteration500: with minibatch training loss = 0.613\n",
      "Iteration550: with minibatch training loss = 0.642\n",
      "Iteration600: with minibatch training loss = 0.563\n",
      "Iteration650: with minibatch training loss = 0.642\n",
      "Iteration700: with minibatch training loss = 0.625\n",
      "Iteration750: with minibatch training loss = 0.568\n",
      "Overall loss = 36.6 \n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f13c29af668>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Pretrain layer_2:\n",
      "Iteration0: with minibatch training loss = 1.01\n",
      "Iteration50: with minibatch training loss = 0.966\n",
      "Iteration100: with minibatch training loss = 0.615\n",
      "Iteration150: with minibatch training loss = 0.217\n",
      "Iteration200: with minibatch training loss = 0.0836\n",
      "Iteration250: with minibatch training loss = 0.066\n",
      "Iteration300: with minibatch training loss = 0.0488\n",
      "Overall loss = 8.74 \n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f13bccb18d0>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Pretrain layer_3:\n",
      "Iteration0: with minibatch training loss = 0.911\n",
      "Iteration50: with minibatch training loss = 0.75\n",
      "Iteration100: with minibatch training loss = 0.256\n",
      "Iteration150: with minibatch training loss = 0.0871\n",
      "Iteration200: with minibatch training loss = 0.0265\n",
      "Overall loss = 5.7 \n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f13bf5dac50>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Pretrain layer_4:\n",
      "Iteration0: with minibatch training loss = 0.675\n",
      "Iteration50: with minibatch training loss = 0.401\n",
      "Iteration100: with minibatch training loss = 0.0878\n",
      "Iteration150: with minibatch training loss = 0.027\n",
      "Overall loss = 3 \n"
     ]
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAYgAAAEWCAYAAAB8LwAVAAAABHNCSVQICAgIfAhkiAAAAAlwSFlzAAALEgAACxIB0t1+/AAAADl0RVh0U29mdHdhcmUAbWF0cGxvdGxpYiB2ZXJzaW9uIDIuMS4yLCBodHRwOi8vbWF0cGxvdGxpYi5vcmcvNQv5yAAAIABJREFUeJzt3Xd4W+XZ+PHvLVmecezYcUz2XhCSkJhMhsOGtoRR1sseTWkLLS8dQOmgLx0USvmVlrakQNmk7A2BhhgIkITsQRKyE2c5y44db+n5/XGOHDmRE41zbDm6P9ely9LR0Tm3j2XderYYY1BKKaUO5mnrAJRSSiUmTRBKKaXC0gShlFIqLE0QSimlwtIEoZRSKixNEEoppcLSBKGUUiosTRBKRUBENojIGW0dh1KtSROEUkqpsDRBKBUHEfmOiKwRkT0i8qaIdLO3i4g8JCJlIlIhIktEZJj93Hki8pWIVIrIFhH5Sdv+FkqFpwlCqRiJyGnAH4BLga7ARmCa/fRZwCnAICAXuAzYbT/3OPBdY0w2MAz4qBXDVipiKW0dgFLt2JXAE8aYBQAichewV0T6AA1ANjAEmGuMWRHyugbgWBFZbIzZC+xt1aiVipCWIJSKXTesUgMAxpgqrFJCd2PMR8DfgEeAHSIyVUQ62rteDJwHbBSRj0VkfCvHrVRENEEoFbutQO/gAxHJAvKBLQDGmIeNMaOB47Cqmn5qb//SGDMZ6AK8DrzYynErFRFNEEpFzici6cEb1gf79SIyUkTSgN8Dc4wxG0TkRBEZKyI+YD9QC/hFJFVErhSRHGNMA7AP8LfZb6TUYWiCUCpy7wI1IbeTgV8CrwDbgP7A5fa+HYF/YbUvbMSqevqT/dzVwAYR2QfcDFzVSvErFRXRBYOUUkqFoyUIpZRSYWmCUEopFZYmCKWUUmFpglBKKRVWux5J3blzZ9OnT5+YXrt//36ysrKcDcghiRqbxhWdRI0LEjc2jSs6scY1f/78XcaYgiPuaIxpt7fRo0ebWM2cOTPm17otUWPTuKKTqHEZk7ixaVzRiTUuYJ6J4DNWq5iUUkqFpQlCKaVUWJoglFJKheVagrDnq5krIotFZLmI/Mbe/qSIrBeRRfZtpL1dRORhe/GVJSIyyq3YlFJKHZmbvZjqgNOMMVX2hGWzROQ9+7mfGmNePmj/c4GB9m0s8A/7p1JKqTbgWgnCbiyvsh/67NvhJn6aDDxtv242kCsiXd2KTyml1OG5OlmfiHiB+cAA4BFjzB0i8iQwHquEMQO40xhTJyJvA/cZY2bZr50B3GGMmXfQMacAUwAKCwtHT5s2jVhUVVXRoUOH2H4xlyVqbBpXdBI1Lkjc2DSu6MQa16RJk+YbY4qOuGMkfWHjvWGtyTsTa/3droAAacBTwK/sfd4BTgp5zQxg9OGOG+s4iO0VNeamv79v6hv9Mb3ebUdbn2u3aVzRS9TYNK7oHBXjIIwx5UAJcI4xZpsdYx3wb2CMvVsp0DPkZT2wVuxy3MJNe/lwYyMPfvC1G4dXSqmjgpu9mApEJNe+nwGcAawMtiuIiAAXAMvsl7wJXGP3ZhoHVBhjtrkR2znDulLcI4V/fryW219cxNWPz2FbRY0bp1JKqXbLzV5MXYGn7HYID/CiMeZtEflIRAqwqpkWYa2oBdZqXecBa4Bq4HoXY+OKoalsb8zgvaXbaQwE+NP0r3nw0hFunlIppdoV1xKEMWYJcEKY7ae1sL8BfuBWPAdL8wpv3XoSAH+avoqpn67jppP7MrRrx9YKQSmlElpSj6T2eT34vB6+V9yf7LQUbn1hIR9/vZPtFbXU1Os68kqp5Naup/t2Sm5mKg9fcQJ3v7aMa5+Y27S9IDuNXnmZjOqVy/eKB+DzCg1+Q15WahtGq5RSrUMThK14cBdm/PhUZqwoY19tA7ur6ti0p5qNu6t54rMNPDt7E3WNfnIyfJT8ZBI5mb62DlkppVylCSJEus/LN4YfOnj76x2VPDFrPVlpKTw+az3//GQtd5wzpA0iVEqp1qMJIgKDCrO57+LhAOyqquPfn63nugl9KOyY3saRKaWUe5K6kToWt585iEAALn30Cx79eC0XPPIZHyzfDsAf31/Jf7/a0cYRKqWUMzRBRKl3fhbP3jSWhsYAf3hvJYs2lzN/414Anpu9kYc/Wt3GESqllDM0QcRgTN88pv/vKUy/7RSy01Ko9wcAqPcHWFJaQene6jaOUCml4qcJIkbZ6T4GH5NNaoqHBjtBNPitmXHfX7a9LUNTSilHaCN1nHxeD/WNAfwBgz9gJYjXF23B5/UwomcuI3vmtnGESikVG00QcfKlWIPngqWI3Ewfy7bsY9mW5WSnp/Dq9yYwsDC7jaNUSqnoaRVTnFK9Hur9gaZ2iBsn9uXhK07gle+NJy3Fy/VPfsme/fVtHKVSSkVPE0ScglVMDY1WgsjJ9HH+iG6M7p3HY9cWUbavjp+8tDi4CJJSSrUbmiDiFGykDpYgfN4Dl3Rkz1zu/sZQPlpZxm/fWUF1fWNbhamUUlHTNog4pXqtBNHQaJoeh7pmfG++2rqPx2et541FW7n1tAF4PcLrC7fw+4uOZ5C2TyilEpQmiDgFq5iaShApzROEiPDHbw/nkqIe3D99Fb9+czkAHoF73lzOczeNxVpcTymlEosmiDj5UjxU1/ipt9sgUr3hP+yL+uTxnynjmL1uDwbD19srueetr/hoZRmnDy1szZCVUioi2gYRp1Svh4bGQFM319SUli+piDC+fz4T+nfmynG96VeQxS9fX6brYSulEpImiDilpojVBhGmkfpwfF4PD19+AvtqG7nm8bns2FfrZphKKRU11xKEiKSLyFwRWSwiy0XkN/b2viIyR0RWi8h/RCTV3p5mP15jP9/Hrdic5AuOg2iMLkEADOuew2PXFrF5bzVn/Plj3li0xa0wlVIqam6WIOqA04wxI4CRwDkiMg74I/CQMWYgsBe40d7/RmCvMWYA8JC9X8ILVjHVR1DFFM64fvm896NTGFyYzW3/WUTJqjI3wlRKqai5liCMpcp+6LNvBjgNeNne/hRwgX1/sv0Y+/nTpR107/GleKj3m6aJ+g7u5hqJvp2zePrGMQwuzOaHLyxkS2XA6TCVUipq4uYIXxHxAvOBAcAjwAPAbLuUgIj0BN4zxgwTkWXAOcaYUvu5tcBYY8yug445BZgCUFhYOHratGkxxVZVVUWHDh1i+8VCPPtVHZ9vbeS6YWn8fVEdv52YQY/s2PLuzuoA986upd4f4OYR6YzsklidzJy6Zk7TuKKXqLFpXNGJNa5JkybNN8YUHXFHY4zrNyAXmAmcDKwJ2d4TWGrfXw70CHluLZB/uOOOHj3axGrmzJkxvzbUb99ebob84j3z2oJS0/uOt826nVVxHW/L3moz4d53zIm//dAYY8ym3fvNawtKnQg1bk5dM6dpXNFL1Ng0rujEGhcwz0Tw2d0qvZiMMeVACTAOyBWR4FfjHsBW+36pnTCwn88B9rRGfPE4dKqN+GrFuuVmMCzfy/46a1qOl+aXctt/FrG1XLvCKqVal5u9mApEJNe+nwGcAazAKkl8297tWuAN+/6b9mPs5z+yM11C83k9NAYMdU0D5eK/pCKC3/7Vg91nS1btjPu4SikVDTdLEF2BmSKyBPgS+NAY8zZwB3C7iKwB8oHH7f0fB/Lt7bcDd7oYm2OC3VqD3/ij7cUUjldoWnwo+HOm9m5SSrUy11pBjTFLgBPCbF8HjAmzvRa4xK143JKW0jxBRDMOoiWeMAniszW7qGv0k5bijfv4SikVCR1JHacDJQh/s8fx8AgEjNWBIJggquv9PDFrA/M2JHyzjFLqKJFY/SjboYOrmOJtpAYrQYBVemgMBMhKtUoNf3x/JQD/vf0UBnTRacKVUu7SEkScgm0OVfWNpHo9jkzdHcwxjQGDPwCZaSm8f9spvPjd8aSmePj3ZxviPodSSh2JJog4BUsM1XWNjpQe4EAJImAM/kAArwg98zIZ0zePC0d255UFpZRX6zrXSil3aYKIU2pIG4QTPZgAPHYpJFiC8HoOJJ7rT+pDbUOApz7f6Mi5lFKqJZog4tRUxVTX6EgDNRyoYgoE7BJESIIYckxHzjv+GB6ZuYYV2/bR4A8QCCT8cBGlVDukCSJOwaRQXe9cgpDQNggDKZ7mVVe/veB4cjJ9XPnYHIb9ejq3v7jIkfMqpVQoTRBxaurFVO9vGhMRr4NLEJ6DEkReVir/77KR9OiUwejenXh90VZmrd4V5khKKRU7TRBxSg0ZKOdUCcLTrBeTOaQEATBxQGfevOUknrjuRHrnZ/KrN5ZR1+h35PxKKQWaIOKW2lTF5MeX4mwvJr+dIDyH6Tqb7vNy7+RhrNu1n//339WOnF8ppUATRNxCk4ITE/XBoQki5QjdZ08ZVMDlJ/bk0Y/X8tbirTrzq1LKEZog4hRareRcLyYrIfiNoTFgmvViasnd3xhK904Z3PrCQibc9xHzN+51JBalVPLSBBGn0FKDc+MgrJ/BEoQ3gtHZ2ek+3r71ZJ65cQxZqV6mzd3kSCxKqeSlCSJOoUnBrSqmSEoQADkZPk4eWMA3h3fjnaXbmuaHUkqpWGiCiJMbVUyxJoigS0/sQXW9n3eWbnMkHqVUctIEEafQEoTPjSomE32CGNWrE/0Ksnh29kbawaJ8SqkEpQkiTqET9DldxdQYYwlCRLj5lP4sKa1g+vLtjsSklEo+miDi5POENlI7Mw4i2CgdMC0PlDuSi0Z1p39BFg9MX0Wjva61UkpFQxNEnDweaSpFOD6S2n/kgXItSfF6+Nk5Q1i7cz8/nLaQmnodZa2Uio4mCAcEE4PTVUxNJYgY15k469hCfn7eEN5btp0L//4Zy7ZUOBKfUio5uJYgRKSniMwUkRUislxEfmRvv0dEtojIIvt2Xshr7hKRNSKySkTOdis2pwUThNON1I0RTLVxOCLClFP688R1J7J7fz2TH9EkoZSKnJsliEbgx8aYocA44Acicqz93EPGmJH27V0A+7nLgeOAc4C/i4jXxfgcE+zJ5Mp6ECa2NohQkwZ34Zkbx+APGNaUVTkQoVIqGbiWIIwx24wxC+z7lcAKoPthXjIZmGaMqTPGrAfWAGPcis9Jwaolp6b7Di1BNPrNIdN9xyLDZ+Vavy4upJSKUEprnERE+gAnAHOAicAtInINMA+rlLEXK3nMDnlZKWESiohMAaYAFBYWUlJSElNMVVVVMb/2YI31tQBs2rCekpLSuI9XW1MDCEuWLqW6pp6dO3bEHevOaqsn01crVpBfuSamYzh5zZykcUUvUWPTuKLjdlyuJwgR6QC8AtxmjNknIv8A7gWM/fNB4AYg3NfkQ77uGmOmAlMBioqKTHFxcUxxlZSUEOtrD5az8GN2VFcxZNAAiif2jft4W976CKhhyNDjSFmznB7dCykuPj6+Y5bXwCcfMWjwYIpP7BXTMZy8Zk7SuKKXqLFpXNFxOy5XezGJiA8rOTxnjHkVwBizwxjjN8YEgH9xoBqpFOgZ8vIewFY343NKUy8mh6qYDiw5GrAHysV/zKYZYnVIhFIqQm72YhLgcWCFMebPIdu7hux2IbDMvv8mcLmIpIlIX2AgMNet+JzU1IvJ6UbqpoFy8R83eIiAMRhjeGTmGtaUVcZ9XKXU0cvNKqaJwNXAUhFZZG/7OXCFiIzEqj7aAHwXwBizXEReBL7C6gH1A2NMuxjdlerSOIh4BsodeswDo7N3VtbxwPRVPPrxWqZeU8S4fvlxH18pdfRxLUEYY2YRvl3h3cO85nfA79yKyS3BqiWn14OId6BcqANVTNYiRAD76/3c/Ox8vrz7DMdKP0qpo4d+KjjAtak24hwo1+yYngMJItjV9dRBBZRXN7B4c3ncx1dKHX00QTjgQBuEw5P1OTRQDmiaEdYYq2QCcNKAzojAp6t3xX18pdTRRxOEA9yqYmoqQTiQIJrWmDCG4Fi5vKxUhvfIZdYaTRBKqUNpgnCAW43U9Y1Wn1QnShAeObSKyeMRTh7QmUWby9lX2xD3OZRSR5cjfqKJyI9EpKNYHheRBSJyVmsE11443c314AQR7YJB4QSPEQiYpiomj8BJAzvjDxi+WLs77nMopY4ukXyi3WCM2QecBRQA1wP3uRpVO+NWFVO937kEcaCb64H5mLwijOrViZwMH28tbhdjEpVSrSiST7Tgp9N5wL+NMYsJ3301ablegnBkHIT102+aVzGlpni4aFR3pi/fzq6qurjPo5Q6ekTyiTZfRD7AShDTRSQb0AkbQvjspUadboOoc7CKSUTwSPMqpmDiuXJsLxr8hpfnxz/RoFLq6BHJQLkbgZHAOmNMtYjkYVUzKVuaw3MxeUQQOVDF5MRAObASTWgvpmDiGdAlmzF98nhi1nqyUr1cPLoHmamtMtGvUiqBRfKJNh5YZYwpF5GrgF8AuixZCKfHQYDVcylYxeTEQDmwShGBg6qYgu44dzAd0lP45RvLufX5hRij60YolewiSRD/AKpFZATwM2Aj8LSrUbUzaT5nSxBgJQUnu7mCVaUUrooJYHTvPD76cTG/+MZQZqws45UFWxw5p1Kq/YrkE63RWF8nJwN/Mcb8Bch2N6z2ZfLI7tx/8XCy032OHbNZCcKpBOER/IEDvZjCHfaGiX0Z0yeP37y1nG0VNY6cVynVPkWSICpF5C6smVnfsdeJdu6T8ChQ2DGdS0/seeQdo+DxyIE2CIcShEesaTYCYaqYQs97/7eH0+g33PnKUq1qUiqJRZIgLgPqsMZDbMdaBvQBV6NSzUoQTvRiAuvDP2Cs+Z0Od9w+nbO489whfPz1Tl7Snk1KJa0jJgg7KTwH5IjIN4FaY4y2QbjM6/E4niC8Is2n2jhM4/fV43ozuncnHpi+iur6RkfOr5RqXyKZauNSrJXdLgEuBeaIyLfdDizZeT1Q53duoBwcKEGYg7q5trTvz88bws7KOh77dL0j51dKtS+RdHa/GzjRGFMGICIFwH+Bl90MLNmluFCCsAbKNZ9q43BG987j7OMKefTjtVw5tpcjMSil2o9I2iA8weRg2x3h61QcPB6ob7RWXHW0iimkDSKSpa5/ds4QahsD/PWjNY7EoJRqPyIpQbwvItOBF+zHl3GYZUOVM1I8Hkcn6wO7iikQ0ospgqqr/gUduOzEnjw3ZyPHTkx3JA6lVPsQSSP1T4GpwHBgBDDVGHOH24ElO484O9138Dj+CHoxHey20weS4vHwxhpdM