VAE SWAT.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "2.2.0\n",
      "0.10.0\n"
     ]
    }
   ],
   "source": [
    "from __future__ import absolute_import\n",
    "from __future__ import division\n",
    "from __future__ import print_function\n",
    "\n",
    "import pandas as pd\n",
    "import numpy as np\n",
    "import os\n",
    "\n",
    "import tensorflow as tf\n",
    "from tensorflow.python import tf2\n",
    "if not tf2.enabled():\n",
    "    import tensorflow.compat.v2 as tf\n",
    "    tf.enable_v2_behavior()\n",
    "    assert tf2.enabled()\n",
    "import tensorflow_probability as tfp\n",
    "\n",
    "tfk = tf.keras\n",
    "tfkl = tf.keras.layers\n",
    "tfpl = tfp.layers\n",
    "tfd = tfp.distributions\n",
    "\n",
    "from tensorflow.keras.utils import plot_model\n",
    "from tensorflow.keras.callbacks import EarlyStopping, ModelCheckpoint\n",
    "\n",
    "import matplotlib.pyplot as plt\n",
    "\n",
    "from sklearn.model_selection import train_test_split \n",
    "from sklearn.linear_model import LogisticRegression\n",
    "from sklearn.metrics import classification_report, accuracy_score, roc_auc_score, roc_curve\n",
    "from sklearn.preprocessing import MinMaxScaler\n",
    "from sklearn.manifold import TSNE\n",
    "\n",
    "%matplotlib inline\n",
    "np.random.seed(0)\n",
    "tf.random.set_seed(0)\n",
    "print(tf.__version__)\n",
    "print(tfp.__version__)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "### Utility Functions\n",
    "## Plots\n",
    "def tsne_plot(x1, y1, name=\"graph.png\"):\n",
    "    tsne = TSNE(n_components=2, random_state=0)\n",
    "    X_t = tsne.fit_transform(x1)\n",
    "\n",
    "    # plt.figure(figsize=(12, 8))\n",
    "    plt.scatter(X_t[np.where(y1 == 0), 0], X_t[np.where(y1 == 0), 1], marker='o', color='g', linewidth='1', alpha=0.8, label='Normal', s=2)\n",
    "    plt.scatter(X_t[np.where(y1 == 1), 0], X_t[np.where(y1 == 1), 1], marker='o', color='r', linewidth='1', alpha=0.8, label='Attack', s=2)\n",
    "\n",
    "    plt.legend(loc='best');\n",
    "    plt.savefig(name);\n",
    "    plt.show();\n",
    "    \n",
    "# Plot Keras training history\n",
    "def plot_loss(hist):\n",
    "    plt.plot(hist.history['loss'])\n",
    "    plt.plot(hist.history['val_loss'])\n",
    "    plt.title('model loss')\n",
    "    plt.ylabel('loss')\n",
    "    plt.xlabel('epoch')\n",
    "    plt.legend(['train', 'test'], loc='upper left')\n",
    "    plt.yscale('log',basey=10)\n",
    "    plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "df = pd.read_excel(\"C:/Users/Ruben/Datasets/SWAT/SWaT_Dataset_Attack_v0.xlsx\")\n",
    "\n",
    "df.columns = [\n",
    "\t'TIMESTAMP','FIT101','LIT101','MV101','P101','P102','AIT201','AIT202','AIT203','FIT201','MV201','P201','P202','P203',\n",
    "    'P204','P205','P206','DPIT301','FIT301','LIT301','MV301','MV302','MV303','MV304','P301','P302','AIT401','AIT402','FIT401',\n",
    "    'LIT401','P401','P402','P403','P404','UV401','AIT501','AIT502','AIT503','AIT504','FIT501','FIT502','FIT503','FIT504',\n",
    "    'P501','P502','PIT501','PIT502','PIT503','FIT601','P601','P602','P603','OUTCOME'\n",
    "]\n",
    "\n",
    "# The first row only contains labels\n",
    "# 'TIMESTAMP' is irrelevant for the thesis\n",
    "# The other dropped columns contain either only 0s, only 1s or only 2s and are therefore irrelevant\n",
    "\n",
    "df = df.iloc[1:]\n",
    "df = df.drop(['TIMESTAMP', 'P202', 'P301', 'P401', 'P404', 'P502', 'P601', 'P603'], axis = 1)\n",
    "\n",
    "# The dataset labels attacks by 'A ttack' and 'Attack', and labels normal data as 'Normal'\n",
    "# To keep the same structure in all datasets, the 'A ttack' and 'Attack' values are changed to '-1' and normal values to '1'\n",
