AE_for_Anomaly_Detection.ipynb

{
  "nbformat": 4,
  "nbformat_minor": 0,
  "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": "3.7.7"
    },
    "colab": {
      "name": "AE_for_Anomaly_Detection.ipynb",
      "provenance": [],
      "toc_visible": true,
      "machine_shape": "hm",
      "include_colab_link": true
    },
    "accelerator": "GPU"
  },
  "cells": [
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "view-in-github",
        "colab_type": "text"
      },
      "source": [
        "<a href=\"https://colab.research.google.com/github/rickqiu/deep-learning/blob/master/AE_for_Anomaly_Detection.ipynb\" target=\"_parent\"><img src=\"https://colab.research.google.com/assets/colab-badge.svg\" alt=\"Open In Colab\"/></a>"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "sAMHDWk0NbHX"
      },
      "source": [
        "# Autoencoder (AE) for Anomaly Detection\n",
        "\n",
        "**by Rick Qiu in May, 2020**"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "ZPl2K8mONbHe",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "0631b0cf-4b03-4916-ce39-2f792ac0f754"
      },
      "source": [
        "# Import Python libraries\n",
        "import sys\n",
        "import random\n",
        "import numpy as np\n",
        "from sklearn.preprocessing import MinMaxScaler\n",
        "# Tensorflow\n",
        "import tensorflow as tf\n",
        "from tensorflow import keras\n",
        "from tensorflow.keras.optimizers import Adam\n",
        "from tensorflow.keras.models import Model, load_model\n",
        "from tensorflow.keras.layers import Input, Dense\n",
        "from tensorflow.keras.callbacks import ModelCheckpoint, TensorBoard\n",
        "from tensorflow.keras.utils import plot_model\n",
        "from tensorflow.keras.preprocessing.sequence import pad_sequences\n",
        "import matplotlib.pyplot as plt\n",
        "%matplotlib inline\n",
        "print(\"Python version: \", sys.version)\n",
        "print(\"TensorFlow version:\", tf.__version__)"
      ],
      "execution_count": 1,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Python version:  3.7.10 (default, Feb 20 2021, 21:17:23) \n",
            "[GCC 7.5.0]\n",
            "TensorFlow version: 2.4.1\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "hkut2yMaNbHi"
      },
      "source": [
        "### Data generation"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "Lp4N7pltNbHi",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "8df135ca-1565-4523-a2f3-fc03f026cdf9"
      },
      "source": [
        "# We refer alphabetic character sequence generation to Alon Agmon 2020\n",
        "# https://github.com/a-agmon/experiments/blob/master/Sequence_Anomaly_Detection-NN.ipynb\n",
        "def get_random_string():\n",
        "    # returns a string of the following format: \n",
        "    #[4 letters A-F][1 digit 0-2][3 letters QWOPZXML]\n",
        "    first_letters =  'ABCDEF'\n",
        "    second_numbers = '120'\n",
        "    last_letters = 'QWOPZXML'\n",
        "    str1 = ''.join(random.choice(first_letters) for i in range(4))\n",
        "    str2 = random.choice(second_numbers)\n",
        "    str3 = ''.join(random.choice(last_letters) for i in range(3))\n",
        "    return str1+str2+str3\n",
        "    \n",
        "print(get_random_string())"
      ],
      "execution_count": 2,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "CCAB1XOP\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "NcPj9ilyNbHl"
      },
      "source": [
        "def generate_data():\n",
        "    normals = 50000 - 10\n",
        "    random_sequences = [get_random_string() for i in range(normals)]\n",
        "    # add some anomalies to our list\n",
        "    random_sequences.extend(['XYDC2DCA', 'TXSX1ABC','RNIU4XRE','AABDXUEI','SDRAC5RF',\n",
        "                             'YXDC2DCB', 'XTSX1ABC','RNUI4XRE','TABDXAID','SERACDRF'])\n",
        "    return random_sequences"
      ],
      "execution_count": 3,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "D5NAkOkSNbHr",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "eed5845e-7813-4929-ce70-d2400cc72f3f"
      },
      "source": [
        "data = generate_data()\n",
        "print(data[0:5])"
      ],
      "execution_count": 4,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "['EAAF1ZMP', 'BEAC2XWM', 'DECE2QXO', 'FEFD1PMW', 'FEEB0QXX']\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "nDHDG03bNbHv"
      },
      "source": [
        "### Data Preprocessing"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "pP5AqjqaNbHw"
      },
      "source": [
        "# We refer sequence encoding and decoding to Alon Agmon 2020\n",
        "# https://github.com/a-agmon/experiments/blob/master/Sequence_Anomaly_Detection-NN.ipynb\n",
        "#Build the char index that we will use to encode seqs to numbers\n",
        "char_index = '0abcdefghijklmnopqrstuvwxyz'\n",
        "char_index +='ABCDEFGHIJKLMNOPQRSTUVWXYZ'\n",
        "char_index += '123456789'\n",
        "char_index += '().,-/+=&$?@#!*:;_[]|%⸏{}\\\"\\'' + ' ' +'\\\\'\n",
        "\n",
        "char_to_int = dict((c, i) for i, c in enumerate(char_index))\n",
        "int_to_char = dict((i, c) for i, c in enumerate(char_index))\n",
        "\n",
        "#function that convert a char seqs to numbers seqs \n",
        "#(it does a little more but lets leave it for now)\n",
        "def encode_sequence_list(seqs, feat_n=0):\n",
        "    encoded_seqs = []\n",
        "    for seq in seqs:\n",
        "        encoded_seq = [char_to_int[c] for c in seq]\n",
        "        encoded_seqs.append(encoded_seq)\n",
        "    if(feat_n > 0):\n",