0KpZBJRVZEx5hVjzO3GmP81xrwWyWtEpKeIzBSRFSKyXER+ZG/PE5EPRWS1/bOTvV1E5GERWSMiS0RkVOy/VvvnThuENJvuO9IpPLp0TOeSoh7M2d7Inv31jsSilEp8LSYIEakUkX1hbpUisi+CYzcCPzbGDAXGAT8QkWOBO4EZxpiBwAz7McC5wED7NgVrio+k5fGETrXh3CyxzabaiCLxXDWuN40BeGneZkdiUUolvhY/eYwx2caYjmFu2caYjkc6sDFmmzFmgX2/EliBNchuMvCUvdtTwAX2/cnA08YyG8gVka5x/G7tWkrISGqHZhG3p9owBOzJ2qPpPjuoMJtBnTw8P3dTUxuGUuro1ipzOotIH+AEYA5QaIzZBlYSEZEu9m7dgdCvp6X2tm0HHWsKVgmDwsJCSkpKYoqpqqoq5te6raqqiv1VXhr81gfx4oULqVzvjfu4NdU1lO2s4asVewCYO3c26zIizz4Tu/j596pq7ps2gwndEmc68ET9WyZqXJC4sWlc0XE7Ltf/y0WkA/AKcJsxZp+0/K013BOHfFU1xkzFajSnqKjIFBcXxxRXSUkJsb7WbSUlJXTKTYWKvQCcWFTE8T1y4j5u9uJPyctNZ+CgQli2lIkTxtM1JyPi1wfMTBZUp/La+hpuvWiio5MTxiNR/5aJGhckbmwaV3TcjsvV8Qwi4sNKDs8ZY161N+8IVh3ZP4NjLEqB0BnvegBJu1ByaPWPo72YAiHjIKIcoe0R4d7Jx7Grqk7HRSiVBFxLEGIVFR4HVhhj/hzy1JvAtfb9a4E3QrZfY/dmGgdUBKuiklFoUnByHITfENU4iIMN75HLhSO78+zsjZRXa48mpY5mkczFdJHdJbUiyl5ME7GmCD9NRBbZt/OA+4AzRWQ1cKb9GKzBd+uANcC/gO/H8gsdLVxJEAImZEW5WI/7nVP6UV3v57k5mxyJSymVmCJpg7gf+JYxZkU0BzbGzCJ8uwLA6WH2N8APojnH0cyNBNE0m6s58DgWQ7t25OSBnXny8w3cdHJf0lLib0BXSiWeSKqYdkSbHFT8QpOCYwsG2W0QJoq5mFpy5dhe7KysY2mpLk+u1NGqxRKEiFxk350nIv8BXsdaOAiAkEZn5YLQBOHYkqMiNAYCcVcxARRkW/MyVdXpWhFKHa0O9x3yW/atI1ANnBWy7Zvuh5bcQqt/nCtBYE21EWMvplBZaVa1UnW9n63lNQy/ZzrLtmhpQqmjSYslCGPM9a0ZiGrO6w0pQTi1YJAER1LHX4LI9Flvnep6Pxt272dfbSNfrN3NsO7xj9dQSiWGSHoxPSUiuSGPO4nIE+6GpVJcaIPwBtektqfaiCfxZKRaJYia+kaq66x1K1Zsi6Rzm1KqvYikmXK4MaY8+MAYsxdr2gzlomYD5bzOliAOVDHFfqzQKqb99prVX2mCUOqoEtGKcsEpucGarptWmsMpmTXr5upgFVPAHijnETjMtCdHlG53bd1f76e63ipBrN1Z1TQDrVKq/Yvkg/5B4HMReRlrbqRLgd+7GpVyZxyE58B03/Ee0+MRMnxeauob2W/3ZGrwG1aXVXJcN22HUOpoEMmKck8DFwM7gJ3ARfY25SJ3EsSBFeWcaPjOTPVSHVKCAFixrTLu4yqlEsMRSxAi8owx5mrgqzDblEvcqGISsRqpA4H4SxAAmWlWgthf10iq14PHow3VSh1NImmDOC70gYh4gdHuhKOCgh/gIs4OlAsErF5MTiSdTF8K1fWN7K9vpEN6CoMLs5m1ehdl+2odiFYp1dYOt+ToXSJSCQwPmaSvEmt67jdaep1yRvAD3KkurnCgiilgDE4USjKCVUx1fjJTvVw9vg/rdlUx6U8lLNpcfuQDKKUS2uGWHP2DMSYbeCBkqdFsY0y+MeauVowxKQW7tjo1SC54rEAA/A5VMWWleamxu7lmpabw7dE9+PB/TyXd5+XhGasdiFgp1ZYiaaS+yx4cN0ZETgneWiO4ZOZGCcIjWG0QDvRiAsjwpTR1c820x0X06ZzF1eN789HKMtaUaYO1Uu1ZJCOpbwI+AaYDv7F/3uNuWCqYGJxqf4ADK8oFHOzFFOzmmpV6oL/D1eN6k5bi4bFP18d9DqVU24mkkfpHwInARmPMJKxR1DtdjUo1JQZHSxBNU204V8UU7OYaHFkNkN8hjQtGdueNRVupbfAf5ghKqUQWSYKoNcbUAohImjFmJTDY3bBUMDE4NQYCrComv92LyYkSRIYvpWmqjdASBMC3RnSjpsHPx1/rdwml2qtIEkSpPVnf68CHIvIGsNXdsJTHhQThDU614VAbhDVQrpH9dQfaIILG9ssjJ8PH9GXb4z6PUqptHHGgnDHmQvvuPSIyE8gB3nc1KtVUgkiJZ9m3g3g8wXEQJq6J+oIy07wEDJRX1x9SgvB5PZx5bCHTl2+nvjFAaopzv4dSqnVE9F8rIqNE5IfAcKDUGFMfwWueEJEyEVkWsu0eEdkiIovs23khz90lImtEZJWInB3LL3M0CVYBOZgfrDWpg1NtOFGC8FmlhoCBzNRDv2ucc9wxVNY28tHKsrjPpZRqfZH0YvoV8BSQD3QG/i0iv4jg2E8C54TZ/pAxZqR9e9c+x7HA5Vijts8B/m6P2E5arpUgjLUmtSMjqUOSQlbaoX+ukwd1pl9BFne8soQlpeWsKatqWg9bKZX4Ivn0uQI40Rjza2PMr4FxwJVHepEx5hNgT4RxTAamGWPqjDHrgTXAmAhfe1QKthE42ATh+EC54KJBEL4EkZbi5anrx+DzCuf/7TPO+PPH/PTlJXGfVynVOiJJEBuA9JDHacDaOM55i4gssauggutMdAc2h+xTam9LWl675OBkCcLrsdajdqoXU2ipIVwJAqBnXiYvfnc8v/jGUK4Y04uX55fyyvzSuM+tlHJfi43UIvJXrPUf6oDlIvKh/fhMYFaM5/sHcK99nHux1pq4AQj3aRW2LkJEpgBTAAoLCykpKYkpkKqqqphf67aqqipWl64CoLp6v2Nxbt5Ujz9g2LlrF9V1JurjHnzNVu0+MMZh3dcrKClveXqNAUC/ToYFnTzc/epiMveuJiPFmeJRov4tEzUuSNzYNK7ouB3X4XoxzbN/zgdeC9keczTGmB3B+yLyL+Bt+2Ep0DNk1x600JXWGDMVmApQVFRkiouLY4qlpKSEWF/rtpKSEo7rOgCWLSa3YzbFxSc5ctyFDV/D2tXkdMqDmgaKiydGHVfoNeu0uRy+/AyAsaNHMqF/5yMeo1P/vVz498+p6Nifc8f0iur8kcaVKBI1Lkjc2DSu6LgdV4sJwhjzlNMnE5Guxpht9sMLgWAPpzeB50Xkz0A3YCAw1+nztyduTbUB0OgPONK20ayKKUwbRDgje+YyoEsHXppfyuUOJQillDsOV8X0ojHmUhFZSpjqHmPM8MMdWEReAIqBziJSCvwaKBaRkfbxNgDftY+1XERexFqUqBH4gTEmqedocGOqjWCCaPAHHOnFlHGEXkzhiAiXjO7BH95bydqdVfQv6BB3HEopdxzua9+P7J/fjOXAxpgrwmx+/DD7/w74XSznOho1TbXh4HTfwUPV+w1pDgxcC46DgPC9mFpy4aju3D99FS/O28xd5w6NOw6llDsOtx7ENvvnxnC31gsxOQV7GTk91QZYVUyOjIOIoYoJoEt2OmcM7cJL80qpa0zqgqJSCS2SgXIXichqEakIWVlOFx52mRuT9R1og3BmHESq19N0nIPnYjqSq8b1Zs/+et5bqnM1KZWoIqlnuB843xiTE7KyXEe3A0t2Xldmcz3QBuFE47eIkOnzkpriweeNrspqYv/O9MnP5JmGGS37AAAgAElEQVTZG3V0tVIJKpL/6h3GmBWuR6Ka8brQSO1paoMI4HXosJlpXrJSo58VxeMRrp/Yl/kb9/Lz15bS6A84E5BSyjGRVBzPE5H/YE33XRfcaIx51bWo1IGpNlzqxeTUWteZqSmkeGL7cL9mfG/KKmt5ZOZayqsbePiKE6IuiSil3BNJgugIVANnhWwzgCYIF7lSgghpg3Aq8WT4vPhiLI6ICD89ewidMlP57TsruP3FxTx8+UjEwZ5bSqnYRbIexPWtEYhqzpUShP3BW+9QLyawFg1qDMT3rf+mk/tRVdfI//vvaq6b0JvRvfMciU0pFZ/DDZT7mTHm/pA5mZoxxvzQ1ciSnDttEM72YgK4ZkIf/IH42w9uOKkvj8xcw/TlOzRBKJUgDleCCDZMzzvMPsolwW/4Tg6Ua6piCjjTiwng/BHdHDlOx3QfEwd05v1l27nr3CFazaRUAjjcXExv2T8dn5NJHZkb3VyD7b8NfuNYLyYnnX3cMdz16lJWbKvk2G7ak1qpthbJQLkiEXlNRBbY6zgsERFd9cVlbo6DAGfbNpxy5rGFiMD05dbguUDAaPdXpdpQJL2YngN+CiwF9L+1lbgxkrpZgkjAKpzOHdLok5/F2p1VAEx5Zh65man86ZIRbRyZUskpkgSx0xjzpuuRqGY8rlQxHTiWk20bTkr3ealtsL6HbNxdzdItFW0ckVLJK5IE8WsReQyYgQ6UazWulyASsIoJIN3naZrAr7bRz459dezYV0thx/QjvFIp5bRIEsT1wBDAx4EqJh0o5zKPG72YQg6VqAOW01O81DbYCcIuSSwpreDMYzVBKNXaIkkQI4wxx7seiWomxe5m5HWwu1H7qGLysKuqEaApUSwpLefMYwvbMiylklIk3yNni8ixrkeimmnqxeTCOIiD7yeStJASRJ1dglhcqu0QSrWFSEoQJwHXish6rDYIAcyRlhxV8QkmBkeXHJX2UYKobfTjDxjq7S6uS0vLMcbo4DmlWlkkCeIc16NQh0jxWIU7r8e5xoL20Ujtpa4h0NRQ3Ssvk017qrn2319y3rBjuHxMrzaOUKnkEclkfbq8aBsI5gUnG5NDc00ijoOAYDdXf1MD9bnDjuHLDXtYvqWCVdv3cdmJPbUkoVQrca0vi4g8ISJlIrIsZFueiHxoL2H6oYh0sreLiDwsImvskdqj3IqrvfB5PaR4hIwo1no+kmZVTAnaiynN56G2MdDUDtG3cxavfn8it505iB376ti0p7qNI1Qqebj5MfEkh1ZP3QnMMMYMxBpXcae9/VxgoH2bAvzDxbjahXSfl+duGsslRT0cO6anPfRiSvFS3xigxk4Q6T5rtbpxfa0ZXues29NmsSmVbFxLEMaYT4CD/5snA8HJ/54CLgjZ/rSxzAZyRaSrW7G1F2P75dMx3efY8dpLGwRARU2D/dh6iw7o0oH8rFRmr9/dZrEplWycq7+ITKExZhuAMWabiHSxt3cHNofsV2pv23bwAURkClYpg8LCQkpKSmIKpKqqKubXus2t2NaV+5vub9q4gZKSrVG9vjWuWekGKzF8Mns+AF+vWE76rlUA9O3g5+OvtlJSUt7qccUiUeOCxI1N44qO23G1doJoSbivs4csUgRgjJkKTAUoKioyxcXFMZ2wpKSEWF/rNrdiyy+tgNmzAOjfrx/FxQMSIq5Q2+ZugpVL6TVgCCxYzImjTmB8/3wANvjWc89bXzFgxBi652bw6epdjO2XxxezPk3Iv2UyvsfipXFFx+24Wrupckew6sj+WWZvLwV6huzXA4ju6606otBmh8TtxWS9Jcurm1cxAZw0sACANxZt5b8ryrjmibk88tGa1g9SqSTR2gniTeBa+/61wBsh26+xezONAyqCVVHKOc2m2kjUXkwpVhtEeVMbhLfpuQFdOlA8uIAnZq3noQ+/BuCJzzZQVR+2sKmUipOb3VxfAL4ABotIqYjcCNwHnCkiq4Ez7ccA7wLrgDXAv4DvuxVXMgtNEIlegqiorrcfe5s9f8ukAezeX89X2/Zx00l92V/fyLvrG1o9TqWSgWttEMaYK1p46vQw+xrgB27FoiyeZuMgEjRBHFKCaP4dpqhPHuP75bOlvIafnTOEPfvreW3hFt5buo1zj0/6jm9KOSpRGqlVK2g+3XdiJoi0g7u5pngP2WfqNaNp8BtSUzzce8Ewlqzfxg+nLeSZrFTG9ctv1XiVOpolaE20ckN7qmI60Eh9aILITveRl5UKQFZaCrePTqd7bgZ3vrKkaQS2Uip+miCSSLuoYjqoBJGWcuS3aKZP+N2Fx7NhdzV/015NSjlGE0QSabYeRGLmh6YEUV5dT2qKJ+IR3xMHdOaCkd2Y+uk69u6vdzNEpZKGJogkEjr/UsJWMdklhoqahqb7kfruqf2pbwzwyoJSN0JTKulogkgiodN9J3oVU8CEb384nKFdOzK6dyeem7MJq2OcUioemiCSiLcdtEGEtjlEmyAArhzbi/W79vPA9FUs26JLlSoVD00QScTTDqqYUux1MODQMRCROO/4rozsmcvfS9byzb/O4jdvLdeeTUrFSMdBJJFm60EkaAkCrJJDVV1jTCWIdJ+X138wkZ2VdTwycw3//mwDtQ1+/nCRLqGuVLS0BJFE2sM4CDhQcgg3SC5SBdlp3HP+cdx4Ul+mfblZq5uUioEmiCTSvBdTGwZyBMEJ+9JiqGI62A9PH0huho973/4q7mMplWw0QSQRaQdTbUBICSKGKqaD5WT4uG5CX+as30NFTQMrt+/jpqfm8fWOyriPrdTRThNEEmlWxZTQCcLb7Ge8CrLTAKip9/Pl+j38d8UOJv/tMz5aucOR4yt1tNIEkUSadXNN6DYIO0FEOVCuJRmp1nFqGvzU2D2a8rJSeXiGTsuh1OFogkgiyVjFBJBhH6em3k9NfQCAC0/ozpLScsqrdVoOpVqiCSKJiEhT43RC92JKCVYxOVWCsHpz1zT4qW5oJDXFw6QhBQQMzFqzy5FzKHU00gSRZIIlh8QuQTjbBhEsQdQ2+Kmt95Ph8zKiRy4d01P45OudjpxDqaORJogkEyw5JHB+aJpuw5UqpgY/maleUrweThrYmU++3qXzNinVAk0QSaYpQSRwhkhzugQR0khdbZcgAE4dVMD2fbV8uWEvAIs3l3PTU/P4zVvLHTmvUu2dTrWRZJqqmBK5DaKpkdqZ7y/pISWI2gZ/0+NvDu/Gnz/8mt+9u4Jvj+7BL19fhs8rNPgNkwZ34ZRBBY6cX6n2qk1KECKyQUSWisgiEZlnb8sTkQ9FZLX9s1NbxHa0CxYc2kUbRBxTbYRqqmJqOFDFBNZypT87ewiLN5fzy9eXcdqQLsz5+Rn07ZzFr95YRmVtwxGPvbOyjo+P0I5R1+jnmdkbue+9lUfcV6lE0pZVTJOMMSONMUX24zuBGcaYgcAM+7FyWDAxtI9eTM4kiMzQXkz1fjJSDxz3whO6M75fPicP7MzfrxxFXlYq/zf5ODbuqeaU+2cy9ZO11NS3PBvs47PWc+0Tc1m3s4ptFTU8/cUGGvyBZvv8o2Qtv3x9GY9+spbr/j2X5+dscuT3UsptidQGMRl4yr7/FHBBG8Zy1AomhsQuQThbxRRs9LbGQfibJR6PR3juprE8c+PYpu0nDyzgte9PZFj3HH7/7kpOfWAmK7btC3vs9buqAHh29ibueGUpv3pjOVOenkd1fSMA++saefLzDZwxtAtf/eYcigcV8PPXlrJg015Hfjel3CRt0YNDRNYDewEDPGqMmSoi5caY3JB99hpjDqlmEpEpwBSAwsLC0dOmTYsphqqqKjp06BDTa93mZmw/mllNRZ3h9ydl0K1DdB/ArXXN/ruxgWdX1POzE9M5Nv/IpYhI4prywX5O7+1jwY5G+uZ4uHlEekSxrNrj568La+nd0cNPT8xg4z4/XTI9ZKRYCfaXn9WwuTJAikCjgeGdvSzd5Wd0oZfrBjby2a40XlhZzy/GpjOgk5eaRsOtH1VT3COFq45NiygGNyTq+1/jik6scU2aNGl+SO1Ni9qqkXqiMWariHQBPhSRlZG+0BgzFZgKUFRUZIqLi2MKoKSkhFhf6zY3Y8v4fAYVdbWMHTuG/gXRvbFa65rt+HITrFjK2KITGN07z5G4sj75gIJjumF2badPjy4UF0e2PkQx4C1Yx2/fWcFr23N4Y9FWMnxeppzSj9vOGMjuj6YzokcOi0sr6JmXwUu3ncpjn67ngemrKMxM5ZNtAcb2zeOmC8c3HfP0bfNYuKmck0859YgluUZ/gFueX8j/jO3laKN5or7/Na7ouB1Xm1QxGWO22j/LgNeAMcAOEekKYP8sa4vYjnZNjdSJ3AZhV/WkOdRIDVZDdXAcRLRtG1eN680xHdN5Y9FWvjm8K0V9OvHwR6vZuLua6no/F57QnZtP7c/9F48gLcXLd0/px/Hdc3hnfQMF2Wk8eOmIZsf75vBulFXWMW/DniOee/a6Pby/fDu/eH0Z9Y2BI+6vlJNaPUGISJaIZAfvA2cBy4A3gWvt3a4F3mjt2JKBpx2MpB7VqxNnDC2kX0GWY8dMT/VS02B1c81MjS5BpPu8PHTZSH7xjaE8fPkJ3HBSX4yBt5dsBaB35yzuPHcI4/vnA9ayqQ9fcQJn907hlZsn0KNTZrPjnTakC+k+D2/Zrz+cd5Zuw+sRNu2p5oW58TVuT5u7iT9/+HVcx1DJpS1KEIXALBFZDMwF3jHGvA/cB5wpIquBM+3HymFNvZgSOEH0zMvksWuLmnofOSHD52VfbSMNftPU7TUa4/vnc9PJ/fB4hBE9rKaytxZvA6B3XuYh+/ftnMUVQ9PolJV6yHNZaSmcd3xXXvyylJXb9/HA9JXc/dpSAoED7YHLtlRQUd3A9OXbOe/4rozrl8fDM1YftkfVkby7bDvPfLEh5ter5NPqbRDGmHXAiDDbdwOnt3Y8ySZYtZTIVUxuyEz1sne/NXNrRpQliIPlZaXSMy+DVTsq8QiHlBAicfd5Q/l41U4u/vvn7Lc/9DtlpvKTswezs7KO8/82i5wMH3urG/jG8V3plOnjsqmzeWVBKVeN6x1T3BU1DeytbmB3VV1Mr1fJJ5G6uapWEMwLniT7y6f7vOxxKEEATaWIrjkZpMawbkV+hzR+f9HxVDf4ufnU/lx+Yk/+NnMNX27Yw9byGgIGquv95GT4KB5cwJi+eQzvkcPjs9YTCBiMMezdX8/zczbxy9eXRVSy2FdjDfxbu3N/1PGq5KRTbSSZ9jDVhhsyQhOEAwPwRvTI5e0l2+idH33pIejs445h0S/PIifTx+6qOqZ9uZklpRX06JQBwFM3jKF3fmZTo/p3Tu