    "\n",
    "df['OUTCOME'].replace(to_replace = ['A ttack', 'Attack'], value = 1, inplace = True)\n",
    "df['OUTCOME'].replace(to_replace = ['Normal'], value = 0, inplace = True)\n",
    "\n",
    "# data types need to be numeric to be encoded to z-scores --> convert column object data types to numerics\n",
    "\n",
    "cols = df.columns[df.columns != 'OUTCOME']\n",
    "df[cols] = df[cols].apply(pd.to_numeric, errors='coerce')\n",
    "\n",
    "# Encoding the feature vectors to z-scores\n",
    "\n",
    "cols = list(df.columns[df.columns != 'OUTCOME'])\n",
    "for col in cols:\n",
    "    df[col] = ((df[col] - df[col].mean())/df[col].std(ddof=0))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [],
   "source": [
    "normal = df[df['OUTCOME'] == 0].sample(10000)\n",
    "attack = df[df['OUTCOME'] == 1].sample(10000)\n",
    "\n",
    "df1 = normal.append(attack).sample(frac=1).reset_index(drop=True)\n",
    "\n",
    "X = df1.drop(['OUTCOME'], axis = 1).values\n",
    "Y = df1[\"OUTCOME\"].values\n",
    "\n",
    "# tsne_plot(X, Y, \"TSNE_SWaT.png\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(20000, 44)\n",
      "44\n",
      "(44,)\n",
      "(20000,)\n"
     ]
    }
   ],
   "source": [
    "print(X.shape)\n",
    "print(X.shape[1])\n",
    "print(X[0].shape)\n",
    "print(Y.shape)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Model: \"encoder\"\n",
      "_________________________________________________________________\n",
      "Layer (type)                 Output Shape              Param #   \n",
      "=================================================================\n",
      "dense_18 (Dense)             (None, 40)                1800      \n",
      "_________________________________________________________________\n",
      "dense_19 (Dense)             (None, 35)                1435      \n",
      "_________________________________________________________________\n",
      "dense_20 (Dense)             (None, 30)                1080      \n",
      "_________________________________________________________________\n",
      "dense_21 (Dense)             (None, 135)               4185      \n",
      "_________________________________________________________________\n",
      "multivariate_normal_tri_l_2  ((None, 15), (None, 15))  0         \n",
      "=================================================================\n",
      "Total params: 8,500\n",
      "Trainable params: 8,500\n",
      "Non-trainable params: 0\n",
      "_________________________________________________________________\n",
      "Model: \"decoder\"\n",
      "_________________________________________________________________\n",
      "Layer (type)                 Output Shape              Param #   \n",
      "=================================================================\n",
      "dense_22 (Dense)             (None, 30)                480       \n",
      "_________________________________________________________________\n",
      "dense_23 (Dense)             (None, 35)                1085      \n",
      "_________________________________________________________________\n",
      "dense_24 (Dense)             (None, 40)                1440      \n",
      "_________________________________________________________________\n",
      "dense_25 (Dense)             (None, 88)                3608      \n",
      "_________________________________________________________________\n",
      "independent_normal_2 (Indepe ((None, 44), (None, 44))  0         \n",
      "=================================================================\n",
      "Total params: 6,613\n",
      "Trainable params: 6,613\n",
      "Non-trainable params: 0\n",
      "_________________________________________________________________\n",