        "        encoded_seqs.append(np.zeros(feat_n))\n",
        "    return pad_sequences(encoded_seqs, padding='post')\n",
        "\n",
        "def decode_sequence_list(seqs):\n",
        "    decoded_seqs = []\n",
        "    for seq in seqs:\n",
        "        decoded_seq = [int_to_char[i] for i in seq]\n",
        "        decoded_seqs.append(decoded_seq)\n",
        "    return decoded_seqs"
      ],
      "execution_count": 5,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "R4_t7WyxNbHz",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "8b4efdcf-b2ce-4041-940b-11d001753faa"
      },
      "source": [
        "# encode each string seq to an integer array [[1],[5],[67]], [[45],[76],[7]\n",
        "encoded_seqs = encode_sequence_list(data)\n",
        "# mix everything up\n",
        "np.random.shuffle(encoded_seqs)\n",
        "print(encoded_seqs.shape)\n",
        "print(encoded_seqs[10])"
      ],
      "execution_count": 6,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "(50000, 8)\n",
            "[30 31 27 29 53 42 39 39]\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "L5Q34Gh2NbH3"
      },
      "source": [
        "# split dataset to train/test (80:20)\n",
        "X_train = encoded_seqs[:40000]\n",
        "X_test = encoded_seqs[40000:]\n",
        "\n",
        "# scale the data to a range [0,1]\n",
        "scaler = MinMaxScaler()\n",
        "scaler = scaler.fit(X_train)\n",
        "scaled_train  = scaler.transform(X_train)\n",
        "scaled_test = scaler.transform(X_test)"
      ],
      "execution_count": 7,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "8fCKc5_yNbH7"
      },
      "source": [
        "### Autoencoder Model"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "W-4KAS8DNbH7"
      },
      "source": [
        "# Define the autoencoder\n",
        "def autoencoder_model():\n",
        "    # Encoder\n",
        "    input_data = Input(shape=(8,))\n",
        "    encoder = Dense(units=5,activation='relu')(input_data)\n",
        "    # Bottleneck\n",
        "    latent_output = Dense(units=2,activation='relu')(encoder)\n",
        "    # Decoder\n",
        "    decoder = Dense(units=5,activation='relu')(latent_output)\n",
        "    decoder = Dense(units=8,activation='linear')(decoder)\n",
        "    # Create Autoencoder model\n",
        "    autoencoder = Model(input_data,decoder)\n",
        "    return autoencoder"
      ],
      "execution_count": 8,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "74lVdHyANbH-"
      },
      "source": [
        "# Define a function which plots training loss and validation loss across the different epochs.\n",
        "def plot_loss(history):\n",
        "    plt.xlabel('Epochs')\n",
        "    plt.ylabel('MSE')\n",
        "    # Plot train loss\n",
        "    plt.plot(history.history[\"loss\"], linewidth=2., label=\"train_loss\")\n",
        "    # Plot val_loss\n",
        "    plt.plot(history.history[\"val_loss\"],linewidth=2., label=\"val_loss\")\n",
        "    plt.legend()\n",
        "    plt.show()"
      ],
      "execution_count": 9,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "eGtLmpeeNbIA",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 884
        },
        "outputId": "6b7e06ed-c7cd-4009-d2e9-87628a7bae23"
      },
      "source": [
        "# Instantiate and compile the model\n",
        "autoencoder = autoencoder_model()\n",
        "opt = Adam(learning_rate=0.001, decay=1e-5)\n",
        "autoencoder.compile(optimizer=opt,loss='mean_squared_error')\n",
        "print(autoencoder.summary())\n",
        "plot_model(autoencoder, to_file='autoencoder.png', show_shapes=True)"
      ],
      "execution_count": 16,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Model: \"model_1\"\n",
            "_________________________________________________________________\n",
            "Layer (type)                 Output Shape              Param #   \n",
            "=================================================================\n",
            "input_2 (InputLayer)         [(None, 8)]               0         \n",
            "_________________________________________________________________\n",
            "dense_4 (Dense)              (None, 5)                 45        \n",
            "_________________________________________________________________\n",
            "dense_5 (Dense)              (None, 2)                 12        \n",
            "_________________________________________________________________\n",
            "dense_6 (Dense)              (None, 5)                 15        \n",
            "_________________________________________________________________\n",
            "dense_7 (Dense)              (None, 8)                 48        \n",
            "=================================================================\n",
            "Total params: 120\n",
            "Trainable params: 120\n",
            "Non-trainable params: 0\n",
            "_________________________________________________________________\n",
            "None\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "execute_result",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<IPython.core.display.Image object>"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 16
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "ZVgnUZIdNbIF"
      },
      "source": [
        "### Model Training"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "iVJADaxbNbIF",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "d50ed030-4a22-494f-fd2f-3f1026b067b6"
      },
      "source": [
        "epochs = 200\n",
        "batch_size = 128\n",
        "\n",