7HrS8s5IyHPmbT7moaQ6qkPAK/mTzssOersBPEmrIqusUctUommiCSzIHZXJMsQdiN1OBQguhplSB658fXkJ6T6QOsaqu0FA/bymtI9Vp/m34FWXTJPjBe49xhx3DywM40+g1nDi2kc4c0ivp04q3F23jis/WcPrSwxa6wxpiQEkQV3RKvS79KQJogkkx7aKR2Q2hScKKKaVj3jnRMT+H47jlxHwusxZy65WawraKWdJ8Xr0fIz2o+kC7F6+GZG8ce8tqhXTsyY+UOHpm5psUEUV3vbypxrCmr4uSQBPHlhj18+vVObjltYEzVZeropQkiybSHqTbcEDr2wYkSRGZqCp/fdTqZDs0XBdA1J51tFTVkpHop6JAW8d8o3edl8ohu/G3mGvbsrycvpOfU+l37McY0/f4esUoQ9LMSQW2Dn9umLWJLeQ2LSiv451WjHO09pto3/bqQZDxJ2gYROvbBiRIEQIe0FEdLYl1zrBLEjn21FOZENhVI0FnHHUPAwIwVO5q27att4LJHv+DOV5ayz56ZdlBhNlvKa6jzW6WJp7/YwJbyGq4a14tZq3dy//urqKn3c8Ejn/Hz15ZS2xB7t9pE8n9vfcXnurxs1PSrQpLxJmkvptBSQ7QD5VpL15x0duyrJTPVG/U0KMd160i3nHQ++GoHlxT1BOBP01dRVllHaoqHimorQYzq3YmV2ysp2dzI7LeW8+KXmykeXMBvLzie+sYAz8/dhD9gWLS5nEWby1m8uZzXvj+xXVc91TX6eeKz9eysqmPCgM5tHU670n7/6iomniQdBxFaanBqGnGndc1NJ2Bg3a79FHaMrgQhIpx13DF88vVOVu+o5Lk5G3lm9kYyfF7KKuuaejCN6WPNbfXCynqem7OJ8f07c6/d++nW0wYSCBiemb2Rs48r5E+XjGD51n3tft3unZXWuI+lpeVtHEn7owkiybSHqTbckN6sBJGYBeduOVb3VmOgsGP0M71eUtQDr0c486FPuPu1ZZw0oDPfPbUf9Y0BNu+tAaxpTF74zjh+NT6dRb86k8euLaKnPRK8Z14ml57Yk1SvhzvPHcrkkd3olOnjjcVHnhIkkZXZCWLD7uqmkpSKTGL+pyjXBEsOkmwlCIcbqd3QNfdAqSHaEgTAcd1ymHXHaTw/ZyOdO6RxaVHPpvme1pRVApCT4aNXfiZ1m71hE+X/nX8ct542gK52svrG8K68PL+U/XWNZKW1/HHxwPSVlO6t4RffOJaC7LabxjycYAkCYOmWCk4aqNVMkdISRJLxeiTpSg/QvIopLUHr04MfyhBbggBrPMUtpw3k8jG98Hik6cN6TVkVIpCdfvjvhCleT7M4Jo/sTm1DgDcWtVyKWLhpL4/MXMsbi7Zy1kMf88Xa3YfsU1HTwB/eW8HtLy7CH2jdNWjKQhLEki1azRQNLUEkGZHka3+AA6WGDJ83YceAdExPISvVy/56f8wJ4mDBgXZryqrIjqHX1ehenRhyTDY/f20pX6zbzXUTejOqVydEhHveXM7G3fvZsa+Oguw0/nVNET95aTHXPjGXh68YyTnDurKrqo7HZ63n+TmbmrWDXD6mlyO/XyR2VtYhYlXhLS2taLXzHg0S86uUco3XI0nXgwkOlCCc6uLqBhGha6717f0YpxKE3Zaxt7qBjhm+qF/v8Qgvf28CP5jUn/9+tYOL//EF9769gl1VdTwzeyOfrN7FV9v28bOzBzOyZy4v3zyeYd07csvzC3l1QSkX/+NzHv14LeP65fH2rScxpk8e909fRVllLWAtA1teXd/UndYYwy6HJxPcWVlLflYaJ/TKZYkmiKhoCSLJeEWSvgSRyLrmpLN5TzUdM5z518xOSyEtxUNdY4CcGBIEWOM9fnr2EL5XPIC7Xl3Ks3M20iE9BX/A8NYtJ+E3hhE9rBHluZmpPHnDGC795xfc/uJislK9vHTzBEb3tlYPvuf84/jmXz9lzO9m0CnTmq3Wep2PWXecxuxtfq7/7X/57in9KMhOY8aKMn581iCK+rS8smDZvloKstNabFfbWVlHl+w0Rva05s9av2s/fTs7t9bI0SwJv0smN49HEraKxU3toQQBVi+jE3rlOtaJQESaShGxJoigDmkp/Oj0gdQ3BvjbR6sZckw2x/fIYWTP5vF2TPfx5PVjOO/4Y3jqhjFNyQHg2G4defOWk/jxmYM4Z1hXfnr2YL5zcl/Kqxv4cv0eFu9sJMUjPPqJtczrsi0VXDZ1dovrWCzctJdxf5jBz+31NBZs2sv2itpm+5RVWlVg54/oRopHeH7OxriuQzLREkSS8UjydXGF9lOC+N8zBzl+zIIOaWzeUxN3ggAY0KUDxYMLKFm1k2+NaHlO2GNy0vn7laPDPjesew7DQuawqm3w89QXG/lszS5W7glw3vFd+fboHmSleRlUmM2Ppi3inre+YmjXjoeUJB6esRoR4YW5m5mxoqypQfr47jncde4QJgzoTNm+OgYXZtOlYzpnHVfIS/NL+fFZgxN2PEwi0QSRZLweSbqZXOHAOIhETxBuCDZUd0yPP0EA3DJpAOt27ufCE7o7crx0n5fRvTrx2sItlNcZxvXLbzbp4F8uH8k3/zqLm5+dT7fcDBr8ht55mQzvmcPMVTv5yVmDKK9uYM76Pfz4rEFU1DTw1Ocb+Z/H5vDgJSPYVVXX1JvrqnG9eXfpdl5buIUr7Iby5+Zs5JGP1vDwFScwvEcuH60so7bBz7DuHRnQJduR33FNWSXZ6T7HOh+0Fk0QScYjyZkg2ksVkxuaqpgynUkQRX3y+ORnkxw5VtCE/vl8sc7qHjuuX/NSQna6j0f+ZxQ/fnExORk+fF4PS7dU8P7y7WSnp3DNhD6HJL9rxvfh3L98yuOz1tMYMHSxE8T4fvmM6JnLr95Yxr6aBrZV1PLk5xvweYXvPD2PnnmZTQ3ZPq9wxzlDuGZ8HwAa/AF83gO18v6ANYW6R4Qt5TX4vMLAwkMTysbd+zn/b5+R4fPy7E1jGdq1Y9hrsLOyjvW79jOgS4dmEy62JU0QScYjgjcJW57aSxWTGwo6ONMG4aYJA/J58EPITZOwDcjDuucw/X9PaXocCBjmbthDhs8btmSU7vNy5rGFTP1kHQBd7G/uIsLTN4zhe8/O5w/vrUQELh7Vg5tP7celj37B+l37+cvlIxnatSP3v7+K376zgr/MWE2Gx8+u6e9x7YQ+nD+iGw9+8DULNu2l+qCV/O46dwhTTumHiFhrcNQ28uMXF+P1CD6vh0sf/YLLinoyYUA+3XIzGFyY3bTvTU/PY/Fma5zGiX06ccnonpw3vCsdwgxQLNtXS+ZhBi46RRNEkvF6krMXk9cjpKZ4kroEEUs319YyvEcu2WkpDM03ETXQezzCuH75h91n0uAuTQkidHR3ToaPp24Yw6rtlfTKz2xKMO/+6GS8HmmqkvvXNaOZuaqM95dtZ93mbYwfVMi/P9vAvz/bQEG2NVK9V14mAWM4Jied95dt5w/vreTFeZvp3imTxZvLm8Z+PHTZCIp653Hv21/x1BcbeGzWesBaeOqOcwZTU+9n8eZyvlfcn6xUL68u3MLPXlnCPW8t55bTBnDdhD7MXb+HdJ+XVdsr+dP0VVw+picTXe6MlXAJQkTOAf4CeIHHjDH3tXFIR5URPXJafSRroshM9SbsTK5uCn44JnIJwuf18MKUcaxdtsCxYxb16UR2egqVtY1NVUyh5xt20GJPoSPIwSptnDakkNOGFFJSspfi4hM4dVABq7ZX8v1JAw65nucN68rYfvl8tGIHW8trOee4YxjQpQMDCztw6qACRISp1xRRUd3A2l1VLN9SwaOfrOPKx+aQn5VKr7xMbj9zED6vhx9MGsCCTXt59ON13P/+Kh784Otm/7cnDejM/4ztzcZlZY5dr3ASKkGIiBd4BDgTKAW+FJE3jTFftW1kR4/rJvbluoltHUXb+MOFxzOgS/KttTmoMJsOaSkMTPDffVj3HHatdq506/N6OHVQAW8v2ebY/FAXjerR4nMej3D1uN5cPa73YY+Rk+ljVK9OjOrViW+P7sndry3l1YVbePCSoU1tHCLC6N55PHp1J95eso0Fm/ZSPLgLHrGqiSf0z0dEcLvDbkIlCGAMsMYYsw5ARKYBkwFNECpu5x7fta1DaBM9OmWy7Ddnt3UYbeL7xQMY2rVjws7gm5Hq5cFLR3DbGYPolZ95yPMiwrdGdDtsl2I3iTGJU90gIt8GzjHG3GQ/vhoYa4y5JWSfKcAUgMLCwtHTpk2L6VxVVVV06JCY36gSNTaNKzqJGhckbmwaV3RijWvSpEnzjTFFR9zRGJMwN+ASrHaH4OOrgb+2tP/o0aNNrGbOnBnza92WqLFpXNFJ1LiMSdzYNK7oxBoXMM9E8JmcaB0eS4GeIY97AO17tRKllGqnEi1BfAkMFJG+IpIKXA682cYxKaVUUkqolhtjTKOI3AJMx+rm+oQxZnkbh6WUUkkpoRIEgDHmXeDdto5DKaWSXaJVMSmllEoQmiCUUkqFpQlCKaVUWAk1UC5aIrITYh5t3hnY5WA4TkrU2DSu6CRqXJC4sWlc0Yk1rt7GmIIj7dSuE0Q8RGSeiWQkYRtI1Ng0rugkalyQuLFpXNFxOy6tYlJKKRWWJgillFJhJXOCmNrWARxGosamcUUnUeOCxI1N44qOq3ElbRuEUkqpw0vmEoRSSqnD0AShlFIqrKRMECJyjoisEpE1InJnG8bRU0RmisgKEVkuIj+yt98jIltEZJF9O68NYtsgIkvt88+zt+WJyIcistr+2akN4hoccl0Wicg+EbmtLa6ZiDwhImUisixkW9hrJJaH7ffcEhEZ1cpxPSAiK+1zvyYiufb2PiJSE3Ld/tnKcbX4dxORu+zrtUpEXF0Sr4XY/hMS1wYRWWRvb81r1tJnROu8zyJZNOJoumHNErsW6AekAouBY9solq7AKPt+NvA1cCxwD/CTNr5OG4DOB227H7jTvn8n8McE+FtuB3q3xTUDTgFGAcuOdI2A84D3AAHGAXNaOa6zgBT7/h9D4uoTul8bXK+wfzf7/2AxkAb0tf9nva0Z20HPPwj8qg2uWUufEa3yPkvGEkTTutfGmHoguO51qzPGbDPGLLDvVwIrgO5tEUuEJgNP2fefAi5ow1gATgfWGmPcXrs9LGPMJ8Cegza3dI0mA08by2wgV0RcWSQ7XFzGmA+MMY32w9lYi3G1qhauV0smA9OMMXXGmPXAGqz/3VaPTUQEuBR4wa3zt+QwnxGt8j5LxgTRHdgc8riUBPhQFpE+wAnAHHvTLXYR8Ym2qMoBDPCBiMwXax1wgEJjzDaw3rhAlzaIK9TlNP+nbetrBi1fo0R6392A9S0zqK+ILBSRj0Xk5DaIJ9zfLZGu18nADmPM6pBtrX7NDvqMaJX3WTImCAmzrU37+opIB+AV4DZjzD7gH0B/YCSwDat429omGmNGAecCPxCRU9oghhaJteLg+cBL9qZEuGaHkxDvOxG5G2gEnrM3bQN6GWNOAG4HnheRjq0YUkt/t4S4XrYraP5FpNWvWZjPiBZ3DbMt5uuWjAkioda9FhEf1h/+ObBfbnsAAATqSURBVGPMqwDGmB3GGL8xJgD8CxeL1i0xxmy1f5YBr9kx7AgWV+2fZa0dV4hzgQXGmB2QGNfM1tI1avP3nYhcC3wTuNLYFdZ2Fc5u+/58rLr+Qa0V02H+bm1+vQBEJAW4CPhPcFtrX7NwnxG00vssGRNEwqx7bddtPg6sMMb8OWR7aJ3hhcCyg1/rclxZIpIdvI/VwLkM6zpda+92LfBGa8Z1kGbf6tr6moVo6Rq9CVxj9zIZB1QEqwhag4icA9wBnG+MqQ7ZXiAiXvt+P2AgsK4V42rp7/YmcLmIpIlIXzuuua0VV4gzgJXGmNLghta8Zi19RtBa77PWaIlPtBtWS//XWJn/7jaM4ySs4t8SYJF9Ow94Blhqb38T6NrKcfXD6kGyGFgevEZAPjADWG3/zGuj65YJ7AZyQra1+jXDSlDbgAasb243tnSNsIr+j9jvuaVAUSvHtQarbjr4Pvunve/F9t94MbAA+FYrx9Xi3w24275eq4BzW/tvaW9/Erj5oH1b85q19BnRKu8znWpDKaVUWMlYxaSUUioCmiCUUkqFpQlCKaVUWJoglFJKhaUJQimlVFiaINRRQ0TOlyPMzisi3UTkZfv+dSLytyjP8fMI9nlSRL4dzXGdJCIlIuLaQvYqeWiCUEcNY8ybxpj7jrDPVmNMPB/eR0wQ7Zk9clgpQBOEagfs+fdXishjIrJMRJ4TkTNE5DN7Pvwx9n5NJQL7W/zDIvK5iKwLfqO3jxU6yrqniLwv1poDvw455+v2RIXLg5MVish9QIZYawA8Z2+7xp5obrGIPBNy3FMOPneY32mFiPzLPscHIpJhP9dUAhCRziKyIeT3e11E3hKR9SJyi4jcbk8aN1tE8kJOcZV9/mUh1yfLnhDvS/s1k0OO+5KIvAV8EM/fSh1l3BydqDe9OXHDmn+/ETge60vNfOAJrFGjk4HX7f2uA/5m338SayI/D9b8+WtCjrUsZP9tWKNSM7CmeSiynwuOTA1uz7cfV4XEdRzWKN/OB70m7Llb+J1G2o9fBK6y75eExNEZ2BAS7xqsdQEKgArsUb7AQ1gTuQVf/y/7/ikhv+/vQ86RizWbQJZ93FLaaGS83hL3piUI1V6sN8YsNdakbsuBGcYYgzWdQJ8WXvO6MSZgjPkKKGxhnw+NMbuNMTXAq1hTGwD8UEQWY62d0BNrvp2DnQa8bIzZBWCMCV1PIJJzrzfGLLLvzz/M7xFqpjGm0hizEytBvGVvP/g6vGDH9AnQUawV5M4C7hRrZbQSIB3oZe//4UHxK4XWN6r2oi7kfiDkcYCW38ehrwk3DTIcOhWyEZFirEnaxhtjqkWkBOvD9GAS5vXRnDt0Hz9WaQWskkXwy9vB5430Ohzye9lxXGyMWRX6hIiMBfa3EKNKYlqCUMnuTLHW983AWpXrMyAH2GsnhyFYSzcGNdjTL4M1SdqlIpIP1jrBDsW0ARht34+1Qf0yABE5CWtGzwpgOnCrPUMoInJCnHGqo5wmCJXsZmHNKLoIeMUYMw94H0gRkSXAvVjVTEFTgSUi8pwxZjnwO+BjuzrqzzjjT8D3RORzrDaIWOy1X/9PrFlTwfpdfFjxL7MfK9Uinc1VKaVUWFqCUEopFZYmCKWUUmFpglBKKRWWJgillFJhaYJQSikVliYIpZRSYWmCUEopFdb/B95KWLXfHgl6AAAAAElFTkSuQmCC\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f13c17f23c8>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "train\n",
      "Iteration0: with minibatch training loss = 2.53 and accuracy of 0.1\n",
      "Epoch 1, Overall loss = 2.4 and accuracy of 0.126\n",
      "Epoch 2, Overall loss = 2.21 and accuracy of 0.184\n",
      "Epoch 3, Overall loss = 2.06 and accuracy of 0.232\n",
      "Epoch 4, Overall loss = 1.89 and accuracy of 0.279\n",
      "Epoch 5, Overall loss = 1.76 and accuracy of 0.32\n",
      "Epoch 6, Overall loss = 1.64 and accuracy of 0.355\n",
      "Epoch 7, Overall loss = 1.54 and accuracy of 0.384\n",
      "Iteration100: with minibatch training loss = 1.45 and accuracy of 0.41\n",
      "Epoch 8, Overall loss = 1.46 and accuracy of 0.411\n",
      "Epoch 9, Overall loss = 1.4 and accuracy of 0.433\n",
      "Epoch 10, Overall loss = 1.35 and accuracy of 0.456\n",
      "Epoch 11, Overall loss = 1.31 and accuracy of 0.465\n",
      "Epoch 12, Overall loss = 1.25 and accuracy of 0.495\n",
      "Epoch 13, Overall loss = 1.19 and accuracy of 0.522\n",
      "Epoch 14, Overall loss = 1.13 and accuracy of 0.536\n",
      "Iteration200: with minibatch training loss = 1.07 and accuracy of 0.59\n",
      "Epoch 15, Overall loss = 1.08 and accuracy of 0.561\n",
      "Epoch 16, Overall loss = 1.05 and accuracy of 0.564\n",
      "Epoch 17, Overall loss = 1.01 and accuracy of 0.573\n",
      "Epoch 18, Overall loss = 0.987 and accuracy of 0.582\n",
      "Epoch 19, Overall loss = 0.948 and accuracy of 0.6\n",
      "Epoch 20, Overall loss = 0.931 and accuracy of 0.611\n",
      "Epoch 21, Overall loss = 0.917 and accuracy of 0.61\n",
      "Iteration300: with minibatch training loss = 0.885 and accuracy of 0.61\n",
      "Epoch 22, Overall loss = 0.915 and accuracy of 0.595\n",
      "Epoch 23, Overall loss = 0.88 and accuracy of 0.627\n",
