study_mnist.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# MNIST Task\n",
    "\n",
    "This notebook contains the main results for the MNIST task, presented as Figure 6 in the paper. **Note:** in this notebook we select the Fourier components slightly differently than in the paper, but the results should be similar."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {
    "scrolled": true
   },
   "outputs": [],
   "source": [
    "from utils_mnist import *\n",
    "import matplotlib as mpl"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Set optimization parameters\n",
    "\n",
    "For consistency, we set the training parameters to be used in the models below."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "epochs = 200\n",
    "batch_size = 512\n",
    "N_classes = 10"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Load dataset"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "mnist_dp = MNISTDataProcessor()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [],
   "source": [
    "data_N16 = mnist_dp.fourier(2)\n",
    "# data_N36 = mnist_dp.fourier(3)\n",
    "# data_N144 = mnist_dp.fourier(6)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Initialize the models"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [],
   "source": [
    "# The ONN with the electro-optic activation function\n",
    "model_eo_N16 = construct_onn_EO_tf(16, L=2)\n",
    "model_eo_N16.compile(optimizer='adam',\n",
    "                 loss='categorical_crossentropy',\n",
    "                 metrics=['accuracy'])\n",
    "\n",
    "# Just a linear ONN\n",
    "model_linear_N16 = construct_onn_linear_tf(16, L=2)\n",
    "model_linear_N16.compile(optimizer='adam',\n",
    "                 loss='categorical_crossentropy',\n",
    "                 metrics=['accuracy'])\n",
    "\n",
    "\n",
    "# model_eo_N36 = construct_onn_EO_tf(36)\n",
    "# model_eo_N36.compile(optimizer='adam',\n",
    "#                  loss='categorical_crossentropy',\n",
    "#                  metrics=['accuracy'])\n",
    "\n",
    "# model_eo_N144 = construct_onn_EO_tf(144)\n",
    "# model_eo_N144.compile(optimizer='adam',\n",
    "#                  loss='categorical_crossentropy',\n",
    "#                  metrics=['accuracy'])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Train a linear ONN\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "WARNING: Logging before flag parsing goes to stderr.\n",
      "W0715 15:43:30.994544 4541961664 deprecation.py:323] From /Users/iwill/miniconda3/lib/python3.7/site-packages/tensorflow/python/ops/math_grad.py:1250: add_dispatch_support.<locals>.wrapper (from tensorflow.python.ops.array_ops) is deprecated and will be removed in a future version.\n",
      "Instructions for updating:\n",
      "Use tf.where in 2.0, which has the same broadcast rule as np.where\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Train on 60000 samples, validate on 10000 samples\n",
      "Epoch 1/200\n",
      "60000/60000 - 5s - loss: 1.8560 - accuracy: 0.3679 - val_loss: 1.5002 - val_accuracy: 0.6124\n",
      "Epoch 2/200\n",
      "60000/60000 - 2s - loss: 1.3794 - accuracy: 0.6647 - val_loss: 1.2713 - val_accuracy: 0.7287\n",
      "Epoch 3/200\n",
      "60000/60000 - 2s - loss: 1.2388 - accuracy: 0.7271 - val_loss: 1.1863 - val_accuracy: 0.7583\n",
      "Epoch 4/200\n",
      "60000/60000 - 2s - loss: 1.1757 - accuracy: 0.7519 - val_loss: 1.1408 - val_accuracy: 0.7737\n",
      "Epoch 5/200\n",
      "60000/60000 - 2s - loss: 1.1435 - accuracy: 0.7642 - val_loss: 1.1177 - val_accuracy: 0.7813\n",
      "Epoch 6/200\n",
      "60000/60000 - 2s - loss: 1.1282 - accuracy: 0.7694 - val_loss: 1.1030 - val_accuracy: 0.7853\n",
      "Epoch 7/200\n",
      "60000/60000 - 2s - loss: 1.1107 - accuracy: 0.7745 - val_loss: 1.0897 - val_accuracy: 0.7871\n",
      "Epoch 8/200\n",
      "60000/60000 - 2s - loss: 1.0996 - accuracy: 0.7790 - val_loss: 1.0807 - val_accuracy: 0.7928\n",
      "Epoch 9/200\n",
      "60000/60000 - 2s - loss: 1.0903 - accuracy: 0.7824 - val_loss: 1.0707 - val_accuracy: 0.7937\n",
      "Epoch 10/200\n",
      "60000/60000 - 2s - loss: 1.0819 - accuracy: 0.7847 - val_loss: 1.0641 - val_accuracy: 0.7929\n",
      "Epoch 11/200\n",
      "60000/60000 - 2s - loss: 1.0752 - accuracy: 0.7865 - val_loss: 1.0561 - val_accuracy: 0.7993\n",
      "Epoch 12/200\n",
      "60000/60000 - 2s - loss: 1.0682 - accuracy: 0.7876 - val_loss: 1.0497 - val_accuracy: 0.7999\n",
      "Epoch 13/200\n",
      "60000/60000 - 2s - loss: 1.0643 - accuracy: 0.7865 - val_loss: 1.0439 - val_accuracy: 0.7982\n",
      "Epoch 14/200\n",
      "60000/60000 - 2s - loss: 1.0563 - accuracy: 0.7908 - val_loss: 1.0381 - val_accuracy: 0.8062\n",
      "Epoch 15/200\n",
      "60000/60000 - 2s - loss: 1.0510 - accuracy: 0.7914 - val_loss: 1.0328 - val_accuracy: 0.8033\n",
      "Epoch 16/200\n",
      "60000/60000 - 2s - loss: 1.0463 - accuracy: 0.7929 - val_loss: 1.0286 - val_accuracy: 0.8029\n",
      "Epoch 17/200\n",
      "60000/60000 - 2s - loss: 1.0415 - accuracy: 0.7946 - val_loss: 1.0238 - val_accuracy: 0.8052\n",
      "Epoch 18/200\n",
      "60000/60000 - 2s - loss: 1.0370 - accuracy: 0.7956 - val_loss: 1.0202 - val_accuracy: 0.8051\n",
      "Epoch 19/200\n",
      "60000/60000 - 2s - loss: 1.0324 - accuracy: 0.7974 - val_loss: 1.0155 - val_accuracy: 0.8064\n",
      "Epoch 20/200\n",
      "60000/60000 - 2s - loss: 1.0285 - accuracy: 0.7986 - val_loss: 1.0112 - val_accuracy: 0.8089\n",
      "Epoch 21/200\n",
      "60000/60000 - 2s - loss: 1.0246 - accuracy: 0.7997 - val_loss: 1.0079 - val_accuracy: 0.8091\n",
      "Epoch 22/200\n",
      "60000/60000 - 2s - loss: 1.0209 - accuracy: 0.8002 - val_loss: 1.0043 - val_accuracy: 0.8127\n",
      "Epoch 23/200\n",
      "60000/60000 - 2s - loss: 1.0172 - accuracy: 0.8018 - val_loss: 1.0013 - val_accuracy: 0.8098\n",
      "Epoch 24/200\n",
      "60000/60000 - 2s - loss: 1.0137 - accuracy: 0.8027 - val_loss: 0.9972 - val_accuracy: 0.8149\n",
      "Epoch 25/200\n",
      "60000/60000 - 1s - loss: 1.0106 - accuracy: 0.8040 - val_loss: 0.9958 - val_accuracy: 0.8102\n",
      "Epoch 26/200\n",
      "60000/60000 - 1s - loss: 1.0079 - accuracy: 0.8028 - val_loss: 0.9903 - val_accuracy: 0.8142\n",
      "Epoch 27/200\n",
      "60000/60000 - 1s - loss: 1.0049 - accuracy: 0.8052 - val_loss: 0.9911 - val_accuracy: 0.8127\n",
      "Epoch 28/200\n",
      "60000/60000 - 1s - loss: 1.0020 - accuracy: 0.8050 - val_loss: 0.9863 - val_accuracy: 0.8149\n",
      "Epoch 29/200\n",
      "60000/60000 - 1s - loss: 0.9993 - accuracy: 0.8063 - val_loss: 0.9836 - val_accuracy: 0.8171\n",
      "Epoch 30/200\n",
      "60000/60000 - 1s - loss: 0.9975 - accuracy: 0.8058 - val_loss: 0.9817 - val_accuracy: 0.8152\n",
      "Epoch 31/200\n",
      "60000/60000 - 1s - loss: 0.9952 - accuracy: 0.8062 - val_loss: 0.9802 - val_accuracy: 0.8196\n",
      "Epoch 32/200\n",
      "60000/60000 - 1s - loss: 0.9939 - accuracy: 0.8070 - val_loss: 0.9777 - val_accuracy: 0.8166\n",
      "Epoch 33/200\n",
      "60000/60000 - 1s - loss: 0.9919 - accuracy: 0.8076 - val_loss: 0.9754 - val_accuracy: 0.8152\n",
      "Epoch 34/200\n",