      "Model: \"vae_mlp\"\n",
      "_________________________________________________________________\n",
      "Layer (type)                 Output Shape              Param #   \n",
      "=================================================================\n",
      "encoder_input (InputLayer)   [(None, 44)]              0         \n",
      "_________________________________________________________________\n",
      "dense_18 (Dense)             (None, 40)                1800      \n",
      "_________________________________________________________________\n",
      "dense_19 (Dense)             (None, 35)                1435      \n",
      "_________________________________________________________________\n",
      "dense_20 (Dense)             (None, 30)                1080      \n",
      "_________________________________________________________________\n",
      "dense_21 (Dense)             (None, 135)               4185      \n",
      "_________________________________________________________________\n",
      "multivariate_normal_tri_l_2  ((None, 15), (None, 15))  0         \n",
      "_________________________________________________________________\n",
      "decoder (Sequential)         (None, 44)                6613      \n",
      "=================================================================\n",
      "Total params: 15,113\n",
      "Trainable params: 15,113\n",
      "Non-trainable params: 0\n",
      "_________________________________________________________________\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<IPython.core.display.Image object>"
      ]
     },
     "execution_count": 16,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "def dense_layers(sizes):\n",
    "    return tfk.Sequential([tfkl.Dense(size, activation=tf.nn.leaky_relu) for size in sizes])\n",
    "\n",
    "original_dim = X.shape[1]\n",
    "input_shape = X[0].shape\n",
    "# intermediary_dims = [25, 20, 15, 10, 8]\n",
    "latent_dim = 15\n",
    "batch_size = 128\n",
    "max_epochs = 1000\n",
    "\n",
    "prior = tfd.MultivariateNormalDiag(\n",
    "        loc=tf.zeros([latent_dim]),\n",
    "        scale_identity_multiplier=1.0)\n",
    "\n",
    "encoder = tfk.Sequential([\n",
    "    tfkl.InputLayer(input_shape=input_shape, name='encoder_input'),\n",
    "    tfkl.Dense(40, activation='relu'),\n",
    "    tfkl.Dense(35, activation='relu'),\n",
    "    tfkl.Dense(30, activation='relu'),\n",
    "    #dense_layers(intermediary_dims),\n",
    "    tfkl.Dense(tfpl.MultivariateNormalTriL.params_size(latent_dim), activation=None),\n",
    "    tfpl.MultivariateNormalTriL(latent_dim, \n",
    "                           activity_regularizer=tfpl.KLDivergenceRegularizer(prior)),\n",
    "], name='encoder')\n",
    "\n",
    "\n",
    "encoder.summary()\n",
    "plot_model(encoder, to_file='vae_mlp_encoder.png', show_shapes=True)\n",
    "\n",
    "decoder = tfk.Sequential([\n",
    "    tfkl.InputLayer(input_shape=[latent_dim]),\n",
    "    tfkl.Dense(30, activation='relu'),\n",
    "    tfkl.Dense(35, activation='relu'),\n",
    "    tfkl.Dense(40, activation='relu'),\n",
    "#    dense_layers(reversed(intermediary_dims)),\n",
    "    tfkl.Dense(tfpl.IndependentNormal.params_size(original_dim), activation=None),\n",
    "    tfpl.IndependentNormal(original_dim),\n",
    "], name='decoder')\n",
    "\n",
    "decoder.summary()\n",
    "plot_model(decoder, to_file='vae_mlp_decoder.png', show_shapes=True)\n",
    "\n",
    "vae = tfk.Model(inputs=encoder.inputs,\n",
    "                outputs=decoder(encoder.outputs[0]),\n",
    "                name='vae_mlp')\n",
    "\n",
    "negloglik = lambda x, rv_x: -rv_x.log_prob(x)\n",
    "\n",