      "Epoch 24, Overall loss = 0.856 and accuracy of 0.642\n",
      "Epoch 25, Overall loss = 0.863 and accuracy of 0.631\n",
      "Epoch 26, Overall loss = 0.844 and accuracy of 0.64\n",
      "Epoch 27, Overall loss = 0.843 and accuracy of 0.639\n",
      "Epoch 28, Overall loss = 0.828 and accuracy of 0.648\n",
      "Iteration400: with minibatch training loss = 0.845 and accuracy of 0.66\n",
      "Epoch 29, Overall loss = 0.819 and accuracy of 0.657\n",
      "Epoch 30, Overall loss = 0.788 and accuracy of 0.661\n",
      "Epoch 31, Overall loss = 0.783 and accuracy of 0.658\n",
      "Epoch 32, Overall loss = 0.773 and accuracy of 0.67\n",
      "Epoch 33, Overall loss = 0.744 and accuracy of 0.692\n",
      "Epoch 34, Overall loss = 0.732 and accuracy of 0.698\n",
      "Epoch 35, Overall loss = 0.734 and accuracy of 0.695\n",
      "Iteration500: with minibatch training loss = 0.713 and accuracy of 0.71\n",
      "Epoch 36, Overall loss = 0.713 and accuracy of 0.705\n",
      "Epoch 37, Overall loss = 0.71 and accuracy of 0.707\n",
      "Epoch 38, Overall loss = 0.681 and accuracy of 0.715\n",
      "Epoch 39, Overall loss = 0.67 and accuracy of 0.722\n",
      "Epoch 40, Overall loss = 0.655 and accuracy of 0.728\n",
      "Epoch 41, Overall loss = 0.654 and accuracy of 0.728\n",
      "Epoch 42, Overall loss = 0.631 and accuracy of 0.744\n",
      "Iteration600: with minibatch training loss = 0.641 and accuracy of 0.73\n",
      "Epoch 43, Overall loss = 0.615 and accuracy of 0.745\n",
      "Epoch 44, Overall loss = 0.605 and accuracy of 0.751\n",
      "Epoch 45, Overall loss = 0.598 and accuracy of 0.752\n",
      "Epoch 46, Overall loss = 0.579 and accuracy of 0.759\n",
      "Epoch 47, Overall loss = 0.576 and accuracy of 0.764\n",
      "Epoch 48, Overall loss = 0.578 and accuracy of 0.765\n",
      "Epoch 49, Overall loss = 0.556 and accuracy of 0.771\n",
      "Epoch 50, Overall loss = 0.538 and accuracy of 0.784\n",
      "Iteration700: with minibatch training loss = 0.556 and accuracy of 0.77\n",
      "Epoch 51, Overall loss = 0.536 and accuracy of 0.779\n",
      "Epoch 52, Overall loss = 0.521 and accuracy of 0.781\n",
      "Epoch 53, Overall loss = 0.511 and accuracy of 0.789\n",
      "Epoch 54, Overall loss = 0.518 and accuracy of 0.784\n",
      "Epoch 55, Overall loss = 0.517 and accuracy of 0.783\n",
      "Epoch 56, Overall loss = 0.484 and accuracy of 0.792\n",
      "Epoch 57, Overall loss = 0.485 and accuracy of 0.794\n",
      "Iteration800: with minibatch training loss = 0.441 and accuracy of 0.81\n",
      "Epoch 58, Overall loss = 0.485 and accuracy of 0.796\n",
      "Epoch 59, Overall loss = 0.474 and accuracy of 0.797\n",
      "Epoch 60, Overall loss = 0.493 and accuracy of 0.8\n",
      "Epoch 61, Overall loss = 0.492 and accuracy of 0.792\n",
      "Epoch 62, Overall loss = 0.472 and accuracy of 0.799\n",
      "Epoch 63, Overall loss = 0.464 and accuracy of 0.807\n",
      "Epoch 64, Overall loss = 0.454 and accuracy of 0.808\n",
      "Iteration900: with minibatch training loss = 0.413 and accuracy of 0.83\n",
      "Epoch 65, Overall loss = 0.479 and accuracy of 0.796\n",
      "Epoch 66, Overall loss = 0.448 and accuracy of 0.804\n",
      "Epoch 67, Overall loss = 0.45 and accuracy of 0.809\n",
      "Epoch 68, Overall loss = 0.439 and accuracy of 0.809\n",
      "Epoch 69, Overall loss = 0.434 and accuracy of 0.811\n",
      "Epoch 70, Overall loss = 0.436 and accuracy of 0.807\n",
      "Epoch 71, Overall loss = 0.438 and accuracy of 0.813\n",
      "Iteration1000: with minibatch training loss = 0.424 and accuracy of 0.81\n",
      "Epoch 72, Overall loss = 0.426 and accuracy of 0.812\n",
      "Epoch 73, Overall loss = 0.435 and accuracy of 0.815\n",
      "Epoch 74, Overall loss = 0.421 and accuracy of 0.815\n",
      "Epoch 75, Overall loss = 0.429 and accuracy of 0.813\n",
      "Epoch 76, Overall loss = 0.43 and accuracy of 0.816\n",
      "Epoch 77, Overall loss = 0.415 and accuracy of 0.816\n",
      "Epoch 78, Overall loss = 0.417 and accuracy of 0.812\n",
      "Iteration1100: with minibatch training loss = 0.466 and accuracy of 0.79\n",
      "Epoch 79, Overall loss = 0.427 and accuracy of 0.811\n",
      "Epoch 80, Overall loss = 0.407 and accuracy of 0.819\n",
      "Epoch 81, Overall loss = 0.414 and accuracy of 0.811\n",
      "Epoch 82, Overall loss = 0.409 and accuracy of 0.82\n",
      "Epoch 83, Overall loss = 0.41 and accuracy of 0.811\n",
      "Epoch 84, Overall loss = 0.418 and accuracy of 0.815\n",
      "Epoch 85, Overall loss = 0.395 and accuracy of 0.818\n",
      "Iteration1200: with minibatch training loss = 0.416 and accuracy of 0.8\n",
      "Epoch 86, Overall loss = 0.418 and accuracy of 0.814\n",
      "Epoch 87, Overall loss = 0.411 and accuracy of 0.811\n",
      "Epoch 88, Overall loss = 0.401 and accuracy of 0.82\n",
      "Epoch 89, Overall loss = 0.399 and accuracy of 0.819\n",
      "Epoch 90, Overall loss = 0.422 and accuracy of 0.813\n",
      "Epoch 91, Overall loss = 0.395 and accuracy of 0.827\n",
      "Epoch 92, Overall loss = 0.392 and accuracy of 0.825\n",
      "Iteration1300: with minibatch training loss = 0.392 and accuracy of 0.82\n",
      "Epoch 93, Overall loss = 0.392 and accuracy of 0.82\n",
      "Epoch 94, Overall loss = 0.392 and accuracy of 0.818\n",
      "Epoch 95, Overall loss = 0.395 and accuracy of 0.823\n",
      "Epoch 96, Overall loss = 0.391 and accuracy of 0.821\n",
      "Epoch 97, Overall loss = 0.388 and accuracy of 0.821\n",
      "Epoch 98, Overall loss = 0.387 and accuracy of 0.824\n",
      "Epoch 99, Overall loss = 0.392 and accuracy of 0.814\n",
      "Epoch 100, Overall loss = 0.371 and accuracy of 0.826\n",
      "Iteration1400: with minibatch training loss = 0.368 and accuracy of 0.83\n",
      "Epoch 101, Overall loss = 0.379 and accuracy of 0.818\n",
      "Epoch 102, Overall loss = 0.382 and accuracy of 0.823\n",
      "Epoch 103, Overall loss = 0.377 and accuracy of 0.825\n",
      "Epoch 104, Overall loss = 0.377 and accuracy of 0.827\n",
      "Epoch 105, Overall loss = 0.37 and accuracy of 0.823\n",
      "Epoch 106, Overall loss = 0.373 and accuracy of 0.819\n",
      "Epoch 107, Overall loss = 0.367 and accuracy of 0.823\n",
      "Iteration1500: with minibatch training loss = 0.369 and accuracy of 0.84\n",
      "Epoch 108, Overall loss = 0.369 and accuracy of 0.824\n",
      "Epoch 109, Overall loss = 0.376 and accuracy of 0.823\n",
      "Epoch 110, Overall loss = 0.38 and accuracy of 0.823\n",
      "Epoch 111, Overall loss = 0.375 and accuracy of 0.824\n",
      "Epoch 112, Overall loss = 0.379 and accuracy of 0.827\n",
      "Epoch 113, Overall loss = 0.37 and accuracy of 0.826\n",
      "Epoch 114, Overall loss = 0.367 and accuracy of 0.818\n",
      "Iteration1600: with minibatch training loss = 0.336 and accuracy of 0.83\n",
      "Epoch 115, Overall loss = 0.373 and accuracy of 0.828\n",
      "Epoch 116, Overall loss = 0.364 and accuracy of 0.826\n",
      "Epoch 117, Overall loss = 0.359 and accuracy of 0.823\n",
      "Epoch 118, Overall loss = 0.353 and accuracy of 0.827\n",
      "Epoch 119, Overall loss = 0.376 and accuracy of 0.821\n",
      "Epoch 120, Overall loss = 0.365 and accuracy of 0.828\n",
      "Epoch 121, Overall loss = 0.376 and accuracy of 0.823\n",
      "Iteration1700: with minibatch training loss = 0.336 and accuracy of 0.82\n",
      "Epoch 122, Overall loss = 0.356 and accuracy of 0.829\n",
      "Epoch 123, Overall loss = 0.362 and accuracy of 0.826\n",
      "Epoch 124, Overall loss = 0.37 and accuracy of 0.822\n",
      "Epoch 125, Overall loss = 0.355 and accuracy of 0.824\n",
      "Epoch 126, Overall loss = 0.364 and accuracy of 0.829\n",
      "Epoch 127, Overall loss = 0.359 and accuracy of 0.83\n",
      "Epoch 128, Overall loss = 0.35 and accuracy of 0.829\n",
      "Iteration1800: with minibatch training loss = 0.389 and accuracy of 0.82\n",
      "Epoch 129, Overall loss = 0.368 and accuracy of 0.826\n",
      "Epoch 130, Overall loss = 0.345 and accuracy of 0.827\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 131, Overall loss = 0.356 and accuracy of 0.829\n",
      "Epoch 132, Overall loss = 0.348 and accuracy of 0.831\n",
      "Epoch 133, Overall loss = 0.345 and accuracy of 0.827\n",
      "Epoch 134, Overall loss = 0.349 and accuracy of 0.826\n",
      "Epoch 135, Overall loss = 0.344 and accuracy of 0.828\n",
      "Iteration1900: with minibatch training loss = 0.355 and accuracy of 0.82\n",
      "Epoch 136, Overall loss = 0.353 and accuracy of 0.827\n",
      "Epoch 137, Overall loss = 0.375 and accuracy of 0.83\n",
      "Epoch 138, Overall loss = 0.36 and accuracy of 0.823\n",
      "Epoch 139, Overall loss = 0.35 and accuracy of 0.829\n",
      "Epoch 140, Overall loss = 0.361 and accuracy of 0.827\n",
      "Epoch 141, Overall loss = 0.339 and accuracy of 0.834\n",
      "Epoch 142, Overall loss = 0.342 and accuracy of 0.829\n",
      "Iteration2000: with minibatch training loss = 0.325 and accuracy of 0.82\n",
      "Epoch 143, Overall loss = 0.329 and accuracy of 0.832\n",
      "Epoch 144, Overall loss = 0.339 and accuracy of 0.829\n",