      "60000/60000 - 1s - loss: 0.9903 - accuracy: 0.8068 - val_loss: 0.9748 - val_accuracy: 0.8169\n",
      "Epoch 35/200\n",
      "60000/60000 - 1s - loss: 0.9883 - accuracy: 0.8082 - val_loss: 0.9732 - val_accuracy: 0.8140\n",
      "Epoch 36/200\n",
      "60000/60000 - 1s - loss: 0.9866 - accuracy: 0.8072 - val_loss: 0.9709 - val_accuracy: 0.8168\n",
      "Epoch 37/200\n",
      "60000/60000 - 1s - loss: 0.9856 - accuracy: 0.8074 - val_loss: 0.9706 - val_accuracy: 0.8189\n",
      "Epoch 38/200\n",
      "60000/60000 - 1s - loss: 0.9866 - accuracy: 0.8071 - val_loss: 0.9687 - val_accuracy: 0.8178\n",
      "Epoch 39/200\n",
      "60000/60000 - 1s - loss: 0.9868 - accuracy: 0.8073 - val_loss: 0.9762 - val_accuracy: 0.8153\n",
      "Epoch 40/200\n",
      "60000/60000 - 1s - loss: 0.9825 - accuracy: 0.8076 - val_loss: 0.9679 - val_accuracy: 0.8167\n",
      "Epoch 41/200\n",
      "60000/60000 - 1s - loss: 0.9823 - accuracy: 0.8084 - val_loss: 0.9672 - val_accuracy: 0.8171\n",
      "Epoch 42/200\n",
      "60000/60000 - 1s - loss: 0.9799 - accuracy: 0.8088 - val_loss: 0.9651 - val_accuracy: 0.8155\n",
      "Epoch 43/200\n",
      "60000/60000 - 1s - loss: 0.9787 - accuracy: 0.8084 - val_loss: 0.9645 - val_accuracy: 0.8146\n",
      "Epoch 44/200\n",
      "60000/60000 - 1s - loss: 0.9777 - accuracy: 0.8088 - val_loss: 0.9641 - val_accuracy: 0.8177\n",
      "Epoch 45/200\n",
      "60000/60000 - 1s - loss: 0.9773 - accuracy: 0.8092 - val_loss: 0.9636 - val_accuracy: 0.8189\n",
      "Epoch 46/200\n",
      "60000/60000 - 1s - loss: 0.9763 - accuracy: 0.8097 - val_loss: 0.9613 - val_accuracy: 0.8177\n",
      "Epoch 47/200\n",
      "60000/60000 - 1s - loss: 0.9754 - accuracy: 0.8095 - val_loss: 0.9617 - val_accuracy: 0.8176\n",
      "Epoch 48/200\n",
      "60000/60000 - 1s - loss: 0.9754 - accuracy: 0.8091 - val_loss: 0.9607 - val_accuracy: 0.8162\n",
      "Epoch 49/200\n",
      "60000/60000 - 1s - loss: 0.9753 - accuracy: 0.8091 - val_loss: 0.9598 - val_accuracy: 0.8141\n",
      "Epoch 50/200\n",
      "60000/60000 - 2s - loss: 0.9734 - accuracy: 0.8095 - val_loss: 0.9589 - val_accuracy: 0.8184\n",
      "Epoch 51/200\n",
      "60000/60000 - 2s - loss: 0.9727 - accuracy: 0.8103 - val_loss: 0.9600 - val_accuracy: 0.8171\n",
      "Epoch 52/200\n",
      "60000/60000 - 2s - loss: 0.9718 - accuracy: 0.8099 - val_loss: 0.9586 - val_accuracy: 0.8182\n",
      "Epoch 53/200\n",
      "60000/60000 - 2s - loss: 0.9716 - accuracy: 0.8097 - val_loss: 0.9575 - val_accuracy: 0.8144\n",
      "Epoch 54/200\n",
      "60000/60000 - 1s - loss: 0.9708 - accuracy: 0.8097 - val_loss: 0.9571 - val_accuracy: 0.8168\n",
      "Epoch 55/200\n",
      "60000/60000 - 2s - loss: 0.9708 - accuracy: 0.8106 - val_loss: 0.9572 - val_accuracy: 0.8188\n",
      "Epoch 56/200\n",
      "60000/60000 - 1s - loss: 0.9705 - accuracy: 0.8099 - val_loss: 0.9590 - val_accuracy: 0.8155\n",
      "Epoch 57/200\n",
      "60000/60000 - 1s - loss: 0.9688 - accuracy: 0.8106 - val_loss: 0.9569 - val_accuracy: 0.8169\n",
      "Epoch 58/200\n",
      "60000/60000 - 1s - loss: 0.9686 - accuracy: 0.8099 - val_loss: 0.9558 - val_accuracy: 0.8141\n",
      "Epoch 59/200\n",
      "60000/60000 - 1s - loss: 0.9686 - accuracy: 0.8103 - val_loss: 0.9559 - val_accuracy: 0.8187\n",
      "Epoch 60/200\n",
      "60000/60000 - 1s - loss: 0.9683 - accuracy: 0.8098 - val_loss: 0.9577 - val_accuracy: 0.8171\n",
      "Epoch 61/200\n",
      "60000/60000 - 2s - loss: 0.9675 - accuracy: 0.8098 - val_loss: 0.9556 - val_accuracy: 0.8178\n",
      "Epoch 62/200\n",
      "60000/60000 - 1s - loss: 0.9670 - accuracy: 0.8101 - val_loss: 0.9544 - val_accuracy: 0.8163\n",
      "Epoch 63/200\n",
      "60000/60000 - 1s - loss: 0.9668 - accuracy: 0.8100 - val_loss: 0.9541 - val_accuracy: 0.8221\n",
      "Epoch 64/200\n",
      "60000/60000 - 2s - loss: 0.9656 - accuracy: 0.8095 - val_loss: 0.9548 - val_accuracy: 0.8178\n",
      "Epoch 65/200\n",
      "60000/60000 - 1s - loss: 0.9665 - accuracy: 0.8094 - val_loss: 0.9529 - val_accuracy: 0.8169\n",
      "Epoch 66/200\n",
      "60000/60000 - 2s - loss: 0.9705 - accuracy: 0.8091 - val_loss: 0.9529 - val_accuracy: 0.8195\n",
      "Epoch 67/200\n",
      "60000/60000 - 2s - loss: 0.9649 - accuracy: 0.8098 - val_loss: 0.9538 - val_accuracy: 0.8172\n",
      "Epoch 68/200\n",
      "60000/60000 - 2s - loss: 0.9660 - accuracy: 0.8104 - val_loss: 0.9521 - val_accuracy: 0.8214\n",
      "Epoch 69/200\n",
      "60000/60000 - 2s - loss: 0.9639 - accuracy: 0.8113 - val_loss: 0.9528 - val_accuracy: 0.8132\n",
      "Epoch 70/200\n",
      "60000/60000 - 2s - loss: 0.9641 - accuracy: 0.8102 - val_loss: 0.9532 - val_accuracy: 0.8147\n",
      "Epoch 71/200\n",
      "60000/60000 - 1s - loss: 0.9646 - accuracy: 0.8101 - val_loss: 0.9504 - val_accuracy: 0.8177\n",
      "Epoch 72/200\n",
      "60000/60000 - 1s - loss: 0.9639 - accuracy: 0.8101 - val_loss: 0.9512 - val_accuracy: 0.8186\n",
      "Epoch 73/200\n",
      "60000/60000 - 2s - loss: 0.9630 - accuracy: 0.8101 - val_loss: 0.9501 - val_accuracy: 0.8181\n",
      "Epoch 74/200\n",
      "60000/60000 - 2s - loss: 0.9625 - accuracy: 0.8096 - val_loss: 0.9517 - val_accuracy: 0.8142\n",
      "Epoch 75/200\n",
      "60000/60000 - 2s - loss: 0.9629 - accuracy: 0.8108 - val_loss: 0.9521 - val_accuracy: 0.8162\n",
      "Epoch 76/200\n",
      "60000/60000 - 2s - loss: 0.9622 - accuracy: 0.8098 - val_loss: 0.9499 - val_accuracy: 0.8208\n",
      "Epoch 77/200\n",
      "60000/60000 - 1s - loss: 0.9623 - accuracy: 0.8097 - val_loss: 0.9495 - val_accuracy: 0.8157\n",
      "Epoch 78/200\n",
      "60000/60000 - 2s - loss: 0.9630 - accuracy: 0.8105 - val_loss: 0.9505 - val_accuracy: 0.8140\n",
      "Epoch 79/200\n",
      "60000/60000 - 2s - loss: 0.9614 - accuracy: 0.8108 - val_loss: 0.9489 - val_accuracy: 0.8198\n",
      "Epoch 80/200\n",
      "60000/60000 - 2s - loss: 0.9616 - accuracy: 0.8102 - val_loss: 0.9502 - val_accuracy: 0.8175\n",
      "Epoch 81/200\n",
      "60000/60000 - 2s - loss: 0.9611 - accuracy: 0.8108 - val_loss: 0.9480 - val_accuracy: 0.8190\n",
      "Epoch 82/200\n",
      "60000/60000 - 2s - loss: 0.9608 - accuracy: 0.8100 - val_loss: 0.9491 - val_accuracy: 0.8185\n",
      "Epoch 83/200\n",
      "60000/60000 - 1s - loss: 0.9604 - accuracy: 0.8106 - val_loss: 0.9490 - val_accuracy: 0.8175\n",
      "Epoch 84/200\n",
      "60000/60000 - 2s - loss: 0.9607 - accuracy: 0.8102 - val_loss: 0.9495 - val_accuracy: 0.8185\n",
      "Epoch 85/200\n",
      "60000/60000 - 2s - loss: 0.9607 - accuracy: 0.8110 - val_loss: 0.9508 - val_accuracy: 0.8130\n",
      "Epoch 86/200\n",
      "60000/60000 - 2s - loss: 0.9608 - accuracy: 0.8099 - val_loss: 0.9517 - val_accuracy: 0.8171\n",
      "Epoch 87/200\n",
      "60000/60000 - 1s - loss: 0.9599 - accuracy: 0.8112 - val_loss: 0.9490 - val_accuracy: 0.8159\n",
      "Epoch 88/200\n",
      "60000/60000 - 2s - loss: 0.9602 - accuracy: 0.8103 - val_loss: 0.9487 - val_accuracy: 0.8157\n",
      "Epoch 89/200\n",
      "60000/60000 - 2s - loss: 0.9598 - accuracy: 0.8094 - val_loss: 0.9485 - val_accuracy: 0.8185\n",
      "Epoch 90/200\n",
      "60000/60000 - 2s - loss: 0.9608 - accuracy: 0.8095 - val_loss: 0.9474 - val_accuracy: 0.8181\n",