        "\n",
        "checkpointer = ModelCheckpoint(filepath=\"ae_model.h5\",\n",
        "                               verbose=0,\n",
        "                               save_best_only=True)\n",
        "\n",
        "tensorboard = TensorBoard(log_dir='logs',\n",
        "                          histogram_freq=0,\n",
        "                          write_graph=True,\n",
        "                          write_images=True)\n",
        "\n",
        "history = autoencoder.fit(scaled_train, scaled_train,\n",
        "                epochs=epochs,\n",
        "                batch_size=batch_size,\n",
        "                shuffle=True,\n",
        "                verbose = 1,\n",
        "                validation_data=(scaled_test, scaled_test),\n",
        "                callbacks=[checkpointer, tensorboard])"
      ],
      "execution_count": 17,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Epoch 1/200\n",
            "313/313 [==============================] - 1s 4ms/step - loss: 0.2078 - val_loss: 0.1094\n",
            "Epoch 2/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0926 - val_loss: 0.0587\n",
            "Epoch 3/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0540 - val_loss: 0.0458\n",
            "Epoch 4/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0448 - val_loss: 0.0438\n",
            "Epoch 5/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0432 - val_loss: 0.0436\n",
            "Epoch 6/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0431 - val_loss: 0.0436\n",
            "Epoch 7/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0432 - val_loss: 0.0436\n",
            "Epoch 8/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0432 - val_loss: 0.0436\n",
            "Epoch 9/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0432 - val_loss: 0.0436\n",
            "Epoch 10/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0433 - val_loss: 0.0436\n",
            "Epoch 11/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0432 - val_loss: 0.0436\n",
            "Epoch 12/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0432 - val_loss: 0.0436\n",
            "Epoch 13/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0432 - val_loss: 0.0436\n",
            "Epoch 14/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0433 - val_loss: 0.0436\n",
            "Epoch 15/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0430 - val_loss: 0.0436\n",
            "Epoch 16/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0431 - val_loss: 0.0436\n",
            "Epoch 17/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0432 - val_loss: 0.0436\n",
            "Epoch 18/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0433 - val_loss: 0.0436\n",
            "Epoch 19/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0433 - val_loss: 0.0436\n",
            "Epoch 20/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0432 - val_loss: 0.0436\n",
            "Epoch 21/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0432 - val_loss: 0.0436\n",
            "Epoch 22/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0432 - val_loss: 0.0436\n",
            "Epoch 23/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0431 - val_loss: 0.0436\n",
            "Epoch 24/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0431 - val_loss: 0.0436\n",
            "Epoch 25/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0432 - val_loss: 0.0436\n",
            "Epoch 26/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0433 - val_loss: 0.0436\n",
            "Epoch 27/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0432 - val_loss: 0.0436\n",
            "Epoch 28/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0432 - val_loss: 0.0436\n",
            "Epoch 29/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0431 - val_loss: 0.0436\n",
            "Epoch 30/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0431 - val_loss: 0.0436\n",
            "Epoch 31/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0434 - val_loss: 0.0436\n",
            "Epoch 32/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0431 - val_loss: 0.0436\n",
            "Epoch 33/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0432 - val_loss: 0.0436\n",
            "Epoch 34/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0431 - val_loss: 0.0436\n",
            "Epoch 35/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0431 - val_loss: 0.0436\n",
            "Epoch 36/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0432 - val_loss: 0.0436\n",
            "Epoch 37/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0431 - val_loss: 0.0436\n",
            "Epoch 38/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0430 - val_loss: 0.0436\n",
            "Epoch 39/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0431 - val_loss: 0.0436\n",
            "Epoch 40/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0433 - val_loss: 0.0436\n",
            "Epoch 41/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0432 - val_loss: 0.0436\n",
            "Epoch 42/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0432 - val_loss: 0.0436\n",
            "Epoch 43/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0431 - val_loss: 0.0436\n",
            "Epoch 44/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0431 - val_loss: 0.0436\n",
            "Epoch 45/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0431 - val_loss: 0.0436\n",
            "Epoch 46/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0431 - val_loss: 0.0436\n",