      "Epoch 145, Overall loss = 0.344 and accuracy of 0.827\n",
      "Epoch 146, Overall loss = 0.335 and accuracy of 0.835\n",
      "Epoch 147, Overall loss = 0.331 and accuracy of 0.834\n",
      "Epoch 148, Overall loss = 0.332 and accuracy of 0.835\n",
      "Epoch 149, Overall loss = 0.333 and accuracy of 0.837\n",
      "Epoch 150, Overall loss = 0.342 and accuracy of 0.834\n",
      "Iteration2100: with minibatch training loss = 0.344 and accuracy of 0.84\n",
      "Epoch 151, Overall loss = 0.331 and accuracy of 0.835\n",
      "Epoch 152, Overall loss = 0.34 and accuracy of 0.835\n",
      "Epoch 153, Overall loss = 0.336 and accuracy of 0.83\n",
      "Epoch 154, Overall loss = 0.323 and accuracy of 0.839\n",
      "Epoch 155, Overall loss = 0.333 and accuracy of 0.836\n",
      "Epoch 156, Overall loss = 0.322 and accuracy of 0.838\n",
      "Epoch 157, Overall loss = 0.337 and accuracy of 0.834\n",
      "Iteration2200: with minibatch training loss = 0.285 and accuracy of 0.84\n",
      "Epoch 158, Overall loss = 0.333 and accuracy of 0.825\n",
      "Epoch 159, Overall loss = 0.328 and accuracy of 0.833\n",
      "Epoch 160, Overall loss = 0.325 and accuracy of 0.833\n",
      "Epoch 161, Overall loss = 0.321 and accuracy of 0.834\n",
      "Epoch 162, Overall loss = 0.327 and accuracy of 0.837\n",
      "Epoch 163, Overall loss = 0.317 and accuracy of 0.844\n",
      "Epoch 164, Overall loss = 0.317 and accuracy of 0.835\n",
      "Iteration2300: with minibatch training loss = 0.362 and accuracy of 0.85\n",
      "Epoch 165, Overall loss = 0.324 and accuracy of 0.838\n",
      "Epoch 166, Overall loss = 0.313 and accuracy of 0.837\n",
      "Epoch 167, Overall loss = 0.324 and accuracy of 0.838\n",
      "Epoch 168, Overall loss = 0.32 and accuracy of 0.834\n",
      "Epoch 169, Overall loss = 0.323 and accuracy of 0.838\n",
      "Epoch 170, Overall loss = 0.317 and accuracy of 0.843\n",
      "Epoch 171, Overall loss = 0.322 and accuracy of 0.835\n",
      "Iteration2400: with minibatch training loss = 0.305 and accuracy of 0.83\n",
      "Epoch 172, Overall loss = 0.317 and accuracy of 0.838\n",
      "Epoch 173, Overall loss = 0.317 and accuracy of 0.837\n",
      "Epoch 174, Overall loss = 0.32 and accuracy of 0.84\n",
      "Epoch 175, Overall loss = 0.31 and accuracy of 0.84\n",
      "Epoch 176, Overall loss = 0.302 and accuracy of 0.847\n",
      "Epoch 177, Overall loss = 0.314 and accuracy of 0.837\n",
      "Epoch 178, Overall loss = 0.323 and accuracy of 0.838\n",
      "Iteration2500: with minibatch training loss = 0.321 and accuracy of 0.85\n",
      "Epoch 179, Overall loss = 0.327 and accuracy of 0.836\n",
      "Epoch 180, Overall loss = 0.31 and accuracy of 0.839\n",
      "Epoch 181, Overall loss = 0.316 and accuracy of 0.84\n",
      "Epoch 182, Overall loss = 0.317 and accuracy of 0.84\n",
      "Epoch 183, Overall loss = 0.304 and accuracy of 0.842\n",
      "Epoch 184, Overall loss = 0.314 and accuracy of 0.839\n",
      "Epoch 185, Overall loss = 0.303 and accuracy of 0.84\n",
      "Iteration2600: with minibatch training loss = 0.282 and accuracy of 0.86\n",
      "Epoch 186, Overall loss = 0.31 and accuracy of 0.847\n",
      "Epoch 187, Overall loss = 0.306 and accuracy of 0.846\n",
      "Epoch 188, Overall loss = 0.303 and accuracy of 0.842\n",
      "Epoch 189, Overall loss = 0.312 and accuracy of 0.845\n",
      "Epoch 190, Overall loss = 0.303 and accuracy of 0.845\n",
      "Epoch 191, Overall loss = 0.311 and accuracy of 0.84\n",
      "Epoch 192, Overall loss = 0.305 and accuracy of 0.838\n",
      "Iteration2700: with minibatch training loss = 0.301 and accuracy of 0.81\n",
      "Epoch 193, Overall loss = 0.299 and accuracy of 0.84\n",
      "Epoch 194, Overall loss = 0.313 and accuracy of 0.847\n",
      "Epoch 195, Overall loss = 0.317 and accuracy of 0.842\n",
      "Epoch 196, Overall loss = 0.31 and accuracy of 0.846\n",
      "Epoch 197, Overall loss = 0.304 and accuracy of 0.846\n",
      "Epoch 198, Overall loss = 0.306 and accuracy of 0.836\n",
      "Epoch 199, Overall loss = 0.296 and accuracy of 0.848\n",
      "Epoch 200, Overall loss = 0.299 and accuracy of 0.844\n",
      "Iteration2800: with minibatch training loss = 0.304 and accuracy of 0.86\n",
      "Epoch 201, Overall loss = 0.3 and accuracy of 0.843\n",
      "Epoch 202, Overall loss = 0.302 and accuracy of 0.839\n",
      "Epoch 203, Overall loss = 0.306 and accuracy of 0.848\n",
      "Epoch 204, Overall loss = 0.306 and accuracy of 0.841\n",
      "Epoch 205, Overall loss = 0.309 and accuracy of 0.845\n",
      "Epoch 206, Overall loss = 0.297 and accuracy of 0.845\n",
      "Epoch 207, Overall loss = 0.307 and accuracy of 0.841\n",
      "Iteration2900: with minibatch training loss = 0.283 and accuracy of 0.85\n",
      "Epoch 208, Overall loss = 0.304 and accuracy of 0.845\n",
      "Epoch 209, Overall loss = 0.314 and accuracy of 0.837\n",
      "Epoch 210, Overall loss = 0.308 and accuracy of 0.84\n",
      "Epoch 211, Overall loss = 0.295 and accuracy of 0.848\n",
      "Epoch 212, Overall loss = 0.305 and accuracy of 0.842\n",
      "Epoch 213, Overall loss = 0.304 and accuracy of 0.844\n",
      "Epoch 214, Overall loss = 0.302 and accuracy of 0.837\n",
      "Iteration3000: with minibatch training loss = 0.273 and accuracy of 0.86\n",
      "Epoch 215, Overall loss = 0.295 and accuracy of 0.846\n",
      "Epoch 216, Overall loss = 0.299 and accuracy of 0.846\n",
      "Epoch 217, Overall loss = 0.291 and accuracy of 0.843\n",
      "Epoch 218, Overall loss = 0.292 and accuracy of 0.849\n",
      "Epoch 219, Overall loss = 0.3 and accuracy of 0.844\n",
      "Epoch 220, Overall loss = 0.301 and accuracy of 0.848\n",
      "Epoch 221, Overall loss = 0.306 and accuracy of 0.85\n",
      "Iteration3100: with minibatch training loss = 0.342 and accuracy of 0.85\n",
      "Epoch 222, Overall loss = 0.306 and accuracy of 0.847\n",
      "Epoch 223, Overall loss = 0.292 and accuracy of 0.847\n",
      "Epoch 224, Overall loss = 0.302 and accuracy of 0.84\n",
      "Epoch 225, Overall loss = 0.304 and accuracy of 0.847\n",
      "Epoch 226, Overall loss = 0.291 and accuracy of 0.842\n",
      "Epoch 227, Overall loss = 0.292 and accuracy of 0.845\n",
      "Epoch 228, Overall loss = 0.292 and accuracy of 0.847\n",
      "Iteration3200: with minibatch training loss = 0.293 and accuracy of 0.85\n",
      "Epoch 229, Overall loss = 0.287 and accuracy of 0.848\n",
      "Epoch 230, Overall loss = 0.295 and accuracy of 0.851\n",
      "Epoch 231, Overall loss = 0.295 and accuracy of 0.845\n",
      "Epoch 232, Overall loss = 0.292 and accuracy of 0.845\n",
      "Epoch 233, Overall loss = 0.287 and accuracy of 0.844\n",
      "Epoch 234, Overall loss = 0.303 and accuracy of 0.848\n",
      "Epoch 235, Overall loss = 0.311 and accuracy of 0.839\n",
      "Iteration3300: with minibatch training loss = 0.309 and accuracy of 0.86\n",
      "Epoch 236, Overall loss = 0.328 and accuracy of 0.837\n",
      "Epoch 237, Overall loss = 0.3 and accuracy of 0.848\n",
      "Epoch 238, Overall loss = 0.3 and accuracy of 0.843\n",
      "Epoch 239, Overall loss = 0.29 and accuracy of 0.844\n",
      "Epoch 240, Overall loss = 0.301 and accuracy of 0.844\n",
      "Epoch 241, Overall loss = 0.295 and accuracy of 0.841\n",
      "Epoch 242, Overall loss = 0.297 and accuracy of 0.84\n",
      "Iteration3400: with minibatch training loss = 0.301 and accuracy of 0.82\n",
      "Epoch 243, Overall loss = 0.282 and accuracy of 0.845\n",
      "Epoch 244, Overall loss = 0.284 and accuracy of 0.849\n",
      "Epoch 245, Overall loss = 0.284 and accuracy of 0.842\n",
      "Epoch 246, Overall loss = 0.28 and accuracy of 0.848\n",
      "Epoch 247, Overall loss = 0.284 and accuracy of 0.846\n",
      "Epoch 248, Overall loss = 0.292 and accuracy of 0.844\n",
      "Epoch 249, Overall loss = 0.293 and accuracy of 0.843\n",
      "Epoch 250, Overall loss = 0.289 and accuracy of 0.844\n",
      "Iteration3500: with minibatch training loss = 0.263 and accuracy of 0.86\n",
      "Epoch 251, Overall loss = 0.291 and accuracy of 0.847\n",
      "Epoch 252, Overall loss = 0.293 and accuracy of 0.845\n",
      "Epoch 253, Overall loss = 0.3 and accuracy of 0.846\n",
      "Epoch 254, Overall loss = 0.288 and accuracy of 0.849\n",
      "Epoch 255, Overall loss = 0.289 and accuracy of 0.842\n",
      "Epoch 256, Overall loss = 0.284 and accuracy of 0.848\n",
      "Epoch 257, Overall loss = 0.302 and accuracy of 0.851\n",
      "Iteration3600: with minibatch training loss = 0.279 and accuracy of 0.87\n",
      "Epoch 258, Overall loss = 0.286 and accuracy of 0.849\n",
      "Epoch 259, Overall loss = 0.283 and accuracy of 0.852\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 260, Overall loss = 0.293 and accuracy of 0.85\n",
      "Epoch 261, Overall loss = 0.282 and accuracy of 0.848\n",
      "Epoch 262, Overall loss = 0.281 and accuracy of 0.844\n",
      "Epoch 263, Overall loss = 0.29 and accuracy of 0.844\n",
      "Epoch 264, Overall loss = 0.288 and accuracy of 0.845\n",