    "vae.compile(optimizer=tf.keras.optimizers.Nadam(), \n",
    "            loss=negloglik, metrics=['accuracy'])\n",
    "\n",
    "vae.summary()\n",
    "plot_model(vae,\n",
    "           to_file='vae_mlp.png',\n",
    "           show_shapes=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [],
   "source": [
    "x = df.drop([\"OUTCOME\"], axis=1)\n",
    "y = df[\"OUTCOME\"].values\n",
    "\n",
    "x_norm, x_fraud = x.values[y == 0], x.values[y == 1]\n",
    "\n",
    "x_norm_sample = x_norm[np.random.randint(x_norm.shape[0], size=100000), :]\n",
    "x_norm_train_sample, x_norm_val_sample = train_test_split(x_norm_sample, test_size=0.2)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(395298, 44)\n",
      "(54621, 44)\n",
      "(100000, 44)\n",
      "(80000, 44)\n",
      "(20000, 44)\n"
     ]
    }
   ],
   "source": [
    "print(x_norm.shape)\n",
    "print(x_fraud.shape)\n",
    "print(x_norm_sample.shape)\n",
    "print(x_norm_train_sample.shape)\n",
    "print(x_norm_val_sample.shape)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {},
   "outputs": [],
   "source": [
    "tf_train = tf.data.Dataset.from_tensor_slices((x_norm_train_sample, x_norm_train_sample)).batch(batch_size).prefetch(tf.data.experimental.AUTOTUNE).shuffle(int(10e4))\n",
    "tf_val = tf.data.Dataset.from_tensor_slices((x_norm_val_sample, x_norm_val_sample)).batch(batch_size).prefetch(tf.data.experimental.AUTOTUNE).shuffle(int(10e4))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/1000\n",
      "625/625 [==============================] - 5s 8ms/step - loss: 9072.3232 - accuracy: 0.0369 - val_loss: 33.3745 - val_accuracy: 0.0371\n",
      "Epoch 2/1000\n",
      "625/625 [==============================] - 5s 8ms/step - loss: 34.1590 - accuracy: 0.0405 - val_loss: 26.9307 - val_accuracy: 0.0459\n",
      "Epoch 3/1000\n",
      "625/625 [==============================] - 5s 8ms/step - loss: 15.5500 - accuracy: 0.0711 - val_loss: 1.4891 - val_accuracy: 0.0877\n",
      "Epoch 4/1000\n",
      "625/625 [==============================] - 5s 8ms/step - loss: -3.0066 - accuracy: 0.0902 - val_loss: -10.5586 - val_accuracy: 0.0956\n",
      "Epoch 5/1000\n",
      "625/625 [==============================] - 5s 8ms/step - loss: 586.8453 - accuracy: 0.0944 - val_loss: 18.4407 - val_accuracy: 0.0848\n",
      "Epoch 6/1000\n",
      "625/625 [==============================] - 5s 7ms/step - loss: 3.8290 - accuracy: 0.1049 - val_loss: -3.1436 - val_accuracy: 0.1033\n",
      "Epoch 7/1000\n",
      "625/625 [==============================] - 5s 7ms/step - loss: -12.1652 - accuracy: 0.1055 - val_loss: -18.0808 - val_accuracy: 0.1076\n",
      "Epoch 8/1000\n",
      "625/625 [==============================] - 5s 8ms/step - loss: -14.8624 - accuracy: 0.1055 - val_loss: -11.0171 - val_accuracy: 0.1048\n",
      "Epoch 9/1000\n",
      "625/625 [==============================] - 5s 7ms/step - loss: -19.2392 - accuracy: 0.1077 - val_loss: -23.9376 - val_accuracy: 0.1053\n",
      "Epoch 10/1000\n",
      "625/625 [==============================] - 5s 8ms/step - loss: -21.9940 - accuracy: 0.1047 - val_loss: -29.4920 - val_accuracy: 0.1107\n",
      "Epoch 11/1000\n",
      "625/625 [==============================] - 5s 7ms/step - loss: -25.8456 - accuracy: 0.1073 - val_loss: -8.5750 - val_accuracy: 0.1079\n",
      "Epoch 12/1000\n",
      "625/625 [==============================] - 4s 7ms/step - loss: -28.5627 - accuracy: 0.1080 - val_loss: -25.6133 - val_accuracy: 0.1085\n",
      "Epoch 13/1000\n",
      "625/625 [==============================] - 4s 7ms/step - loss: -31.2721 - accuracy: 0.1082 - val_loss: -27.7826 - val_accuracy: 0.1090\n",
      "Epoch 14/1000\n",