            "Epoch 47/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0432 - val_loss: 0.0436\n",
            "Epoch 48/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0431 - val_loss: 0.0436\n",
            "Epoch 49/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0430 - val_loss: 0.0436\n",
            "Epoch 50/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0433 - val_loss: 0.0436\n",
            "Epoch 51/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0431 - val_loss: 0.0436\n",
            "Epoch 52/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0434 - val_loss: 0.0436\n",
            "Epoch 53/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0432 - val_loss: 0.0436\n",
            "Epoch 54/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0433 - val_loss: 0.0436\n",
            "Epoch 55/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0431 - val_loss: 0.0436\n",
            "Epoch 56/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0430 - val_loss: 0.0436\n",
            "Epoch 57/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0433 - val_loss: 0.0436\n",
            "Epoch 58/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0433 - val_loss: 0.0436\n",
            "Epoch 59/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0431 - val_loss: 0.0436\n",
            "Epoch 60/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0432 - val_loss: 0.0436\n",
            "Epoch 61/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0432 - val_loss: 0.0436\n",
            "Epoch 62/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0431 - val_loss: 0.0436\n",
            "Epoch 63/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0431 - val_loss: 0.0436\n",
            "Epoch 64/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0433 - val_loss: 0.0436\n",
            "Epoch 65/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0433 - val_loss: 0.0436\n",
            "Epoch 66/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0432 - val_loss: 0.0436\n",
            "Epoch 67/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0431 - val_loss: 0.0436\n",
            "Epoch 68/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0432 - val_loss: 0.0436\n",
            "Epoch 69/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0432 - val_loss: 0.0436\n",
            "Epoch 70/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0434 - val_loss: 0.0436\n",
            "Epoch 71/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0433 - val_loss: 0.0436\n",
            "Epoch 72/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0430 - val_loss: 0.0436\n",
            "Epoch 73/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0430 - val_loss: 0.0436\n",
            "Epoch 74/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0430 - val_loss: 0.0436\n",
            "Epoch 75/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0433 - val_loss: 0.0436\n",
            "Epoch 76/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0432 - val_loss: 0.0436\n",
            "Epoch 77/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0433 - val_loss: 0.0436\n",
            "Epoch 78/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0432 - val_loss: 0.0436\n",
            "Epoch 79/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0432 - val_loss: 0.0436\n",
            "Epoch 80/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0432 - val_loss: 0.0436\n",
            "Epoch 81/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0433 - val_loss: 0.0436\n",
            "Epoch 82/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0430 - val_loss: 0.0436\n",
            "Epoch 83/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0431 - val_loss: 0.0436\n",
            "Epoch 84/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0432 - val_loss: 0.0436\n",
            "Epoch 85/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0432 - val_loss: 0.0436\n",
            "Epoch 86/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0433 - val_loss: 0.0436\n",
            "Epoch 87/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0431 - val_loss: 0.0436\n",
            "Epoch 88/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0432 - val_loss: 0.0436\n",
            "Epoch 89/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0433 - val_loss: 0.0436\n",
            "Epoch 90/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0430 - val_loss: 0.0436\n",
            "Epoch 91/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0432 - val_loss: 0.0436\n",
            "Epoch 92/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0431 - val_loss: 0.0436\n",
            "Epoch 93/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0432 - val_loss: 0.0436\n",
            "Epoch 94/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0432 - val_loss: 0.0436\n",
            "Epoch 95/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0431 - val_loss: 0.0436\n",
            "Epoch 96/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0434 - val_loss: 0.0436\n",
            "Epoch 97/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0431 - val_loss: 0.0436\n",
            "Epoch 98/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0432 - val_loss: 0.0436\n",
            "Epoch 99/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0430 - val_loss: 0.0436\n",
            "Epoch 100/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0432 - val_loss: 0.0436\n",
            "Epoch 101/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0433 - val_loss: 0.0436\n",