      "Epoch 91/200\n",
      "60000/60000 - 2s - loss: 0.9592 - accuracy: 0.8097 - val_loss: 0.9477 - val_accuracy: 0.8200\n",
      "Epoch 92/200\n",
      "60000/60000 - 2s - loss: 0.9589 - accuracy: 0.8103 - val_loss: 0.9488 - val_accuracy: 0.8145\n",
      "Epoch 93/200\n",
      "60000/60000 - 2s - loss: 0.9590 - accuracy: 0.8105 - val_loss: 0.9470 - val_accuracy: 0.8175\n",
      "Epoch 94/200\n",
      "60000/60000 - 2s - loss: 0.9587 - accuracy: 0.8102 - val_loss: 0.9466 - val_accuracy: 0.8214\n",
      "Epoch 95/200\n",
      "60000/60000 - 2s - loss: 0.9590 - accuracy: 0.8097 - val_loss: 0.9476 - val_accuracy: 0.8181\n",
      "Epoch 96/200\n",
      "60000/60000 - 2s - loss: 0.9581 - accuracy: 0.8102 - val_loss: 0.9462 - val_accuracy: 0.8191\n",
      "Epoch 97/200\n",
      "60000/60000 - 2s - loss: 0.9587 - accuracy: 0.8102 - val_loss: 0.9476 - val_accuracy: 0.8165\n",
      "Epoch 98/200\n",
      "60000/60000 - 2s - loss: 0.9582 - accuracy: 0.8102 - val_loss: 0.9482 - val_accuracy: 0.8184\n",
      "Epoch 99/200\n",
      "60000/60000 - 2s - loss: 0.9587 - accuracy: 0.8102 - val_loss: 0.9491 - val_accuracy: 0.8181\n",
      "Epoch 100/200\n",
      "60000/60000 - 2s - loss: 0.9582 - accuracy: 0.8096 - val_loss: 0.9468 - val_accuracy: 0.8194\n",
      "Epoch 101/200\n",
      "60000/60000 - 2s - loss: 0.9581 - accuracy: 0.8106 - val_loss: 0.9467 - val_accuracy: 0.8159\n",
      "Epoch 102/200\n",
      "60000/60000 - 2s - loss: 0.9579 - accuracy: 0.8103 - val_loss: 0.9450 - val_accuracy: 0.8194\n",
      "Epoch 103/200\n",
      "60000/60000 - 2s - loss: 0.9580 - accuracy: 0.8110 - val_loss: 0.9465 - val_accuracy: 0.8182\n",
      "Epoch 104/200\n",
      "60000/60000 - 2s - loss: 0.9575 - accuracy: 0.8103 - val_loss: 0.9465 - val_accuracy: 0.8187\n",
      "Epoch 105/200\n",
      "60000/60000 - 2s - loss: 0.9576 - accuracy: 0.8104 - val_loss: 0.9460 - val_accuracy: 0.8214\n",
      "Epoch 106/200\n",
      "60000/60000 - 2s - loss: 0.9574 - accuracy: 0.8102 - val_loss: 0.9454 - val_accuracy: 0.8202\n",
      "Epoch 107/200\n",
      "60000/60000 - 2s - loss: 0.9571 - accuracy: 0.8106 - val_loss: 0.9455 - val_accuracy: 0.8130\n",
      "Epoch 108/200\n",
      "60000/60000 - 2s - loss: 0.9585 - accuracy: 0.8107 - val_loss: 0.9459 - val_accuracy: 0.8173\n",
      "Epoch 109/200\n",
      "60000/60000 - 2s - loss: 0.9573 - accuracy: 0.8112 - val_loss: 0.9468 - val_accuracy: 0.8158\n",
      "Epoch 110/200\n",
      "60000/60000 - 2s - loss: 0.9573 - accuracy: 0.8104 - val_loss: 0.9459 - val_accuracy: 0.8168\n",
      "Epoch 111/200\n",
      "60000/60000 - 2s - loss: 0.9569 - accuracy: 0.8109 - val_loss: 0.9446 - val_accuracy: 0.8171\n",
      "Epoch 112/200\n",
      "60000/60000 - 2s - loss: 0.9568 - accuracy: 0.8106 - val_loss: 0.9447 - val_accuracy: 0.8188\n",
      "Epoch 113/200\n",
      "60000/60000 - 2s - loss: 0.9572 - accuracy: 0.8096 - val_loss: 0.9464 - val_accuracy: 0.8196\n",
      "Epoch 114/200\n",
      "60000/60000 - 2s - loss: 0.9569 - accuracy: 0.8096 - val_loss: 0.9441 - val_accuracy: 0.8164\n",
      "Epoch 115/200\n",
      "60000/60000 - 2s - loss: 0.9565 - accuracy: 0.8106 - val_loss: 0.9438 - val_accuracy: 0.8183\n",
      "Epoch 116/200\n",
      "60000/60000 - 2s - loss: 0.9569 - accuracy: 0.8101 - val_loss: 0.9446 - val_accuracy: 0.8182\n",
      "Epoch 117/200\n",
      "60000/60000 - 2s - loss: 0.9567 - accuracy: 0.8106 - val_loss: 0.9450 - val_accuracy: 0.8200\n",
      "Epoch 118/200\n",
      "60000/60000 - 2s - loss: 0.9570 - accuracy: 0.8104 - val_loss: 0.9460 - val_accuracy: 0.8176\n",
      "Epoch 119/200\n",
      "60000/60000 - 2s - loss: 0.9566 - accuracy: 0.8119 - val_loss: 0.9452 - val_accuracy: 0.8202\n",
      "Epoch 120/200\n",
      "60000/60000 - 2s - loss: 0.9563 - accuracy: 0.8100 - val_loss: 0.9440 - val_accuracy: 0.8192\n",
      "Epoch 121/200\n",
      "60000/60000 - 2s - loss: 0.9562 - accuracy: 0.8112 - val_loss: 0.9445 - val_accuracy: 0.8186\n",
      "Epoch 122/200\n",
      "60000/60000 - 2s - loss: 0.9567 - accuracy: 0.8110 - val_loss: 0.9442 - val_accuracy: 0.8174\n",
      "Epoch 123/200\n",
      "60000/60000 - 2s - loss: 0.9563 - accuracy: 0.8101 - val_loss: 0.9443 - val_accuracy: 0.8181\n",
      "Epoch 124/200\n",
      "60000/60000 - 2s - loss: 0.9565 - accuracy: 0.8100 - val_loss: 0.9454 - val_accuracy: 0.8199\n",
      "Epoch 125/200\n",
      "60000/60000 - 2s - loss: 0.9566 - accuracy: 0.8109 - val_loss: 0.9436 - val_accuracy: 0.8178\n",
      "Epoch 126/200\n",
      "60000/60000 - 2s - loss: 0.9563 - accuracy: 0.8105 - val_loss: 0.9438 - val_accuracy: 0.8188\n",
      "Epoch 127/200\n",
      "60000/60000 - 2s - loss: 0.9565 - accuracy: 0.8101 - val_loss: 0.9428 - val_accuracy: 0.8195\n",
      "Epoch 128/200\n",
      "60000/60000 - 2s - loss: 0.9564 - accuracy: 0.8099 - val_loss: 0.9458 - val_accuracy: 0.8178\n",
      "Epoch 129/200\n",
      "60000/60000 - 2s - loss: 0.9561 - accuracy: 0.8102 - val_loss: 0.9438 - val_accuracy: 0.8203\n",
      "Epoch 130/200\n",
      "60000/60000 - 2s - loss: 0.9561 - accuracy: 0.8105 - val_loss: 0.9453 - val_accuracy: 0.8185\n",
      "Epoch 131/200\n",
      "60000/60000 - 2s - loss: 0.9563 - accuracy: 0.8100 - val_loss: 0.9437 - val_accuracy: 0.8189\n",
      "Epoch 132/200\n",
      "60000/60000 - 2s - loss: 0.9561 - accuracy: 0.8093 - val_loss: 0.9441 - val_accuracy: 0.8170\n",
      "Epoch 133/200\n",
      "60000/60000 - 2s - loss: 0.9563 - accuracy: 0.8111 - val_loss: 0.9453 - val_accuracy: 0.8165\n",
      "Epoch 134/200\n",
      "60000/60000 - 2s - loss: 0.9563 - accuracy: 0.8099 - val_loss: 0.9432 - val_accuracy: 0.8177\n",
      "Epoch 135/200\n",
      "60000/60000 - 2s - loss: 0.9562 - accuracy: 0.8109 - val_loss: 0.9425 - val_accuracy: 0.8204\n",
      "Epoch 136/200\n",
      "60000/60000 - 2s - loss: 0.9563 - accuracy: 0.8104 - val_loss: 0.9434 - val_accuracy: 0.8198\n",
      "Epoch 137/200\n",
      "60000/60000 - 2s - loss: 0.9559 - accuracy: 0.8099 - val_loss: 0.9425 - val_accuracy: 0.8198\n",
      "Epoch 138/200\n",
      "60000/60000 - 2s - loss: 0.9560 - accuracy: 0.8109 - val_loss: 0.9444 - val_accuracy: 0.8169\n",
      "Epoch 139/200\n",
      "60000/60000 - 2s - loss: 0.9563 - accuracy: 0.8104 - val_loss: 0.9431 - val_accuracy: 0.8202\n",
      "Epoch 140/200\n",
      "60000/60000 - 2s - loss: 0.9561 - accuracy: 0.8105 - val_loss: 0.9441 - val_accuracy: 0.8177\n",
      "Epoch 141/200\n",
      "60000/60000 - 2s - loss: 0.9565 - accuracy: 0.8101 - val_loss: 0.9440 - val_accuracy: 0.8213\n",
      "Epoch 142/200\n",
      "60000/60000 - 2s - loss: 0.9563 - accuracy: 0.8109 - val_loss: 0.9434 - val_accuracy: 0.8187\n",
      "Epoch 143/200\n",
      "60000/60000 - 2s - loss: 0.9560 - accuracy: 0.8105 - val_loss: 0.9438 - val_accuracy: 0.8181\n",
      "Epoch 144/200\n",
      "60000/60000 - 2s - loss: 0.9574 - accuracy: 0.8106 - val_loss: 0.9426 - val_accuracy: 0.8199\n",
      "Epoch 145/200\n",
      "60000/60000 - 2s - loss: 0.9562 - accuracy: 0.8107 - val_loss: 0.9427 - val_accuracy: 0.8205\n",
      "Epoch 146/200\n",
      "60000/60000 - 2s - loss: 0.9556 - accuracy: 0.8111 - val_loss: 0.9434 - val_accuracy: 0.8183\n",
      "Epoch 147/200\n",