      "625/625 [==============================] - 4s 7ms/step - loss: -35.0331 - accuracy: 0.1103 - val_loss: -40.2736 - val_accuracy: 0.1107\n",
      "Epoch 15/1000\n",
      "625/625 [==============================] - 5s 8ms/step - loss: 1665.0616 - accuracy: 0.0823 - val_loss: 36.8735 - val_accuracy: 0.0785\n",
      "Epoch 16/1000\n",
      "625/625 [==============================] - 5s 8ms/step - loss: 35528.0820 - accuracy: 0.0659 - val_loss: 41.8894 - val_accuracy: 0.0557\n",
      "Epoch 17/1000\n",
      "625/625 [==============================] - 5s 8ms/step - loss: 35.5384 - accuracy: 0.0620 - val_loss: 23.9103 - val_accuracy: 0.0826\n",
      "Epoch 18/1000\n",
      "625/625 [==============================] - 5s 8ms/step - loss: 7.1032 - accuracy: 0.0996 - val_loss: -5.6129 - val_accuracy: 0.1057\n",
      "Epoch 19/1000\n",
      "625/625 [==============================] - 5s 7ms/step - loss: -12.7815 - accuracy: 0.1058 - val_loss: -17.9034 - val_accuracy: 0.1069\n",
      "Epoch 20/1000\n",
      "625/625 [==============================] - 5s 8ms/step - loss: -20.0566 - accuracy: 0.1068 - val_loss: -17.3534 - val_accuracy: 0.1064\n",
      "Epoch 21/1000\n",
      "625/625 [==============================] - 5s 7ms/step - loss: -24.1564 - accuracy: 0.1050 - val_loss: -22.4888 - val_accuracy: 0.1091\n",
      "Epoch 22/1000\n",
      "625/625 [==============================] - 5s 8ms/step - loss: -25.4355 - accuracy: 0.1069 - val_loss: -15.0303 - val_accuracy: 0.1046\n",
      "Epoch 23/1000\n",
      "625/625 [==============================] - 5s 7ms/step - loss: -29.1158 - accuracy: 0.1060 - val_loss: -32.8847 - val_accuracy: 0.1063\n",
      "Epoch 24/1000\n",
      "625/625 [==============================] - 5s 7ms/step - loss: -30.9887 - accuracy: 0.1054 - val_loss: -31.7615 - val_accuracy: 0.1080\n",
      "Epoch 25/1000\n",
      "625/625 [==============================] - 5s 7ms/step - loss: -33.6762 - accuracy: 0.1045 - val_loss: -36.9000 - val_accuracy: 0.1088\n",
      "Epoch 26/1000\n",
      "625/625 [==============================] - 5s 7ms/step - loss: -35.9744 - accuracy: 0.1091 - val_loss: -40.7335 - val_accuracy: 0.1077\n",
      "Epoch 27/1000\n",
      "625/625 [==============================] - 5s 8ms/step - loss: -37.6845 - accuracy: 0.1068 - val_loss: -30.6742 - val_accuracy: 0.1019\n",
      "Epoch 28/1000\n",
      "625/625 [==============================] - 5s 7ms/step - loss: -39.9983 - accuracy: 0.1062 - val_loss: -40.9552 - val_accuracy: 0.1036\n",
      "Epoch 29/1000\n",
      "625/625 [==============================] - 5s 8ms/step - loss: -42.2964 - accuracy: 0.1081 - val_loss: -37.5455 - val_accuracy: 0.1062\n",
      "Epoch 30/1000\n",
      "625/625 [==============================] - 5s 8ms/step - loss: -42.9407 - accuracy: 0.1083 - val_loss: -48.8095 - val_accuracy: 0.1079\n",
      "Epoch 31/1000\n",
      "625/625 [==============================] - 5s 8ms/step - loss: -46.7332 - accuracy: 0.1111 - val_loss: -54.8774 - val_accuracy: 0.1125\n",
      "Epoch 32/1000\n",
      "625/625 [==============================] - 5s 8ms/step - loss: -46.7468 - accuracy: 0.1088 - val_loss: -55.0115 - val_accuracy: 0.1089\n",
      "Epoch 33/1000\n",
      "625/625 [==============================] - 5s 8ms/step - loss: -49.1859 - accuracy: 0.1123 - val_loss: -55.1908 - val_accuracy: 0.1176\n",
      "Epoch 34/1000\n",
      "625/625 [==============================] - 5s 8ms/step - loss: -51.0280 - accuracy: 0.1158 - val_loss: -52.6285 - val_accuracy: 0.1132\n",
      "Epoch 35/1000\n",
      "625/625 [==============================] - 5s 7ms/step - loss: -53.0801 - accuracy: 0.1204 - val_loss: -63.3812 - val_accuracy: 0.1294\n",
      "Epoch 36/1000\n",
      "625/625 [==============================] - 5s 8ms/step - loss: -53.7570 - accuracy: 0.1248 - val_loss: -62.6462 - val_accuracy: 0.1286\n",