      "Iteration3700: with minibatch training loss = 0.283 and accuracy of 0.86\n",
      "Epoch 265, Overall loss = 0.279 and accuracy of 0.849\n",
      "Epoch 266, Overall loss = 0.287 and accuracy of 0.844\n",
      "Epoch 267, Overall loss = 0.277 and accuracy of 0.852\n",
      "Epoch 268, Overall loss = 0.286 and accuracy of 0.846\n",
      "Epoch 269, Overall loss = 0.3 and accuracy of 0.845\n",
      "Epoch 270, Overall loss = 0.273 and accuracy of 0.853\n",
      "Epoch 271, Overall loss = 0.287 and accuracy of 0.848\n",
      "Iteration3800: with minibatch training loss = 0.306 and accuracy of 0.86\n",
      "Epoch 272, Overall loss = 0.283 and accuracy of 0.853\n",
      "Epoch 273, Overall loss = 0.283 and accuracy of 0.855\n",
      "Epoch 274, Overall loss = 0.283 and accuracy of 0.854\n",
      "Epoch 275, Overall loss = 0.28 and accuracy of 0.847\n",
      "Epoch 276, Overall loss = 0.279 and accuracy of 0.854\n",
      "Epoch 277, Overall loss = 0.28 and accuracy of 0.848\n",
      "Epoch 278, Overall loss = 0.275 and accuracy of 0.851\n",
      "Iteration3900: with minibatch training loss = 0.287 and accuracy of 0.84\n",
      "Epoch 279, Overall loss = 0.271 and accuracy of 0.853\n",
      "Epoch 280, Overall loss = 0.267 and accuracy of 0.856\n",
      "Epoch 281, Overall loss = 0.273 and accuracy of 0.852\n",
      "Epoch 282, Overall loss = 0.268 and accuracy of 0.848\n",
      "Epoch 283, Overall loss = 0.274 and accuracy of 0.849\n",
      "Epoch 284, Overall loss = 0.275 and accuracy of 0.859\n",
      "Epoch 285, Overall loss = 0.277 and accuracy of 0.851\n",
      "Iteration4000: with minibatch training loss = 0.283 and accuracy of 0.84\n",
      "Epoch 286, Overall loss = 0.265 and accuracy of 0.854\n",
      "Epoch 287, Overall loss = 0.266 and accuracy of 0.857\n",
      "Epoch 288, Overall loss = 0.258 and accuracy of 0.859\n",
      "Epoch 289, Overall loss = 0.262 and accuracy of 0.857\n",
      "Epoch 290, Overall loss = 0.263 and accuracy of 0.858\n",
      "Epoch 291, Overall loss = 0.248 and accuracy of 0.856\n",
      "Epoch 292, Overall loss = 0.264 and accuracy of 0.86\n",
      "Iteration4100: with minibatch training loss = 0.25 and accuracy of 0.87\n",
      "Epoch 293, Overall loss = 0.261 and accuracy of 0.859\n",
      "Epoch 294, Overall loss = 0.261 and accuracy of 0.853\n",
      "Epoch 295, Overall loss = 0.273 and accuracy of 0.848\n",
      "Epoch 296, Overall loss = 0.25 and accuracy of 0.86\n",
      "Epoch 297, Overall loss = 0.265 and accuracy of 0.854\n",
      "Epoch 298, Overall loss = 0.26 and accuracy of 0.857\n",
      "Epoch 299, Overall loss = 0.274 and accuracy of 0.854\n",
      "Epoch 300, Overall loss = 0.255 and accuracy of 0.858\n",
      "Iteration4200: with minibatch training loss = 0.305 and accuracy of 0.86\n",
      "Epoch 301, Overall loss = 0.256 and accuracy of 0.864\n",
      "Epoch 302, Overall loss = 0.263 and accuracy of 0.86\n",
      "Epoch 303, Overall loss = 0.258 and accuracy of 0.863\n",
      "Epoch 304, Overall loss = 0.255 and accuracy of 0.858\n",
      "Epoch 305, Overall loss = 0.241 and accuracy of 0.866\n",
      "Epoch 306, Overall loss = 0.263 and accuracy of 0.853\n",
      "Epoch 307, Overall loss = 0.255 and accuracy of 0.87\n",
      "Iteration4300: with minibatch training loss = 0.242 and accuracy of 0.85\n",
      "Epoch 308, Overall loss = 0.27 and accuracy of 0.859\n",
      "Epoch 309, Overall loss = 0.277 and accuracy of 0.86\n",
      "Epoch 310, Overall loss = 0.257 and accuracy of 0.867\n",
      "Epoch 311, Overall loss = 0.259 and accuracy of 0.86\n",
      "Epoch 312, Overall loss = 0.249 and accuracy of 0.864\n",
      "Epoch 313, Overall loss = 0.243 and accuracy of 0.863\n",
      "Epoch 314, Overall loss = 0.249 and accuracy of 0.863\n",
      "Iteration4400: with minibatch training loss = 0.238 and accuracy of 0.87\n",
      "Epoch 315, Overall loss = 0.246 and accuracy of 0.862\n",
      "Epoch 316, Overall loss = 0.246 and accuracy of 0.866\n",
      "Epoch 317, Overall loss = 0.252 and accuracy of 0.86\n",
      "Epoch 318, Overall loss = 0.247 and accuracy of 0.866\n",
      "Epoch 319, Overall loss = 0.239 and accuracy of 0.869\n",
      "Epoch 320, Overall loss = 0.236 and accuracy of 0.87\n",
      "Epoch 321, Overall loss = 0.246 and accuracy of 0.866\n",
      "Iteration4500: with minibatch training loss = 0.222 and accuracy of 0.84\n",
      "Epoch 322, Overall loss = 0.235 and accuracy of 0.865\n",
      "Epoch 323, Overall loss = 0.24 and accuracy of 0.871\n",
      "Epoch 324, Overall loss = 0.239 and accuracy of 0.872\n",
      "Epoch 325, Overall loss = 0.25 and accuracy of 0.866\n",
      "Epoch 326, Overall loss = 0.226 and accuracy of 0.869\n",
      "Epoch 327, Overall loss = 0.232 and accuracy of 0.875\n",
      "Epoch 328, Overall loss = 0.239 and accuracy of 0.865\n",
      "Iteration4600: with minibatch training loss = 0.281 and accuracy of 0.85\n",
      "Epoch 329, Overall loss = 0.222 and accuracy of 0.869\n",
      "Epoch 330, Overall loss = 0.233 and accuracy of 0.867\n",
      "Epoch 331, Overall loss = 0.236 and accuracy of 0.864\n",
      "Epoch 332, Overall loss = 0.231 and accuracy of 0.873\n",
      "Epoch 333, Overall loss = 0.233 and accuracy of 0.87\n",
      "Epoch 334, Overall loss = 0.228 and accuracy of 0.87\n",
      "Epoch 335, Overall loss = 0.225 and accuracy of 0.869\n",
      "Iteration4700: with minibatch training loss = 0.199 and accuracy of 0.88\n",
      "Epoch 336, Overall loss = 0.219 and accuracy of 0.87\n",
      "Epoch 337, Overall loss = 0.229 and accuracy of 0.866\n",
      "Epoch 338, Overall loss = 0.234 and accuracy of 0.87\n",
      "Epoch 339, Overall loss = 0.224 and accuracy of 0.863\n",
      "Epoch 340, Overall loss = 0.219 and accuracy of 0.877\n",
      "Epoch 341, Overall loss = 0.215 and accuracy of 0.872\n",
      "Epoch 342, Overall loss = 0.223 and accuracy of 0.868\n",
      "Iteration4800: with minibatch training loss = 0.269 and accuracy of 0.85\n",
      "Epoch 343, Overall loss = 0.229 and accuracy of 0.871\n",
      "Epoch 344, Overall loss = 0.211 and accuracy of 0.875\n",
      "Epoch 345, Overall loss = 0.223 and accuracy of 0.873\n",
      "Epoch 346, Overall loss = 0.227 and accuracy of 0.871\n",
      "Epoch 347, Overall loss = 0.234 and accuracy of 0.866\n",
      "Epoch 348, Overall loss = 0.233 and accuracy of 0.872\n",
      "Epoch 349, Overall loss = 0.233 and accuracy of 0.872\n",
      "Epoch 350, Overall loss = 0.227 and accuracy of 0.875\n",
      "Iteration4900: with minibatch training loss = 0.306 and accuracy of 0.86\n",
      "Epoch 351, Overall loss = 0.232 and accuracy of 0.869\n",
      "Epoch 352, Overall loss = 0.231 and accuracy of 0.868\n",
      "Epoch 353, Overall loss = 0.231 and accuracy of 0.875\n",
      "Epoch 354, Overall loss = 0.225 and accuracy of 0.873\n",
      "Epoch 355, Overall loss = 0.223 and accuracy of 0.875\n",
      "Epoch 356, Overall loss = 0.226 and accuracy of 0.875\n",
      "Epoch 357, Overall loss = 0.21 and accuracy of 0.872\n",
      "Iteration5000: with minibatch training loss = 0.182 and accuracy of 0.88\n",
      "Epoch 358, Overall loss = 0.225 and accuracy of 0.869\n",
      "Epoch 359, Overall loss = 0.218 and accuracy of 0.877\n",
      "Epoch 360, Overall loss = 0.212 and accuracy of 0.877\n",
      "Epoch 361, Overall loss = 0.226 and accuracy of 0.871\n",
      "Epoch 362, Overall loss = 0.219 and accuracy of 0.877\n",
      "Epoch 363, Overall loss = 0.231 and accuracy of 0.87\n",
      "Epoch 364, Overall loss = 0.229 and accuracy of 0.875\n",
      "Iteration5100: with minibatch training loss = 0.173 and accuracy of 0.89\n",
      "Epoch 365, Overall loss = 0.216 and accuracy of 0.874\n",
      "Epoch 366, Overall loss = 0.222 and accuracy of 0.876\n",
      "Epoch 367, Overall loss = 0.21 and accuracy of 0.873\n",
      "Epoch 368, Overall loss = 0.215 and accuracy of 0.875\n",
      "Epoch 369, Overall loss = 0.211 and accuracy of 0.878\n",
      "Epoch 370, Overall loss = 0.213 and accuracy of 0.87\n",
      "Epoch 371, Overall loss = 0.211 and accuracy of 0.877\n",
      "Iteration5200: with minibatch training loss = 0.215 and accuracy of 0.87\n",
      "Epoch 372, Overall loss = 0.22 and accuracy of 0.875\n",
      "Epoch 373, Overall loss = 0.218 and accuracy of 0.874\n",
      "Epoch 374, Overall loss = 0.215 and accuracy of 0.874\n",
      "Epoch 375, Overall loss = 0.209 and accuracy of 0.871\n",
      "Epoch 376, Overall loss = 0.223 and accuracy of 0.877\n",
      "Epoch 377, Overall loss = 0.208 and accuracy of 0.878\n",
      "Epoch 378, Overall loss = 0.207 and accuracy of 0.873\n",
      "Iteration5300: with minibatch training loss = 0.242 and accuracy of 0.86\n",
      "Epoch 379, Overall loss = 0.219 and accuracy of 0.882\n",
      "Epoch 380, Overall loss = 0.217 and accuracy of 0.874\n",
      "Epoch 381, Overall loss = 0.217 and accuracy of 0.875\n",
      "Epoch 382, Overall loss = 0.215 and accuracy of 0.88\n",
      "Epoch 383, Overall loss = 0.211 and accuracy of 0.874\n",
      "Epoch 384, Overall loss = 0.21 and accuracy of 0.873\n",