      "60000/60000 - 2s - loss: 0.9559 - accuracy: 0.8099 - val_loss: 0.9436 - val_accuracy: 0.8201\n",
      "Epoch 148/200\n",
      "60000/60000 - 2s - loss: 0.9564 - accuracy: 0.8104 - val_loss: 0.9441 - val_accuracy: 0.8214\n",
      "Epoch 149/200\n",
      "60000/60000 - 2s - loss: 0.9558 - accuracy: 0.8101 - val_loss: 0.9435 - val_accuracy: 0.8185\n",
      "Epoch 150/200\n",
      "60000/60000 - 2s - loss: 0.9557 - accuracy: 0.8112 - val_loss: 0.9450 - val_accuracy: 0.8149\n",
      "Epoch 151/200\n",
      "60000/60000 - 2s - loss: 0.9554 - accuracy: 0.8102 - val_loss: 0.9433 - val_accuracy: 0.8198\n",
      "Epoch 152/200\n",
      "60000/60000 - 2s - loss: 0.9558 - accuracy: 0.8111 - val_loss: 0.9436 - val_accuracy: 0.8195\n",
      "Epoch 153/200\n",
      "60000/60000 - 2s - loss: 0.9557 - accuracy: 0.8106 - val_loss: 0.9432 - val_accuracy: 0.8194\n",
      "Epoch 154/200\n",
      "60000/60000 - 2s - loss: 0.9559 - accuracy: 0.8106 - val_loss: 0.9442 - val_accuracy: 0.8204\n",
      "Epoch 155/200\n",
      "60000/60000 - 2s - loss: 0.9555 - accuracy: 0.8109 - val_loss: 0.9440 - val_accuracy: 0.8171\n",
      "Epoch 156/200\n",
      "60000/60000 - 2s - loss: 0.9618 - accuracy: 0.8080 - val_loss: 0.9534 - val_accuracy: 0.8176\n",
      "Epoch 157/200\n",
      "60000/60000 - 2s - loss: 0.9568 - accuracy: 0.8096 - val_loss: 0.9439 - val_accuracy: 0.8191\n",
      "Epoch 158/200\n",
      "60000/60000 - 2s - loss: 0.9557 - accuracy: 0.8109 - val_loss: 0.9463 - val_accuracy: 0.8197\n",
      "Epoch 159/200\n",
      "60000/60000 - 2s - loss: 0.9558 - accuracy: 0.8105 - val_loss: 0.9444 - val_accuracy: 0.8208\n",
      "Epoch 160/200\n",
      "60000/60000 - 2s - loss: 0.9554 - accuracy: 0.8111 - val_loss: 0.9431 - val_accuracy: 0.8210\n",
      "Epoch 161/200\n",
      "60000/60000 - 2s - loss: 0.9555 - accuracy: 0.8106 - val_loss: 0.9445 - val_accuracy: 0.8162\n",
      "Epoch 162/200\n",
      "60000/60000 - 2s - loss: 0.9554 - accuracy: 0.8115 - val_loss: 0.9431 - val_accuracy: 0.8207\n",
      "Epoch 163/200\n",
      "60000/60000 - 2s - loss: 0.9571 - accuracy: 0.8103 - val_loss: 0.9460 - val_accuracy: 0.8175\n",
      "Epoch 164/200\n",
      "60000/60000 - 2s - loss: 0.9559 - accuracy: 0.8113 - val_loss: 0.9435 - val_accuracy: 0.8193\n",
      "Epoch 165/200\n",
      "60000/60000 - 2s - loss: 0.9554 - accuracy: 0.8102 - val_loss: 0.9430 - val_accuracy: 0.8220\n",
      "Epoch 166/200\n",
      "60000/60000 - 2s - loss: 0.9557 - accuracy: 0.8109 - val_loss: 0.9451 - val_accuracy: 0.8193\n",
      "Epoch 167/200\n",
      "60000/60000 - 2s - loss: 0.9554 - accuracy: 0.8103 - val_loss: 0.9432 - val_accuracy: 0.8196\n",
      "Epoch 168/200\n",
      "60000/60000 - 2s - loss: 0.9558 - accuracy: 0.8106 - val_loss: 0.9439 - val_accuracy: 0.8161\n",
      "Epoch 169/200\n",
      "60000/60000 - 2s - loss: 0.9557 - accuracy: 0.8110 - val_loss: 0.9437 - val_accuracy: 0.8204\n",
      "Epoch 170/200\n",
      "60000/60000 - 2s - loss: 0.9556 - accuracy: 0.8102 - val_loss: 0.9436 - val_accuracy: 0.8177\n",
      "Epoch 171/200\n",
      "60000/60000 - 2s - loss: 0.9561 - accuracy: 0.8103 - val_loss: 0.9425 - val_accuracy: 0.8204\n",
      "Epoch 172/200\n",
      "60000/60000 - 1s - loss: 0.9555 - accuracy: 0.8112 - val_loss: 0.9431 - val_accuracy: 0.8200\n",
      "Epoch 173/200\n",
      "60000/60000 - 2s - loss: 0.9560 - accuracy: 0.8099 - val_loss: 0.9429 - val_accuracy: 0.8195\n",
      "Epoch 174/200\n",
      "60000/60000 - 2s - loss: 0.9557 - accuracy: 0.8104 - val_loss: 0.9428 - val_accuracy: 0.8208\n",
      "Epoch 175/200\n",
      "60000/60000 - 2s - loss: 0.9557 - accuracy: 0.8113 - val_loss: 0.9434 - val_accuracy: 0.8207\n",
      "Epoch 176/200\n",
      "60000/60000 - 2s - loss: 0.9555 - accuracy: 0.8102 - val_loss: 0.9453 - val_accuracy: 0.8161\n",
      "Epoch 177/200\n",
      "60000/60000 - 2s - loss: 0.9559 - accuracy: 0.8109 - val_loss: 0.9433 - val_accuracy: 0.8195\n",
      "Epoch 178/200\n",
      "60000/60000 - 2s - loss: 0.9551 - accuracy: 0.8110 - val_loss: 0.9433 - val_accuracy: 0.8220\n",
      "Epoch 179/200\n",
      "60000/60000 - 2s - loss: 0.9556 - accuracy: 0.8107 - val_loss: 0.9451 - val_accuracy: 0.8207\n",
      "Epoch 180/200\n",
      "60000/60000 - 2s - loss: 0.9553 - accuracy: 0.8096 - val_loss: 0.9430 - val_accuracy: 0.8186\n",
      "Epoch 181/200\n",
      "60000/60000 - 2s - loss: 0.9554 - accuracy: 0.8104 - val_loss: 0.9439 - val_accuracy: 0.8202\n",
      "Epoch 182/200\n",
      "60000/60000 - 2s - loss: 0.9556 - accuracy: 0.8109 - val_loss: 0.9437 - val_accuracy: 0.8193\n",
      "Epoch 183/200\n",
      "60000/60000 - 2s - loss: 0.9554 - accuracy: 0.8105 - val_loss: 0.9432 - val_accuracy: 0.8193\n",
      "Epoch 184/200\n",
      "60000/60000 - 2s - loss: 0.9555 - accuracy: 0.8108 - val_loss: 0.9423 - val_accuracy: 0.8204\n",
      "Epoch 185/200\n",
      "60000/60000 - 2s - loss: 0.9551 - accuracy: 0.8105 - val_loss: 0.9433 - val_accuracy: 0.8213\n",
      "Epoch 186/200\n",
      "60000/60000 - 2s - loss: 0.9554 - accuracy: 0.8098 - val_loss: 0.9419 - val_accuracy: 0.8223\n",
      "Epoch 187/200\n",
      "60000/60000 - 2s - loss: 0.9552 - accuracy: 0.8110 - val_loss: 0.9432 - val_accuracy: 0.8210\n",
      "Epoch 188/200\n",
      "60000/60000 - 2s - loss: 0.9555 - accuracy: 0.8109 - val_loss: 0.9436 - val_accuracy: 0.8188\n",
      "Epoch 189/200\n",
      "60000/60000 - 2s - loss: 0.9558 - accuracy: 0.8104 - val_loss: 0.9508 - val_accuracy: 0.8161\n",
      "Epoch 190/200\n",
      "60000/60000 - 2s - loss: 0.9558 - accuracy: 0.8108 - val_loss: 0.9429 - val_accuracy: 0.8199\n",
      "Epoch 191/200\n",
      "60000/60000 - 2s - loss: 0.9559 - accuracy: 0.8100 - val_loss: 0.9436 - val_accuracy: 0.8215\n",
      "Epoch 192/200\n",
      "60000/60000 - 2s - loss: 0.9556 - accuracy: 0.8113 - val_loss: 0.9434 - val_accuracy: 0.8191\n",
      "Epoch 193/200\n",
      "60000/60000 - 1s - loss: 0.9556 - accuracy: 0.8099 - val_loss: 0.9450 - val_accuracy: 0.8183\n",
      "Epoch 194/200\n",
      "60000/60000 - 2s - loss: 0.9559 - accuracy: 0.8107 - val_loss: 0.9435 - val_accuracy: 0.8196\n",
      "Epoch 195/200\n",
      "60000/60000 - 2s - loss: 0.9584 - accuracy: 0.8110 - val_loss: 0.9476 - val_accuracy: 0.8206\n",
      "Epoch 196/200\n",
      "60000/60000 - 2s - loss: 0.9563 - accuracy: 0.8098 - val_loss: 0.9433 - val_accuracy: 0.8180\n",
      "Epoch 197/200\n",
      "60000/60000 - 2s - loss: 0.9556 - accuracy: 0.8113 - val_loss: 0.9428 - val_accuracy: 0.8200\n",
      "Epoch 198/200\n",
      "60000/60000 - 2s - loss: 0.9551 - accuracy: 0.8110 - val_loss: 0.9442 - val_accuracy: 0.8208\n",
      "Epoch 199/200\n",
      "60000/60000 - 2s - loss: 0.9558 - accuracy: 0.8106 - val_loss: 0.9431 - val_accuracy: 0.8195\n",
      "Epoch 200/200\n",
      "60000/60000 - 2s - loss: 0.9557 - accuracy: 0.8105 - val_loss: 0.9440 - val_accuracy: 0.8215\n"
     ]
    }
   ],
   "source": [
    "history_linear = model_linear_N16.fit(data_N16.x_train,\n",
    "                          data_N16.y_train,\n",
    "                          epochs=epochs,\n",
    "                          batch_size=batch_size,\n",