      "Epoch 37/1000\n",
      "625/625 [==============================] - 5s 7ms/step - loss: -55.1052 - accuracy: 0.1275 - val_loss: -55.9917 - val_accuracy: 0.1277\n",
      "Epoch 38/1000\n",
      "625/625 [==============================] - 5s 7ms/step - loss: -59.1762 - accuracy: 0.1308 - val_loss: 137.7358 - val_accuracy: 0.1387\n",
      "Epoch 39/1000\n",
      "625/625 [==============================] - 5s 7ms/step - loss: -58.0127 - accuracy: 0.1337 - val_loss: -65.9150 - val_accuracy: 0.1324\n",
      "Epoch 40/1000\n",
      "625/625 [==============================] - 5s 8ms/step - loss: -59.1490 - accuracy: 0.1389 - val_loss: -56.9073 - val_accuracy: 0.1347\n",
      "Epoch 41/1000\n",
      "625/625 [==============================] - 5s 7ms/step - loss: -61.3049 - accuracy: 0.1449 - val_loss: -56.6020 - val_accuracy: 0.1343\n",
      "Epoch 42/1000\n",
      "625/625 [==============================] - 5s 7ms/step - loss: -63.1279 - accuracy: 0.1473 - val_loss: -61.5069 - val_accuracy: 0.1317\n",
      "Epoch 43/1000\n",
      "625/625 [==============================] - 5s 8ms/step - loss: -63.9893 - accuracy: 0.1454 - val_loss: -68.8619 - val_accuracy: 0.1465\n",
      "Epoch 44/1000\n",
      "625/625 [==============================] - 5s 8ms/step - loss: -66.3384 - accuracy: 0.1516 - val_loss: -71.6559 - val_accuracy: 0.1454\n",
      "Epoch 45/1000\n",
      "625/625 [==============================] - 5s 8ms/step - loss: -65.7462 - accuracy: 0.1474 - val_loss: -68.6615 - val_accuracy: 0.1534\n",
      "Epoch 46/1000\n",
      "625/625 [==============================] - 5s 8ms/step - loss: -67.9929 - accuracy: 0.1508 - val_loss: -74.9092 - val_accuracy: 0.1492\n",
      "Epoch 47/1000\n",
      "625/625 [==============================] - 5s 8ms/step - loss: -64.6928 - accuracy: 0.1528 - val_loss: -76.5880 - val_accuracy: 0.1657\n",
      "Epoch 48/1000\n",
      "625/625 [==============================] - 5s 8ms/step - loss: -69.1402 - accuracy: 0.1552 - val_loss: -79.3824 - val_accuracy: 0.1586\n",
      "Epoch 49/1000\n",
      "625/625 [==============================] - 5s 8ms/step - loss: -70.6856 - accuracy: 0.1584 - val_loss: -60.9762 - val_accuracy: 0.1455\n",
      "Epoch 50/1000\n",
      "625/625 [==============================] - 5s 7ms/step - loss: -71.3217 - accuracy: 0.1578 - val_loss: -65.4935 - val_accuracy: 0.1337\n",
      "Epoch 51/1000\n",
      "625/625 [==============================] - 5s 7ms/step - loss: -70.8331 - accuracy: 0.1549 - val_loss: -63.8904 - val_accuracy: 0.1701\n",
      "Epoch 52/1000\n",
      "625/625 [==============================] - 4s 7ms/step - loss: -75.0701 - accuracy: 0.1630 - val_loss: -78.1892 - val_accuracy: 0.1608\n",
      "Epoch 53/1000\n",
      "625/625 [==============================] - 5s 8ms/step - loss: -74.3752 - accuracy: 0.1601 - val_loss: -73.1178 - val_accuracy: 0.1485\n",
      "Epoch 54/1000\n",
      "625/625 [==============================] - 5s 7ms/step - loss: -77.3853 - accuracy: 0.1649 - val_loss: -76.8670 - val_accuracy: 0.1570\n",
      "Epoch 55/1000\n",
      "625/625 [==============================] - 4s 7ms/step - loss: -78.6030 - accuracy: 0.1667 - val_loss: -87.8070 - val_accuracy: 0.1792\n",
      "Epoch 56/1000\n",
      "625/625 [==============================] - 4s 7ms/step - loss: -78.4751 - accuracy: 0.1657 - val_loss: -75.0651 - val_accuracy: 0.1633\n",
      "Epoch 57/1000\n",
      "625/625 [==============================] - 4s 7ms/step - loss: 120.1325 - accuracy: 0.1042 - val_loss: -27.3749 - val_accuracy: 0.1032\n",
      "Epoch 58/1000\n",
      "625/625 [==============================] - 4s 7ms/step - loss: -47.0870 - accuracy: 0.1101 - val_loss: 483.5472 - val_accuracy: 0.1098\n",