      "Epoch 385, Overall loss = 0.207 and accuracy of 0.88\n",
      "Iteration5400: with minibatch training loss = 0.186 and accuracy of 0.88\n",
      "Epoch 386, Overall loss = 0.216 and accuracy of 0.873\n",
      "Epoch 387, Overall loss = 0.217 and accuracy of 0.88\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 388, Overall loss = 0.21 and accuracy of 0.875\n",
      "Epoch 389, Overall loss = 0.206 and accuracy of 0.882\n",
      "Epoch 390, Overall loss = 0.221 and accuracy of 0.876\n",
      "Epoch 391, Overall loss = 0.214 and accuracy of 0.874\n",
      "Epoch 392, Overall loss = 0.212 and accuracy of 0.874\n",
      "Iteration5500: with minibatch training loss = 0.227 and accuracy of 0.88\n",
      "Epoch 393, Overall loss = 0.223 and accuracy of 0.875\n",
      "Epoch 394, Overall loss = 0.206 and accuracy of 0.871\n",
      "Epoch 395, Overall loss = 0.225 and accuracy of 0.88\n",
      "Epoch 396, Overall loss = 0.206 and accuracy of 0.872\n",
      "Epoch 397, Overall loss = 0.21 and accuracy of 0.874\n",
      "Epoch 398, Overall loss = 0.214 and accuracy of 0.873\n",
      "Epoch 399, Overall loss = 0.204 and accuracy of 0.877\n",
      "Epoch 400, Overall loss = 0.204 and accuracy of 0.881\n",
      "Iteration5600: with minibatch training loss = 0.193 and accuracy of 0.88\n",
      "Epoch 401, Overall loss = 0.211 and accuracy of 0.878\n",
      "Epoch 402, Overall loss = 0.217 and accuracy of 0.873\n",
      "Epoch 403, Overall loss = 0.204 and accuracy of 0.877\n",
      "Epoch 404, Overall loss = 0.211 and accuracy of 0.875\n",
      "Epoch 405, Overall loss = 0.213 and accuracy of 0.871\n",
      "Epoch 406, Overall loss = 0.212 and accuracy of 0.87\n",
      "Epoch 407, Overall loss = 0.205 and accuracy of 0.879\n",
      "Iteration5700: with minibatch training loss = 0.19 and accuracy of 0.89\n",
      "Epoch 408, Overall loss = 0.207 and accuracy of 0.876\n",
      "Epoch 409, Overall loss = 0.212 and accuracy of 0.88\n",
      "Epoch 410, Overall loss = 0.203 and accuracy of 0.883\n",
      "Epoch 411, Overall loss = 0.201 and accuracy of 0.877\n",
      "Epoch 412, Overall loss = 0.206 and accuracy of 0.883\n",
      "Epoch 413, Overall loss = 0.206 and accuracy of 0.877\n",
      "Epoch 414, Overall loss = 0.207 and accuracy of 0.876\n",
      "Iteration5800: with minibatch training loss = 0.165 and accuracy of 0.9\n",
      "Epoch 415, Overall loss = 0.204 and accuracy of 0.88\n",
      "Epoch 416, Overall loss = 0.201 and accuracy of 0.884\n",
      "Epoch 417, Overall loss = 0.199 and accuracy of 0.883\n",
      "Epoch 418, Overall loss = 0.199 and accuracy of 0.878\n",
      "Epoch 419, Overall loss = 0.209 and accuracy of 0.883\n",
      "Epoch 420, Overall loss = 0.201 and accuracy of 0.877\n",
      "Epoch 421, Overall loss = 0.2 and accuracy of 0.883\n",
      "Iteration5900: with minibatch training loss = 0.219 and accuracy of 0.85\n",
      "Epoch 422, Overall loss = 0.201 and accuracy of 0.882\n",
      "Epoch 423, Overall loss = 0.206 and accuracy of 0.875\n",
      "Epoch 424, Overall loss = 0.193 and accuracy of 0.883\n",
      "Epoch 425, Overall loss = 0.213 and accuracy of 0.876\n",
      "Epoch 426, Overall loss = 0.206 and accuracy of 0.88\n",
      "Epoch 427, Overall loss = 0.202 and accuracy of 0.878\n",
      "Epoch 428, Overall loss = 0.194 and accuracy of 0.878\n",
      "Iteration6000: with minibatch training loss = 0.212 and accuracy of 0.88\n",
      "Epoch 429, Overall loss = 0.202 and accuracy of 0.881\n",
      "Epoch 430, Overall loss = 0.205 and accuracy of 0.873\n",
      "Epoch 431, Overall loss = 0.197 and accuracy of 0.879\n",
      "Epoch 432, Overall loss = 0.204 and accuracy of 0.874\n",
      "Epoch 433, Overall loss = 0.196 and accuracy of 0.875\n",
      "Epoch 434, Overall loss = 0.204 and accuracy of 0.874\n",
      "Epoch 435, Overall loss = 0.21 and accuracy of 0.874\n",
      "Iteration6100: with minibatch training loss = 0.187 and accuracy of 0.86\n",
      "Epoch 436, Overall loss = 0.195 and accuracy of 0.877\n",
      "Epoch 437, Overall loss = 0.198 and accuracy of 0.88\n",
      "Epoch 438, Overall loss = 0.208 and accuracy of 0.88\n",
      "Epoch 439, Overall loss = 0.2 and accuracy of 0.876\n",
      "Epoch 440, Overall loss = 0.215 and accuracy of 0.872\n",
      "Epoch 441, Overall loss = 0.207 and accuracy of 0.876\n",
      "Epoch 442, Overall loss = 0.198 and accuracy of 0.881\n",
      "Iteration6200: with minibatch training loss = 0.246 and accuracy of 0.87\n",
      "Epoch 443, Overall loss = 0.21 and accuracy of 0.873\n",
      "Epoch 444, Overall loss = 0.204 and accuracy of 0.873\n",
      "Epoch 445, Overall loss = 0.201 and accuracy of 0.879\n",
      "Epoch 446, Overall loss = 0.205 and accuracy of 0.875\n",
      "Epoch 447, Overall loss = 0.205 and accuracy of 0.876\n",
      "Epoch 448, Overall loss = 0.203 and accuracy of 0.878\n",
      "Epoch 449, Overall loss = 0.208 and accuracy of 0.877\n",
      "Epoch 450, Overall loss = 0.191 and accuracy of 0.883\n",
      "Iteration6300: with minibatch training loss = 0.183 and accuracy of 0.88\n",
      "Epoch 451, Overall loss = 0.202 and accuracy of 0.874\n",
      "Epoch 452, Overall loss = 0.202 and accuracy of 0.878\n",
      "Epoch 453, Overall loss = 0.198 and accuracy of 0.882\n",
      "Epoch 454, Overall loss = 0.192 and accuracy of 0.889\n",
      "Epoch 455, Overall loss = 0.216 and accuracy of 0.875\n",
      "Epoch 456, Overall loss = 0.199 and accuracy of 0.886\n",
      "Epoch 457, Overall loss = 0.206 and accuracy of 0.879\n",
      "Iteration6400: with minibatch training loss = 0.191 and accuracy of 0.89\n",
      "Epoch 458, Overall loss = 0.199 and accuracy of 0.877\n",
      "Epoch 459, Overall loss = 0.202 and accuracy of 0.877\n",
      "Epoch 460, Overall loss = 0.202 and accuracy of 0.882\n",
      "Epoch 461, Overall loss = 0.22 and accuracy of 0.879\n",
      "Epoch 462, Overall loss = 0.214 and accuracy of 0.876\n",
      "Epoch 463, Overall loss = 0.2 and accuracy of 0.874\n",
      "Epoch 464, Overall loss = 0.199 and accuracy of 0.879\n",
      "Iteration6500: with minibatch training loss = 0.203 and accuracy of 0.88\n",
      "Epoch 465, Overall loss = 0.199 and accuracy of 0.873\n",
      "Epoch 466, Overall loss = 0.2 and accuracy of 0.881\n",
      "Epoch 467, Overall loss = 0.209 and accuracy of 0.879\n",
      "Epoch 468, Overall loss = 0.204 and accuracy of 0.877\n",
      "Epoch 469, Overall loss = 0.196 and accuracy of 0.882\n",
      "Epoch 470, Overall loss = 0.202 and accuracy of 0.884\n",
      "Epoch 471, Overall loss = 0.197 and accuracy of 0.882\n",
      "Iteration6600: with minibatch training loss = 0.235 and accuracy of 0.87\n",
      "Epoch 472, Overall loss = 0.203 and accuracy of 0.879\n",
      "Epoch 473, Overall loss = 0.2 and accuracy of 0.877\n",
      "Epoch 474, Overall loss = 0.195 and accuracy of 0.886\n",
      "Epoch 475, Overall loss = 0.202 and accuracy of 0.879\n",
      "Epoch 476, Overall loss = 0.199 and accuracy of 0.881\n",
      "Epoch 477, Overall loss = 0.197 and accuracy of 0.878\n",
      "Epoch 478, Overall loss = 0.197 and accuracy of 0.885\n",
      "Iteration6700: with minibatch training loss = 0.236 and accuracy of 0.88\n",
      "Epoch 479, Overall loss = 0.202 and accuracy of 0.882\n",
      "Epoch 480, Overall loss = 0.202 and accuracy of 0.881\n",
      "Epoch 481, Overall loss = 0.199 and accuracy of 0.874\n",
      "Epoch 482, Overall loss = 0.207 and accuracy of 0.875\n",
      "Epoch 483, Overall loss = 0.214 and accuracy of 0.876\n",
      "Epoch 484, Overall loss = 0.21 and accuracy of 0.88\n",
      "Epoch 485, Overall loss = 0.205 and accuracy of 0.873\n",
      "Iteration6800: with minibatch training loss = 0.208 and accuracy of 0.84\n",
      "Epoch 486, Overall loss = 0.202 and accuracy of 0.876\n",
      "Epoch 487, Overall loss = 0.192 and accuracy of 0.874\n",
      "Epoch 488, Overall loss = 0.2 and accuracy of 0.88\n",
      "Epoch 489, Overall loss = 0.205 and accuracy of 0.881\n",
      "Epoch 490, Overall loss = 0.199 and accuracy of 0.875\n",
      "Epoch 491, Overall loss = 0.196 and accuracy of 0.883\n",
      "Epoch 492, Overall loss = 0.196 and accuracy of 0.881\n",
      "Iteration6900: with minibatch training loss = 0.209 and accuracy of 0.89\n",
      "Epoch 493, Overall loss = 0.193 and accuracy of 0.877\n",
      "Epoch 494, Overall loss = 0.191 and accuracy of 0.883\n",
      "Epoch 495, Overall loss = 0.195 and accuracy of 0.884\n",
      "Epoch 496, Overall loss = 0.19 and accuracy of 0.881\n",
      "Epoch 497, Overall loss = 0.203 and accuracy of 0.881\n",
      "Epoch 498, Overall loss = 0.189 and accuracy of 0.875\n",
      "Epoch 499, Overall loss = 0.187 and accuracy of 0.88\n",
      "Epoch 500, Overall loss = 0.192 and accuracy of 0.884\n",
      "Iteration7000: with minibatch training loss = 0.23 and accuracy of 0.89\n",
      "Epoch 501, Overall loss = 0.195 and accuracy of 0.879\n",
      "Epoch 502, Overall loss = 0.193 and accuracy of 0.884\n",
      "Epoch 503, Overall loss = 0.19 and accuracy of 0.884\n",
      "Epoch 504, Overall loss = 0.202 and accuracy of 0.883\n",