    "                          validation_data=(data_N16.x_test, data_N16.y_test),\n",
    "                          verbose=2)\n",
    "\n",
    "# model_linear_N16_vars = [var.numpy() for var in model_linear_N16.variables]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Model: \"sequential_1\"\n",
      "_________________________________________________________________\n",
      "Layer (type)                 Output Shape              Param #   \n",
      "=================================================================\n",
      "rm_2 (RM)                    multiple                  272       \n",
      "_________________________________________________________________\n",
      "rm_3 (RM)                    multiple                  272       \n",
      "_________________________________________________________________\n",
      "activation_1 (Activation)    multiple                  0         \n",
      "_________________________________________________________________\n",
      "lambda_1 (Lambda)            multiple                  0         \n",
      "=================================================================\n",
      "Total params: 544\n",
      "Trainable params: 544\n",
      "Non-trainable params: 0\n",
      "_________________________________________________________________\n"
     ]
    }
   ],
   "source": [
    "model_linear_N16.summary()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Acc = 0.8178\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 288x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "cf_matrix = calc_confusion_matrix_tf(model_linear_N16, data_N16.x_test, data_N16.y_test)\n",
    "fig = plot_confusion_matrix(cf_matrix);"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Training an ONN with the electro-optic activation\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Train on 60000 samples, validate on 10000 samples\n",
      "Epoch 1/200\n",
      "60000/60000 - 5s - loss: 2.8950 - accuracy: 0.1628 - val_loss: 2.0011 - val_accuracy: 0.2943\n",
      "Epoch 2/200\n",
      "60000/60000 - 2s - loss: 1.6967 - accuracy: 0.4252 - val_loss: 1.4306 - val_accuracy: 0.5423\n",
      "Epoch 3/200\n",
      "60000/60000 - 2s - loss: 1.2792 - accuracy: 0.6064 - val_loss: 1.1332 - val_accuracy: 0.6684\n",
      "Epoch 4/200\n",
      "60000/60000 - 2s - loss: 1.0563 - accuracy: 0.6909 - val_loss: 0.9551 - val_accuracy: 0.7355\n",
      "Epoch 5/200\n",
      "60000/60000 - 2s - loss: 0.9136 - accuracy: 0.7365 - val_loss: 0.8494 - val_accuracy: 0.7627\n",
      "Epoch 6/200\n",
      "60000/60000 - 2s - loss: 0.8207 - accuracy: 0.7631 - val_loss: 0.7774 - val_accuracy: 0.7786\n",
      "Epoch 7/200\n",
      "60000/60000 - 2s - loss: 0.7560 - accuracy: 0.7819 - val_loss: 0.7288 - val_accuracy: 0.7952\n",
      "Epoch 8/200\n",
      "60000/60000 - 2s - loss: 0.7095 - accuracy: 0.7954 - val_loss: 0.6924 - val_accuracy: 0.8065\n",
      "Epoch 9/200\n",
      "60000/60000 - 2s - loss: 0.6740 - accuracy: 0.8060 - val_loss: 0.6628 - val_accuracy: 0.8126\n",
      "Epoch 10/200\n",
      "60000/60000 - 2s - loss: 0.6456 - accuracy: 0.8134 - val_loss: 0.6377 - val_accuracy: 0.8184\n",
      "Epoch 11/200\n",
      "60000/60000 - 2s - loss: 0.6222 - accuracy: 0.8195 - val_loss: 0.6168 - val_accuracy: 0.8268\n",
      "Epoch 12/200\n",
      "60000/60000 - 2s - loss: 0.6013 - accuracy: 0.8252 - val_loss: 0.5956 - val_accuracy: 0.8326\n",
      "Epoch 13/200\n",
      "60000/60000 - 2s - loss: 0.5826 - accuracy: 0.8309 - val_loss: 0.5775 - val_accuracy: 0.8355\n",
      "Epoch 14/200\n",
      "60000/60000 - 2s - loss: 0.5664 - accuracy: 0.8353 - val_loss: 0.5615 - val_accuracy: 0.8424\n",
      "Epoch 15/200\n",
      "60000/60000 - 2s - loss: 0.5511 - accuracy: 0.8405 - val_loss: 0.5484 - val_accuracy: 0.8461\n",
      "Epoch 16/200\n",
      "60000/60000 - 2s - loss: 0.5375 - accuracy: 0.8442 - val_loss: 0.5341 - val_accuracy: 0.8501\n",
      "Epoch 17/200\n",
      "60000/60000 - 2s - loss: 0.5255 - accuracy: 0.8479 - val_loss: 0.5233 - val_accuracy: 0.8522\n",
      "Epoch 18/200\n",
      "60000/60000 - 2s - loss: 0.5153 - accuracy: 0.8512 - val_loss: 0.5165 - val_accuracy: 0.8540\n",
      "Epoch 19/200\n",
      "60000/60000 - 2s - loss: 0.5059 - accuracy: 0.8533 - val_loss: 0.5060 - val_accuracy: 0.8570\n",
      "Epoch 20/200\n",
      "60000/60000 - 2s - loss: 0.4985 - accuracy: 0.8555 - val_loss: 0.4946 - val_accuracy: 0.8598\n",
      "Epoch 21/200\n",
      "60000/60000 - 2s - loss: 0.4905 - accuracy: 0.8571 - val_loss: 0.4883 - val_accuracy: 0.8614\n",
      "Epoch 22/200\n",
      "60000/60000 - 2s - loss: 0.4841 - accuracy: 0.8601 - val_loss: 0.4809 - val_accuracy: 0.8638\n",
      "Epoch 23/200\n",
      "60000/60000 - 2s - loss: 0.4785 - accuracy: 0.8612 - val_loss: 0.4749 - val_accuracy: 0.8662\n",
      "Epoch 24/200\n",
      "60000/60000 - 2s - loss: 0.4724 - accuracy: 0.8623 - val_loss: 0.4699 - val_accuracy: 0.8677\n",
      "Epoch 25/200\n",
      "60000/60000 - 2s - loss: 0.4679 - accuracy: 0.8626 - val_loss: 0.4662 - val_accuracy: 0.8666\n",
      "Epoch 26/200\n",
      "60000/60000 - 2s - loss: 0.4632 - accuracy: 0.8644 - val_loss: 0.4603 - val_accuracy: 0.8698\n",
      "Epoch 27/200\n",
      "60000/60000 - 2s - loss: 0.4591 - accuracy: 0.8661 - val_loss: 0.4566 - val_accuracy: 0.8696\n",
      "Epoch 28/200\n",
      "60000/60000 - 2s - loss: 0.4550 - accuracy: 0.8674 - val_loss: 0.4518 - val_accuracy: 0.8710\n",
      "Epoch 29/200\n",
      "60000/60000 - 2s - loss: 0.4514 - accuracy: 0.8686 - val_loss: 0.4499 - val_accuracy: 0.8709\n",
      "Epoch 30/200\n",
      "60000/60000 - 2s - loss: 0.4476 - accuracy: 0.8684 - val_loss: 0.4448 - val_accuracy: 0.8719\n",
      "Epoch 31/200\n",
      "60000/60000 - 2s - loss: 0.4441 - accuracy: 0.8700 - val_loss: 0.4412 - val_accuracy: 0.8721\n",
      "Epoch 32/200\n",
      "60000/60000 - 2s - loss: 0.4411 - accuracy: 0.8707 - val_loss: 0.4375 - val_accuracy: 0.8748\n",
      "Epoch 33/200\n",
      "60000/60000 - 2s - loss: 0.4383 - accuracy: 0.8710 - val_loss: 0.4337 - val_accuracy: 0.8753\n",
      "Epoch 34/200\n",
      "60000/60000 - 2s - loss: 0.4343 - accuracy: 0.8729 - val_loss: 0.4300 - val_accuracy: 0.8761\n",
      "Epoch 35/200\n",
      "60000/60000 - 2s - loss: 0.4311 - accuracy: 0.8730 - val_loss: 0.4256 - val_accuracy: 0.8766\n",
      "Epoch 36/200\n",
      "60000/60000 - 2s - loss: 0.4280 - accuracy: 0.8745 - val_loss: 0.4239 - val_accuracy: 0.8760\n",
      "Epoch 37/200\n",
      "60000/60000 - 2s - loss: 0.4257 - accuracy: 0.8736 - val_loss: 0.4203 - val_accuracy: 0.8776\n",
      "Epoch 38/200\n",
      "60000/60000 - 2s - loss: 0.4225 - accuracy: 0.8753 - val_loss: 0.4156 - val_accuracy: 0.8779\n",
      "Epoch 39/200\n",
      "60000/60000 - 2s - loss: 0.4199 - accuracy: 0.8766 - val_loss: 0.4141 - val_accuracy: 0.8775\n",
      "Epoch 40/200\n",
      "60000/60000 - 2s - loss: 0.4169 - accuracy: 0.8766 - val_loss: 0.4134 - val_accuracy: 0.8778\n",
      "Epoch 41/200\n",
      "60000/60000 - 2s - loss: 0.4149 - accuracy: 0.8779 - val_loss: 0.4089 - val_accuracy: 0.8811\n",
      "Epoch 42/200\n",
      "60000/60000 - 2s - loss: 0.4114 - accuracy: 0.8781 - val_loss: 0.4065 - val_accuracy: 0.8800\n",