      "Epoch 59/1000\n",
      "625/625 [==============================] - 4s 7ms/step - loss: 3880.2429 - accuracy: 0.1081 - val_loss: -3.2002 - val_accuracy: 0.1089\n",
      "Epoch 60/1000\n",
      "625/625 [==============================] - 4s 7ms/step - loss: -6.1771 - accuracy: 0.1027 - val_loss: -16.0493 - val_accuracy: 0.0980\n",
      "Epoch 61/1000\n",
      "625/625 [==============================] - 4s 7ms/step - loss: -23.0821 - accuracy: 0.1030 - val_loss: -34.5324 - val_accuracy: 0.1101\n",
      "Epoch 62/1000\n",
      "625/625 [==============================] - 4s 7ms/step - loss: -39.4844 - accuracy: 0.1133 - val_loss: -42.3986 - val_accuracy: 0.1086\n",
      "Epoch 63/1000\n",
      "625/625 [==============================] - 4s 7ms/step - loss: -48.6796 - accuracy: 0.1193 - val_loss: -54.3272 - val_accuracy: 0.1206\n",
      "Epoch 64/1000\n",
      "625/625 [==============================] - 4s 7ms/step - loss: -55.8528 - accuracy: 0.1274 - val_loss: -60.6082 - val_accuracy: 0.1424\n",
      "Epoch 65/1000\n",
      "625/625 [==============================] - 4s 7ms/step - loss: -59.7762 - accuracy: 0.1336 - val_loss: -45.5810 - val_accuracy: 0.1404\n",
      "Epoch 66/1000\n",
      "625/625 [==============================] - 4s 7ms/step - loss: -64.4590 - accuracy: 0.1467 - val_loss: -68.9975 - val_accuracy: 0.1421\n",
      "Epoch 67/1000\n",
      "625/625 [==============================] - 4s 7ms/step - loss: 1630.9517 - accuracy: 0.1385 - val_loss: -21.5606 - val_accuracy: 0.1394\n",
      "Epoch 68/1000\n",
      "625/625 [==============================] - 4s 7ms/step - loss: -40.2879 - accuracy: 0.1365 - val_loss: -48.6220 - val_accuracy: 0.1433\n",
      "Epoch 69/1000\n",
      "625/625 [==============================] - 4s 7ms/step - loss: -56.5241 - accuracy: 0.1467 - val_loss: -60.9427 - val_accuracy: 0.1508\n",
      "Epoch 70/1000\n",
      "625/625 [==============================] - 4s 7ms/step - loss: -61.0392 - accuracy: 0.1493 - val_loss: -66.4483 - val_accuracy: 0.1426\n",
      "Epoch 71/1000\n",
      "625/625 [==============================] - 4s 7ms/step - loss: -65.8099 - accuracy: 0.1511 - val_loss: -70.0241 - val_accuracy: 0.1454\n",
      "Epoch 72/1000\n",
      "625/625 [==============================] - 4s 7ms/step - loss: -68.7165 - accuracy: 0.1563 - val_loss: -71.8670 - val_accuracy: 0.1540\n",
      "Epoch 73/1000\n",
      "625/625 [==============================] - 4s 7ms/step - loss: -62.1356 - accuracy: 0.1522 - val_loss: -73.4358 - val_accuracy: 0.1431\n",
      "Epoch 74/1000\n",
      "625/625 [==============================] - 5s 7ms/step - loss: -71.4563 - accuracy: 0.1599 - val_loss: -75.7804 - val_accuracy: 0.1608\n",
      "Epoch 75/1000\n",
      "625/625 [==============================] - 4s 7ms/step - loss: -74.5049 - accuracy: 0.1631 - val_loss: -0.6491 - val_accuracy: 0.1665\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "checkpointer = ModelCheckpoint(filepath='bestmodel.h5', verbose=0, save_best_only=True)\n",
    "earlystopper = EarlyStopping(monitor='val_loss', mode='min', min_delta=0.005, patience=20, verbose=0, restore_best_weights=True)\n",
    "\n",
    "hist = vae.fit(tf_train, epochs=max_epochs, shuffle=True, verbose=1,\n",
    "               validation_data=tf_val, callbacks=[checkpointer, earlystopper])\n",
    "\n",
    "plot_loss(hist)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [],
   "source": [
    "batch_size = 16\n",
    "max_epochs = 100"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3.7 (tensorflow)",
   "language": "python",
   "name": "tensorflow"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.7.7"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 4
}