      "Epoch 505, Overall loss = 0.201 and accuracy of 0.88\n",
      "Epoch 506, Overall loss = 0.197 and accuracy of 0.879\n",
      "Epoch 507, Overall loss = 0.199 and accuracy of 0.883\n",
      "Iteration7100: with minibatch training loss = 0.176 and accuracy of 0.87\n",
      "Epoch 508, Overall loss = 0.194 and accuracy of 0.882\n",
      "Epoch 509, Overall loss = 0.198 and accuracy of 0.876\n",
      "Epoch 510, Overall loss = 0.196 and accuracy of 0.881\n",
      "Epoch 511, Overall loss = 0.195 and accuracy of 0.883\n",
      "Epoch 512, Overall loss = 0.197 and accuracy of 0.881\n",
      "Epoch 513, Overall loss = 0.192 and accuracy of 0.882\n",
      "Epoch 514, Overall loss = 0.193 and accuracy of 0.881\n",
      "Iteration7200: with minibatch training loss = 0.198 and accuracy of 0.89\n",
      "Epoch 515, Overall loss = 0.196 and accuracy of 0.879\n",
      "Epoch 516, Overall loss = 0.192 and accuracy of 0.88\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 517, Overall loss = 0.187 and accuracy of 0.882\n",
      "Epoch 518, Overall loss = 0.192 and accuracy of 0.881\n",
      "Epoch 519, Overall loss = 0.199 and accuracy of 0.875\n",
      "Epoch 520, Overall loss = 0.199 and accuracy of 0.879\n",
      "Epoch 521, Overall loss = 0.186 and accuracy of 0.883\n",
      "Iteration7300: with minibatch training loss = 0.213 and accuracy of 0.88\n",
      "Epoch 522, Overall loss = 0.193 and accuracy of 0.885\n",
      "Epoch 523, Overall loss = 0.2 and accuracy of 0.881\n",
      "Epoch 524, Overall loss = 0.194 and accuracy of 0.882\n",
      "Epoch 525, Overall loss = 0.197 and accuracy of 0.88\n",
      "Epoch 526, Overall loss = 0.202 and accuracy of 0.88\n",
      "Epoch 527, Overall loss = 0.199 and accuracy of 0.882\n",
      "Epoch 528, Overall loss = 0.189 and accuracy of 0.881\n",
      "Iteration7400: with minibatch training loss = 0.198 and accuracy of 0.87\n",
      "Epoch 529, Overall loss = 0.188 and accuracy of 0.882\n",
      "Epoch 530, Overall loss = 0.188 and accuracy of 0.88\n",
      "Epoch 531, Overall loss = 0.195 and accuracy of 0.881\n",
      "Epoch 532, Overall loss = 0.188 and accuracy of 0.884\n",
      "Epoch 533, Overall loss = 0.189 and accuracy of 0.882\n",
      "Epoch 534, Overall loss = 0.197 and accuracy of 0.883\n",
      "Epoch 535, Overall loss = 0.201 and accuracy of 0.882\n",
      "Iteration7500: with minibatch training loss = 0.223 and accuracy of 0.87\n",
      "Epoch 536, Overall loss = 0.203 and accuracy of 0.882\n",
      "Epoch 537, Overall loss = 0.191 and accuracy of 0.876\n",
      "Epoch 538, Overall loss = 0.193 and accuracy of 0.887\n",
      "Epoch 539, Overall loss = 0.211 and accuracy of 0.879\n",
      "Epoch 540, Overall loss = 0.195 and accuracy of 0.885\n",
      "Epoch 541, Overall loss = 0.188 and accuracy of 0.883\n",
      "Epoch 542, Overall loss = 0.201 and accuracy of 0.883\n",
      "Iteration7600: with minibatch training loss = 0.21 and accuracy of 0.88\n",
      "Epoch 543, Overall loss = 0.204 and accuracy of 0.886\n",
      "Epoch 544, Overall loss = 0.194 and accuracy of 0.885\n",
      "Epoch 545, Overall loss = 0.197 and accuracy of 0.883\n",
      "Epoch 546, Overall loss = 0.196 and accuracy of 0.886\n",
      "Epoch 547, Overall loss = 0.194 and accuracy of 0.89\n",
      "Epoch 548, Overall loss = 0.185 and accuracy of 0.888\n",
      "Epoch 549, Overall loss = 0.189 and accuracy of 0.883\n",
      "Epoch 550, Overall loss = 0.194 and accuracy of 0.884\n",
      "Iteration7700: with minibatch training loss = 0.197 and accuracy of 0.88\n",
      "Epoch 551, Overall loss = 0.185 and accuracy of 0.884\n",
      "Epoch 552, Overall loss = 0.202 and accuracy of 0.88\n",
      "Epoch 553, Overall loss = 0.197 and accuracy of 0.883\n",
      "Epoch 554, Overall loss = 0.203 and accuracy of 0.885\n",
      "Epoch 555, Overall loss = 0.196 and accuracy of 0.88\n",
      "Epoch 556, Overall loss = 0.195 and accuracy of 0.887\n",
      "Epoch 557, Overall loss = 0.205 and accuracy of 0.882\n",
      "Iteration7800: with minibatch training loss = 0.192 and accuracy of 0.88\n",
      "Epoch 558, Overall loss = 0.195 and accuracy of 0.887\n",
      "Epoch 559, Overall loss = 0.197 and accuracy of 0.884\n",
      "Epoch 560, Overall loss = 0.187 and accuracy of 0.89\n",
      "Epoch 561, Overall loss = 0.195 and accuracy of 0.889\n",
      "Epoch 562, Overall loss = 0.19 and accuracy of 0.884\n",
      "Epoch 563, Overall loss = 0.193 and accuracy of 0.889\n",
      "Epoch 564, Overall loss = 0.188 and accuracy of 0.892\n",
      "Iteration7900: with minibatch training loss = 0.145 and accuracy of 0.91\n",
      "Epoch 565, Overall loss = 0.188 and accuracy of 0.886\n",
      "Epoch 566, Overall loss = 0.181 and accuracy of 0.889\n",
      "Epoch 567, Overall loss = 0.191 and accuracy of 0.889\n",
      "Epoch 568, Overall loss = 0.188 and accuracy of 0.89\n",
      "Epoch 569, Overall loss = 0.188 and accuracy of 0.887\n",
      "Epoch 570, Overall loss = 0.19 and accuracy of 0.89\n",
      "Epoch 571, Overall loss = 0.192 and accuracy of 0.891\n",
      "Iteration8000: with minibatch training loss = 0.182 and accuracy of 0.9\n",
      "Epoch 572, Overall loss = 0.184 and accuracy of 0.894\n",
      "Epoch 573, Overall loss = 0.18 and accuracy of 0.895\n",
      "Epoch 574, Overall loss = 0.179 and accuracy of 0.9\n",
      "Epoch 575, Overall loss = 0.199 and accuracy of 0.892\n",
      "Epoch 576, Overall loss = 0.186 and accuracy of 0.903\n",
      "Epoch 577, Overall loss = 0.185 and accuracy of 0.899\n",
      "Epoch 578, Overall loss = 0.182 and accuracy of 0.903\n",
      "Iteration8100: with minibatch training loss = 0.234 and accuracy of 0.89\n",
      "Epoch 579, Overall loss = 0.19 and accuracy of 0.895\n",
      "Epoch 580, Overall loss = 0.177 and accuracy of 0.898\n",
      "Epoch 581, Overall loss = 0.174 and accuracy of 0.907\n",
      "Epoch 582, Overall loss = 0.187 and accuracy of 0.902\n",
      "Epoch 583, Overall loss = 0.19 and accuracy of 0.901\n",
      "Epoch 584, Overall loss = 0.178 and accuracy of 0.909\n",
      "Epoch 585, Overall loss = 0.174 and accuracy of 0.913\n",
      "Iteration8200: with minibatch training loss = 0.196 and accuracy of 0.91\n",
      "Epoch 586, Overall loss = 0.172 and accuracy of 0.913\n",
      "Epoch 587, Overall loss = 0.171 and accuracy of 0.914\n",
      "Epoch 588, Overall loss = 0.171 and accuracy of 0.917\n",
      "Epoch 589, Overall loss = 0.169 and accuracy of 0.916\n",
      "Epoch 590, Overall loss = 0.175 and accuracy of 0.913\n",
      "Epoch 591, Overall loss = 0.166 and accuracy of 0.922\n",
      "Epoch 592, Overall loss = 0.15 and accuracy of 0.93\n",
      "Iteration8300: with minibatch training loss = 0.165 and accuracy of 0.92\n",
      "Epoch 593, Overall loss = 0.165 and accuracy of 0.928\n",
      "Epoch 594, Overall loss = 0.151 and accuracy of 0.927\n",
      "Epoch 595, Overall loss = 0.159 and accuracy of 0.934\n",
      "Epoch 596, Overall loss = 0.154 and accuracy of 0.933\n",
      "Epoch 597, Overall loss = 0.157 and accuracy of 0.933\n",
      "Epoch 598, Overall loss = 0.146 and accuracy of 0.934\n",
      "Epoch 599, Overall loss = 0.137 and accuracy of 0.936\n",
      "Epoch 600, Overall loss = 0.132 and accuracy of 0.944\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f13bf592dd8>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "eval\n",
      "Epoch 1, Overall loss = 0.164 and accuracy of 0.973\n"
     ]
    }
   ],
   "source": [
    "with tf.Session() as sess:\n",
    "    writer.add_graph(sess.graph)\n",
    "    sess.run(init)\n",
    "    print(\"Pretrain layer_1:\")\n",
    "    run_autoencoder(sess, layer_1_para['loss'], b_train_data, layer_1_para['y_pred'],\n",
    "                    1, 800, 512, 50, \n",
    "                    layer_1_para['optimizer'], True)\n",
    "    print(\"Pretrain layer_2:\")\n",
    "    run_autoencoder(sess, layer_2_para['loss'], b_train_data, layer_2_para['y_pred'],\n",
    "                   1, 350, 512, 50,\n",
    "                   layer_2_para['optimizer'], True)\n",
    "    print(\"Pretrain layer_3:\")\n",
    "    run_autoencoder(sess, layer_3_para['loss'], b_train_data, layer_3_para['y_pred'],\n",
    "                    1, 250, 512, 50,\n",
    "                    layer_3_para['optimizer'], True)\n",
    "    print(\"Pretrain layer_4:\")\n",
    "    run_autoencoder(sess, layer_4_para['loss'], b_train_data, layer_4_para['y_pred'],\n",
    "                    1, 200, 512, 50,\n",
    "                    layer_4_para['optimizer'], True)\n",
    "    print(\"train\")\n",
    "    dcf.run_model(sess, dcf.out, dcf.loss, b_train_data, b_train_labels.eval(), 0.55,\n",
    "                 600, 512, 100,\n",
    "                 dcf.train_step, True)\n",
    "    print(\"eval\")\n",
    "    dcf.run_model(sess, dcf.out, dcf.loss, b_eval_data, b_eval_labels.eval(), 1, 1, 512)\n",
    "    \n",
    "    \n",
    "    save_path = saver.save(sess, \"/tmp/model.ckpt\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.6.4"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}