      "Epoch 43/200\n",
      "60000/60000 - 2s - loss: 0.4094 - accuracy: 0.8791 - val_loss: 0.4048 - val_accuracy: 0.8814\n",
      "Epoch 44/200\n",
      "60000/60000 - 2s - loss: 0.4069 - accuracy: 0.8795 - val_loss: 0.4016 - val_accuracy: 0.8819\n",
      "Epoch 45/200\n",
      "60000/60000 - 2s - loss: 0.4050 - accuracy: 0.8810 - val_loss: 0.3994 - val_accuracy: 0.8822\n",
      "Epoch 46/200\n",
      "60000/60000 - 2s - loss: 0.4027 - accuracy: 0.8803 - val_loss: 0.3958 - val_accuracy: 0.8827\n",
      "Epoch 47/200\n",
      "60000/60000 - 2s - loss: 0.4006 - accuracy: 0.8822 - val_loss: 0.3959 - val_accuracy: 0.8834\n",
      "Epoch 48/200\n",
      "60000/60000 - 2s - loss: 0.3989 - accuracy: 0.8828 - val_loss: 0.3933 - val_accuracy: 0.8835\n",
      "Epoch 49/200\n",
      "60000/60000 - 2s - loss: 0.3965 - accuracy: 0.8825 - val_loss: 0.3929 - val_accuracy: 0.8824\n",
      "Epoch 50/200\n",
      "60000/60000 - 2s - loss: 0.3949 - accuracy: 0.8836 - val_loss: 0.3919 - val_accuracy: 0.8846\n",
      "Epoch 51/200\n",
      "60000/60000 - 2s - loss: 0.3931 - accuracy: 0.8832 - val_loss: 0.3897 - val_accuracy: 0.8842\n",
      "Epoch 52/200\n",
      "60000/60000 - 2s - loss: 0.3909 - accuracy: 0.8839 - val_loss: 0.3874 - val_accuracy: 0.8862\n",
      "Epoch 53/200\n",
      "60000/60000 - 2s - loss: 0.3894 - accuracy: 0.8856 - val_loss: 0.3866 - val_accuracy: 0.8872\n",
      "Epoch 54/200\n",
      "60000/60000 - 2s - loss: 0.3882 - accuracy: 0.8855 - val_loss: 0.3841 - val_accuracy: 0.8854\n",
      "Epoch 55/200\n",
      "60000/60000 - 2s - loss: 0.3867 - accuracy: 0.8859 - val_loss: 0.3837 - val_accuracy: 0.8890\n",
      "Epoch 56/200\n",
      "60000/60000 - 2s - loss: 0.3851 - accuracy: 0.8856 - val_loss: 0.3803 - val_accuracy: 0.8875\n",
      "Epoch 57/200\n",
      "60000/60000 - 2s - loss: 0.3833 - accuracy: 0.8857 - val_loss: 0.3831 - val_accuracy: 0.8867\n",
      "Epoch 58/200\n",
      "60000/60000 - 2s - loss: 0.3823 - accuracy: 0.8863 - val_loss: 0.3809 - val_accuracy: 0.8870\n",
      "Epoch 59/200\n",
      "60000/60000 - 2s - loss: 0.3810 - accuracy: 0.8867 - val_loss: 0.3776 - val_accuracy: 0.8887\n",
      "Epoch 60/200\n",
      "60000/60000 - 2s - loss: 0.3796 - accuracy: 0.8870 - val_loss: 0.3747 - val_accuracy: 0.8889\n",
      "Epoch 61/200\n",
      "60000/60000 - 2s - loss: 0.3784 - accuracy: 0.8870 - val_loss: 0.3755 - val_accuracy: 0.8879\n",
      "Epoch 62/200\n",
      "60000/60000 - 2s - loss: 0.3770 - accuracy: 0.8877 - val_loss: 0.3748 - val_accuracy: 0.8896\n",
      "Epoch 63/200\n",
      "60000/60000 - 2s - loss: 0.3754 - accuracy: 0.8885 - val_loss: 0.3756 - val_accuracy: 0.8877\n",
      "Epoch 64/200\n",
      "60000/60000 - 2s - loss: 0.3738 - accuracy: 0.8881 - val_loss: 0.3723 - val_accuracy: 0.8920\n",
      "Epoch 65/200\n",
      "60000/60000 - 2s - loss: 0.3739 - accuracy: 0.8885 - val_loss: 0.3705 - val_accuracy: 0.8915\n",
      "Epoch 66/200\n",
      "60000/60000 - 2s - loss: 0.3719 - accuracy: 0.8891 - val_loss: 0.3700 - val_accuracy: 0.8911\n",
      "Epoch 67/200\n",
      "60000/60000 - 2s - loss: 0.3708 - accuracy: 0.8897 - val_loss: 0.3699 - val_accuracy: 0.8916\n",
      "Epoch 68/200\n",
      "60000/60000 - 2s - loss: 0.3692 - accuracy: 0.8904 - val_loss: 0.3681 - val_accuracy: 0.8925\n",
      "Epoch 69/200\n",
      "60000/60000 - 2s - loss: 0.3685 - accuracy: 0.8907 - val_loss: 0.3667 - val_accuracy: 0.8928\n",
      "Epoch 70/200\n",
      "60000/60000 - 2s - loss: 0.3684 - accuracy: 0.8907 - val_loss: 0.3654 - val_accuracy: 0.8934\n",
      "Epoch 71/200\n",
      "60000/60000 - 2s - loss: 0.3667 - accuracy: 0.8915 - val_loss: 0.3642 - val_accuracy: 0.8927\n",
      "Epoch 72/200\n",
      "60000/60000 - 2s - loss: 0.3651 - accuracy: 0.8913 - val_loss: 0.3667 - val_accuracy: 0.8924\n",
      "Epoch 73/200\n",
      "60000/60000 - 2s - loss: 0.3650 - accuracy: 0.8913 - val_loss: 0.3648 - val_accuracy: 0.8935\n",
      "Epoch 74/200\n",
      "60000/60000 - 2s - loss: 0.3633 - accuracy: 0.8918 - val_loss: 0.3634 - val_accuracy: 0.8938\n",
      "Epoch 75/200\n",
      "60000/60000 - 2s - loss: 0.3621 - accuracy: 0.8924 - val_loss: 0.3618 - val_accuracy: 0.8949\n",
      "Epoch 76/200\n",
      "60000/60000 - 2s - loss: 0.3612 - accuracy: 0.8922 - val_loss: 0.3626 - val_accuracy: 0.8949\n",
      "Epoch 77/200\n",
      "60000/60000 - 2s - loss: 0.3610 - accuracy: 0.8920 - val_loss: 0.3623 - val_accuracy: 0.8944\n",
      "Epoch 78/200\n",
      "60000/60000 - 2s - loss: 0.3604 - accuracy: 0.8925 - val_loss: 0.3596 - val_accuracy: 0.8939\n",
      "Epoch 79/200\n",
      "60000/60000 - 2s - loss: 0.3592 - accuracy: 0.8933 - val_loss: 0.3588 - val_accuracy: 0.8958\n",
      "Epoch 80/200\n",
      "60000/60000 - 2s - loss: 0.3580 - accuracy: 0.8933 - val_loss: 0.3592 - val_accuracy: 0.8944\n",
      "Epoch 81/200\n",
      "60000/60000 - 2s - loss: 0.3573 - accuracy: 0.8938 - val_loss: 0.3567 - val_accuracy: 0.8962\n",
      "Epoch 82/200\n",
      "60000/60000 - 2s - loss: 0.3564 - accuracy: 0.8944 - val_loss: 0.3575 - val_accuracy: 0.8958\n",
      "Epoch 83/200\n",
      "60000/60000 - 2s - loss: 0.3559 - accuracy: 0.8950 - val_loss: 0.3584 - val_accuracy: 0.8950\n",
      "Epoch 84/200\n",
      "60000/60000 - 2s - loss: 0.3550 - accuracy: 0.8945 - val_loss: 0.3564 - val_accuracy: 0.8957\n",
      "Epoch 85/200\n",
      "60000/60000 - 2s - loss: 0.3542 - accuracy: 0.8945 - val_loss: 0.3564 - val_accuracy: 0.8963\n",
      "Epoch 86/200\n",
      "60000/60000 - 2s - loss: 0.3534 - accuracy: 0.8950 - val_loss: 0.3529 - val_accuracy: 0.8972\n",
      "Epoch 87/200\n",
      "60000/60000 - 2s - loss: 0.3525 - accuracy: 0.8958 - val_loss: 0.3533 - val_accuracy: 0.8978\n",
      "Epoch 88/200\n",
      "60000/60000 - 2s - loss: 0.3527 - accuracy: 0.8954 - val_loss: 0.3518 - val_accuracy: 0.8976\n",
      "Epoch 89/200\n",
      "60000/60000 - 2s - loss: 0.3506 - accuracy: 0.8962 - val_loss: 0.3537 - val_accuracy: 0.8983\n",
      "Epoch 90/200\n",
      "60000/60000 - 2s - loss: 0.3500 - accuracy: 0.8961 - val_loss: 0.3547 - val_accuracy: 0.8953\n",
      "Epoch 91/200\n",
      "60000/60000 - 2s - loss: 0.3492 - accuracy: 0.8960 - val_loss: 0.3532 - val_accuracy: 0.8954\n",
      "Epoch 92/200\n",
      "60000/60000 - 2s - loss: 0.3497 - accuracy: 0.8966 - val_loss: 0.3514 - val_accuracy: 0.8963\n",
      "Epoch 93/200\n",
      "60000/60000 - 2s - loss: 0.3489 - accuracy: 0.8961 - val_loss: 0.3504 - val_accuracy: 0.8992\n",
      "Epoch 94/200\n",
      "60000/60000 - 2s - loss: 0.3478 - accuracy: 0.8964 - val_loss: 0.3496 - val_accuracy: 0.8976\n",
      "Epoch 95/200\n",
      "60000/60000 - 2s - loss: 0.3476 - accuracy: 0.8967 - val_loss: 0.3482 - val_accuracy: 0.8971\n",
      "Epoch 96/200\n",
      "60000/60000 - 2s - loss: 0.3464 - accuracy: 0.8971 - val_loss: 0.3495 - val_accuracy: 0.8989\n",
      "Epoch 97/200\n",
      "60000/60000 - 2s - loss: 0.3463 - accuracy: 0.8974 - val_loss: 0.3492 - val_accuracy: 0.8973\n",
      "Epoch 98/200\n",
      "60000/60000 - 2s - loss: 0.3456 - accuracy: 0.8975 - val_loss: 0.3467 - val_accuracy: 0.8989\n",
      "Epoch 99/200\n",
      "60000/60000 - 2s - loss: 0.3446 - accuracy: 0.8977 - val_loss: 0.3485 - val_accuracy: 0.9000\n",
      "Epoch 100/200\n",
      "60000/60000 - 2s - loss: 0.3440 - accuracy: 0.8983 - val_loss: 0.3493 - val_accuracy: 0.8967\n",
      "Epoch 101/200\n",
      "60000/60000 - 2s - loss: 0.3436 - accuracy: 0.8984 - val_loss: 0.3449 - val_accuracy: 0.8999\n",
      "Epoch 102/200\n",
      "60000/60000 - 2s - loss: 0.3432 - accuracy: 0.8984 - val_loss: 0.3471 - val_accuracy: 0.9014\n",
      "Epoch 103/200\n",
      "60000/60000 - 2s - loss: 0.3428 - accuracy: 0.8986 - val_loss: 0.3444 - val_accuracy: 0.8997\n",
      "Epoch 104/200\n",
      "60000/60000 - 2s - loss: 0.3418 - accuracy: 0.8994 - val_loss: 0.3447 - val_accuracy: 0.9015\n",
      "Epoch 105/200\n",
      "60000/60000 - 2s - loss: 0.3413 - accuracy: 0.8984 - val_loss: 0.3433 - val_accuracy: 0.9005\n",
      "Epoch 106/200\n",
      "60000/60000 - 2s - loss: 0.3411 - accuracy: 0.8990 - val_loss: 0.3448 - val_accuracy: 0.9010\n",
      "Epoch 107/200\n",
      "60000/60000 - 2s - loss: 0.3403 - accuracy: 0.8985 - val_loss: 0.3424 - val_accuracy: 0.9012\n",
      "Epoch 108/200\n",
      "60000/60000 - 2s - loss: 0.3400 - accuracy: 0.8992 - val_loss: 0.3440 - val_accuracy: 0.8999\n",
      "Epoch 109/200\n",
      "60000/60000 - 2s - loss: 0.3400 - accuracy: 0.8997 - val_loss: 0.3421 - val_accuracy: 0.9022\n",
      "Epoch 110/200\n",
      "60000/60000 - 2s - loss: 0.3391 - accuracy: 0.9000 - val_loss: 0.3426 - val_accuracy: 0.9006\n",
      "Epoch 111/200\n",
      "60000/60000 - 2s - loss: 0.3390 - accuracy: 0.8999 - val_loss: 0.3410 - val_accuracy: 0.9006\n",
      "Epoch 112/200\n",
      "60000/60000 - 2s - loss: 0.3386 - accuracy: 0.9004 - val_loss: 0.3416 - val_accuracy: 0.9017\n",
      "Epoch 113/200\n",
      "60000/60000 - 2s - loss: 0.3381 - accuracy: 0.9006 - val_loss: 0.3411 - val_accuracy: 0.9005\n",
      "Epoch 114/200\n",
      "60000/60000 - 2s - loss: 0.3380 - accuracy: 0.8995 - val_loss: 0.3398 - val_accuracy: 0.9021\n",
      "Epoch 115/200\n",
      "60000/60000 - 2s - loss: 0.3371 - accuracy: 0.8999 - val_loss: 0.3374 - val_accuracy: 0.9022\n",
      "Epoch 116/200\n",
      "60000/60000 - 2s - loss: 0.3362 - accuracy: 0.9001 - val_loss: 0.3388 - val_accuracy: 0.9010\n",
      "Epoch 117/200\n",
      "60000/60000 - 2s - loss: 0.3369 - accuracy: 0.9003 - val_loss: 0.3389 - val_accuracy: 0.9014\n",
      "Epoch 118/200\n",
      "60000/60000 - 2s - loss: 0.3353 - accuracy: 0.9009 - val_loss: 0.3388 - val_accuracy: 0.9025\n",
      "Epoch 119/200\n",
      "60000/60000 - 2s - loss: 0.3354 - accuracy: 0.9011 - val_loss: 0.3384 - val_accuracy: 0.9030\n",
      "Epoch 120/200\n",
      "60000/60000 - 2s - loss: 0.3356 - accuracy: 0.9008 - val_loss: 0.3375 - val_accuracy: 0.9016\n",
      "Epoch 121/200\n",
      "60000/60000 - 2s - loss: 0.3349 - accuracy: 0.9012 - val_loss: 0.3396 - val_accuracy: 0.8995\n",
      "Epoch 122/200\n",
      "60000/60000 - 2s - loss: 0.3344 - accuracy: 0.9010 - val_loss: 0.3363 - val_accuracy: 0.9029\n",
      "Epoch 123/200\n",
      "60000/60000 - 2s - loss: 0.3334 - accuracy: 0.9010 - val_loss: 0.3387 - val_accuracy: 0.9005\n",
      "Epoch 124/200\n",
      "60000/60000 - 2s - loss: 0.3344 - accuracy: 0.9015 - val_loss: 0.3366 - val_accuracy: 0.9022\n",
      "Epoch 125/200\n",
      "60000/60000 - 2s - loss: 0.3326 - accuracy: 0.9021 - val_loss: 0.3366 - val_accuracy: 0.9014\n",
      "Epoch 126/200\n",
      "60000/60000 - 2s - loss: 0.3323 - accuracy: 0.9014 - val_loss: 0.3344 - val_accuracy: 0.9033\n",
      "Epoch 127/200\n",
      "60000/60000 - 2s - loss: 0.3320 - accuracy: 0.9020 - val_loss: 0.3345 - val_accuracy: 0.9030\n",
      "Epoch 128/200\n",
      "60000/60000 - 2s - loss: 0.3323 - accuracy: 0.9022 - val_loss: 0.3355 - val_accuracy: 0.9014\n",
      "Epoch 129/200\n",
      "60000/60000 - 2s - loss: 0.3316 - accuracy: 0.9024 - val_loss: 0.3342 - val_accuracy: 0.9013\n",
      "Epoch 130/200\n",
      "60000/60000 - 2s - loss: 0.3317 - accuracy: 0.9018 - val_loss: 0.3342 - val_accuracy: 0.9021\n",
      "Epoch 131/200\n",
      "60000/60000 - 2s - loss: 0.3309 - accuracy: 0.9028 - val_loss: 0.3316 - val_accuracy: 0.9032\n",
      "Epoch 132/200\n",
      "60000/60000 - 2s - loss: 0.3302 - accuracy: 0.9028 - val_loss: 0.3348 - val_accuracy: 0.9022\n",
      "Epoch 133/200\n",
      "60000/60000 - 2s - loss: 0.3305 - accuracy: 0.9025 - val_loss: 0.3330 - val_accuracy: 0.9023\n",
      "Epoch 134/200\n",
      "60000/60000 - 2s - loss: 0.3298 - accuracy: 0.9022 - val_loss: 0.3326 - val_accuracy: 0.9018\n",
      "Epoch 135/200\n",
      "60000/60000 - 2s - loss: 0.3305 - accuracy: 0.9018 - val_loss: 0.3347 - val_accuracy: 0.9039\n",
      "Epoch 136/200\n",
      "60000/60000 - 2s - loss: 0.3292 - accuracy: 0.9030 - val_loss: 0.3340 - val_accuracy: 0.9025\n",
      "Epoch 137/200\n",
      "60000/60000 - 2s - loss: 0.3298 - accuracy: 0.9030 - val_loss: 0.3308 - val_accuracy: 0.9027\n",
      "Epoch 138/200\n",
      "60000/60000 - 2s - loss: 0.3284 - accuracy: 0.9035 - val_loss: 0.3317 - val_accuracy: 0.9035\n",
      "Epoch 139/200\n",
      "60000/60000 - 2s - loss: 0.3288 - accuracy: 0.9030 - val_loss: 0.3307 - val_accuracy: 0.9019\n",
      "Epoch 140/200\n",
      "60000/60000 - 2s - loss: 0.3286 - accuracy: 0.9032 - val_loss: 0.3304 - val_accuracy: 0.9017\n",
      "Epoch 141/200\n",
      "60000/60000 - 2s - loss: 0.3278 - accuracy: 0.9028 - val_loss: 0.3305 - val_accuracy: 0.9024\n",
      "Epoch 142/200\n",
      "60000/60000 - 2s - loss: 0.3284 - accuracy: 0.9034 - val_loss: 0.3376 - val_accuracy: 0.9000\n",
      "Epoch 143/200\n",
      "60000/60000 - 2s - loss: 0.3281 - accuracy: 0.9025 - val_loss: 0.3305 - val_accuracy: 0.9031\n",
      "Epoch 144/200\n",
      "60000/60000 - 2s - loss: 0.3276 - accuracy: 0.9029 - val_loss: 0.3336 - val_accuracy: 0.9029\n",
      "Epoch 145/200\n",
      "60000/60000 - 2s - loss: 0.3274 - accuracy: 0.9030 - val_loss: 0.3304 - val_accuracy: 0.9033\n",
      "Epoch 146/200\n",
      "60000/60000 - 2s - loss: 0.3267 - accuracy: 0.9034 - val_loss: 0.3287 - val_accuracy: 0.9040\n",
      "Epoch 147/200\n",
      "60000/60000 - 2s - loss: 0.3266 - accuracy: 0.9029 - val_loss: 0.3298 - val_accuracy: 0.9034\n",
      "Epoch 148/200\n",
      "60000/60000 - 2s - loss: 0.3266 - accuracy: 0.9034 - val_loss: 0.3285 - val_accuracy: 0.9045\n",
      "Epoch 149/200\n",
      "60000/60000 - 2s - loss: 0.3263 - accuracy: 0.9038 - val_loss: 0.3274 - val_accuracy: 0.9026\n",
      "Epoch 150/200\n",
      "60000/60000 - 2s - loss: 0.3258 - accuracy: 0.9039 - val_loss: 0.3304 - val_accuracy: 0.9021\n",
      "Epoch 151/200\n",
      "60000/60000 - 2s - loss: 0.3257 - accuracy: 0.9039 - val_loss: 0.3286 - val_accuracy: 0.9034\n",
      "Epoch 152/200\n",
      "60000/60000 - 2s - loss: 0.3259 - accuracy: 0.9037 - val_loss: 0.3293 - val_accuracy: 0.9028\n",
      "Epoch 153/200\n",
      "60000/60000 - 2s - loss: 0.3261 - accuracy: 0.9032 - val_loss: 0.3280 - val_accuracy: 0.9043\n",
      "Epoch 154/200\n",
      "60000/60000 - 2s - loss: 0.3252 - accuracy: 0.9045 - val_loss: 0.3294 - val_accuracy: 0.9023\n",
      "Epoch 155/200\n",
      "60000/60000 - 2s - loss: 0.3255 - accuracy: 0.9035 - val_loss: 0.3263 - val_accuracy: 0.9040\n",
      "Epoch 156/200\n",
      "60000/60000 - 2s - loss: 0.3244 - accuracy: 0.9038 - val_loss: 0.3268 - val_accuracy: 0.9035\n",
      "Epoch 157/200\n",
      "60000/60000 - 2s - loss: 0.3244 - accuracy: 0.9044 - val_loss: 0.3271 - val_accuracy: 0.9042\n",
      "Epoch 158/200\n",
      "60000/60000 - 2s - loss: 0.3243 - accuracy: 0.9046 - val_loss: 0.3275 - val_accuracy: 0.9028\n",
      "Epoch 159/200\n",
      "60000/60000 - 2s - loss: 0.3242 - accuracy: 0.9041 - val_loss: 0.3279 - val_accuracy: 0.9026\n",
      "Epoch 160/200\n",
      "60000/60000 - 2s - loss: 0.3235 - accuracy: 0.9042 - val_loss: 0.3272 - val_accuracy: 0.9036\n",
      "Epoch 161/200\n",
      "60000/60000 - 2s - loss: 0.3236 - accuracy: 0.9043 - val_loss: 0.3261 - val_accuracy: 0.9030\n",
      "Epoch 162/200\n",
      "60000/60000 - 2s - loss: 0.3241 - accuracy: 0.9035 - val_loss: 0.3287 - val_accuracy: 0.9022\n",
      "Epoch 163/200\n",
      "60000/60000 - 2s - loss: 0.3237 - accuracy: 0.9039 - val_loss: 0.3273 - val_accuracy: 0.9018\n",
      "Epoch 164/200\n",
      "60000/60000 - 2s - loss: 0.3232 - accuracy: 0.9045 - val_loss: 0.3253 - val_accuracy: 0.9036\n",
      "Epoch 165/200\n",
      "60000/60000 - 2s - loss: 0.3236 - accuracy: 0.9043 - val_loss: 0.3267 - val_accuracy: 0.9036\n",
      "Epoch 166/200\n",
      "60000/60000 - 2s - loss: 0.3221 - accuracy: 0.9044 - val_loss: 0.3262 - val_accuracy: 0.9041\n",
      "Epoch 167/200\n",
      "60000/60000 - 2s - loss: 0.3233 - accuracy: 0.9044 - val_loss: 0.3256 - val_accuracy: 0.9025\n",
      "Epoch 168/200\n",
      "60000/60000 - 2s - loss: 0.3228 - accuracy: 0.9042 - val_loss: 0.3227 - val_accuracy: 0.9044\n",
      "Epoch 169/200\n",
      "60000/60000 - 2s - loss: 0.3221 - accuracy: 0.9049 - val_loss: 0.3251 - val_accuracy: 0.9048\n",
      "Epoch 170/200\n",
      "60000/60000 - 2s - loss: 0.3224 - accuracy: 0.9051 - val_loss: 0.3238 - val_accuracy: 0.9034\n",
      "Epoch 171/200\n",
      "60000/60000 - 2s - loss: 0.3219 - accuracy: 0.9052 - val_loss: 0.3243 - val_accuracy: 0.9036\n",
      "Epoch 172/200\n",
      "60000/60000 - 2s - loss: 0.3214 - accuracy: 0.9046 - val_loss: 0.3251 - val_accuracy: 0.9039\n",
      "Epoch 173/200\n",
      "60000/60000 - 2s - loss: 0.3218 - accuracy: 0.9045 - val_loss: 0.3258 - val_accuracy: 0.9030\n",
      "Epoch 174/200\n",
      "60000/60000 - 2s - loss: 0.3223 - accuracy: 0.9045 - val_loss: 0.3230 - val_accuracy: 0.9042\n",
      "Epoch 175/200\n",
      "60000/60000 - 2s - loss: 0.3216 - accuracy: 0.9047 - val_loss: 0.3233 - val_accuracy: 0.9045\n",
      "Epoch 176/200\n",
      "60000/60000 - 2s - loss: 0.3212 - accuracy: 0.9050 - val_loss: 0.3236 - val_accuracy: 0.9041\n",
      "Epoch 177/200\n",
      "60000/60000 - 2s - loss: 0.3220 - accuracy: 0.9041 - val_loss: 0.3239 - val_accuracy: 0.9047\n",
      "Epoch 178/200\n",
      "60000/60000 - 2s - loss: 0.3208 - accuracy: 0.9049 - val_loss: 0.3221 - val_accuracy: 0.9053\n",
      "Epoch 179/200\n",
      "60000/60000 - 2s - loss: 0.3206 - accuracy: 0.9052 - val_loss: 0.3268 - val_accuracy: 0.9020\n",
      "Epoch 180/200\n",
      "60000/60000 - 2s - loss: 0.3220 - accuracy: 0.9039 - val_loss: 0.3242 - val_accuracy: 0.9039\n",
      "Epoch 181/200\n",
      "60000/60000 - 2s - loss: 0.3199 - accuracy: 0.9055 - val_loss: 0.3230 - val_accuracy: 0.9037\n",
      "Epoch 182/200\n",
      "60000/60000 - 2s - loss: 0.3207 - accuracy: 0.9043 - val_loss: 0.3217 - val_accuracy: 0.9046\n",
      "Epoch 183/200\n",
      "60000/60000 - 2s - loss: 0.3201 - accuracy: 0.9052 - val_loss: 0.3208 - val_accuracy: 0.9044\n",
      "Epoch 184/200\n",
      "60000/60000 - 2s - loss: 0.3197 - accuracy: 0.9048 - val_loss: 0.3203 - val_accuracy: 0.9048\n",
      "Epoch 185/200\n",
      "60000/60000 - 2s - loss: 0.3197 - accuracy: 0.9049 - val_loss: 0.3241 - val_accuracy: 0.9049\n",
      "Epoch 186/200\n",
      "60000/60000 - 2s - loss: 0.3202 - accuracy: 0.9052 - val_loss: 0.3218 - val_accuracy: 0.9045\n",
      "Epoch 187/200\n",
      "60000/60000 - 2s - loss: 0.3195 - accuracy: 0.9047 - val_loss: 0.3216 - val_accuracy: 0.9057\n",
      "Epoch 188/200\n",
      "60000/60000 - 2s - loss: 0.3195 - accuracy: 0.9049 - val_loss: 0.3211 - val_accuracy: 0.9047\n",
      "Epoch 189/200\n",
      "60000/60000 - 2s - loss: 0.3195 - accuracy: 0.9050 - val_loss: 0.3221 - val_accuracy: 0.9062\n",
      "Epoch 190/200\n",
      "60000/60000 - 2s - loss: 0.3191 - accuracy: 0.9048 - val_loss: 0.3223 - val_accuracy: 0.9043\n",
      "Epoch 191/200\n",
      "60000/60000 - 2s - loss: 0.3193 - accuracy: 0.9049 - val_loss: 0.3197 - val_accuracy: 0.9059\n",
      "Epoch 192/200\n",
      "60000/60000 - 2s - loss: 0.3188 - accuracy: 0.9054 - val_loss: 0.3200 - val_accuracy: 0.9057\n",
      "Epoch 193/200\n",
      "60000/60000 - 2s - loss: 0.3184 - accuracy: 0.9058 - val_loss: 0.3221 - val_accuracy: 0.9054\n",
      "Epoch 194/200\n",
      "60000/60000 - 2s - loss: 0.3191 - accuracy: 0.9052 - val_loss: 0.3274 - val_accuracy: 0.9017\n",
      "Epoch 195/200\n",
      "60000/60000 - 2s - loss: 0.3191 - accuracy: 0.9046 - val_loss: 0.3221 - val_accuracy: 0.9028\n",
      "Epoch 196/200\n",
      "60000/60000 - 2s - loss: 0.3180 - accuracy: 0.9059 - val_loss: 0.3231 - val_accuracy: 0.9037\n",
      "Epoch 197/200\n",
      "60000/60000 - 2s - loss: 0.3180 - accuracy: 0.9056 - val_loss: 0.3205 - val_accuracy: 0.9046\n",
      "Epoch 198/200\n",
      "60000/60000 - 2s - loss: 0.3182 - accuracy: 0.9053 - val_loss: 0.3279 - val_accuracy: 0.9027\n",
      "Epoch 199/200\n",
      "60000/60000 - 2s - loss: 0.3188 - accuracy: 0.9058 - val_loss: 0.3198 - val_accuracy: 0.9057\n",
      "Epoch 200/200\n",
      "60000/60000 - 2s - loss: 0.3186 - accuracy: 0.9053 - val_loss: 0.3184 - val_accuracy: 0.9066\n"
     ]
    }
   ],
   "source": [
    "history_eo = model_eo_N16.fit(data_N16.x_train,\n",
    "                          data_N16.y_train,\n",
    "                          epochs=epochs,\n",
    "                          batch_size=batch_size,\n",
    "                          validation_data=(data_N16.x_test, data_N16.y_test),\n",
    "                          verbose=2)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Acc = 0.9053\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 288x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "cf_matrix = calc_confusion_matrix_tf(model_eo_N16, data_N16.x_test, data_N16.y_test)\n",
    "fig = plot_confusion_matrix(cf_matrix);"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Compare the training history of the two models:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 252x252 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "plot_model_comparison([history_linear, history_eo], \n",
    "                          ['without activation', 'with activation'],\n",
    "                          figsize=(3.5,3.5),\n",
    "                          axs=None,\n",
    "                          ylim_acc=[40,100],\n",
    "                          height_ratios=[1,0.33]);"
   ]
  }
 ],
 "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": "3.7.3"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}