            "Epoch 102/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0433 - val_loss: 0.0436\n",
            "Epoch 103/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0432 - val_loss: 0.0436\n",
            "Epoch 104/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0432 - val_loss: 0.0436\n",
            "Epoch 105/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0433 - val_loss: 0.0436\n",
            "Epoch 106/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0431 - val_loss: 0.0436\n",
            "Epoch 107/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0431 - val_loss: 0.0436\n",
            "Epoch 108/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0431 - val_loss: 0.0436\n",
            "Epoch 109/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0431 - val_loss: 0.0436\n",
            "Epoch 110/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0432 - val_loss: 0.0436\n",
            "Epoch 111/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0433 - val_loss: 0.0436\n",
            "Epoch 112/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0431 - val_loss: 0.0436\n",
            "Epoch 113/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0433 - val_loss: 0.0436\n",
            "Epoch 114/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0432 - val_loss: 0.0436\n",
            "Epoch 115/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0429 - val_loss: 0.0436\n",
            "Epoch 116/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0431 - val_loss: 0.0436\n",
            "Epoch 117/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0430 - val_loss: 0.0436\n",
            "Epoch 118/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0432 - val_loss: 0.0436\n",
            "Epoch 119/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0431 - val_loss: 0.0436\n",
            "Epoch 120/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0431 - val_loss: 0.0436\n",
            "Epoch 121/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0432 - val_loss: 0.0436\n",
            "Epoch 122/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0432 - val_loss: 0.0436\n",
            "Epoch 123/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0431 - val_loss: 0.0436\n",
            "Epoch 124/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0430 - val_loss: 0.0436\n",
            "Epoch 125/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0432 - val_loss: 0.0436\n",
            "Epoch 126/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0433 - val_loss: 0.0436\n",
            "Epoch 127/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0431 - val_loss: 0.0436\n",
            "Epoch 128/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0431 - val_loss: 0.0436\n",
            "Epoch 129/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0432 - val_loss: 0.0436\n",
            "Epoch 130/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0431 - val_loss: 0.0436\n",
            "Epoch 131/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0433 - val_loss: 0.0436\n",
            "Epoch 132/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0431 - val_loss: 0.0436\n",
            "Epoch 133/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0432 - val_loss: 0.0436\n",
            "Epoch 134/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0432 - val_loss: 0.0436\n",
            "Epoch 135/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0433 - val_loss: 0.0436\n",
            "Epoch 136/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0431 - val_loss: 0.0436\n",
            "Epoch 137/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0431 - val_loss: 0.0436\n",
            "Epoch 138/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0433 - val_loss: 0.0436\n",
            "Epoch 139/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0431 - val_loss: 0.0436\n",
            "Epoch 140/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0429 - val_loss: 0.0436\n",
            "Epoch 141/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0432 - val_loss: 0.0436\n",
            "Epoch 142/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0431 - val_loss: 0.0436\n",
            "Epoch 143/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0432 - val_loss: 0.0436\n",
            "Epoch 144/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0432 - val_loss: 0.0436\n",
            "Epoch 145/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0432 - val_loss: 0.0436\n",
            "Epoch 146/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0432 - val_loss: 0.0436\n",
            "Epoch 147/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0431 - val_loss: 0.0436\n",
            "Epoch 148/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0431 - val_loss: 0.0436\n",
            "Epoch 149/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0432 - val_loss: 0.0436\n",
            "Epoch 150/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0432 - val_loss: 0.0436\n",
            "Epoch 151/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0431 - val_loss: 0.0436\n",
            "Epoch 152/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0432 - val_loss: 0.0436\n",
            "Epoch 153/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0430 - val_loss: 0.0436\n",
            "Epoch 154/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0432 - val_loss: 0.0436\n",
            "Epoch 155/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0433 - val_loss: 0.0436\n",
            "Epoch 156/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0433 - val_loss: 0.0436\n",
            "Epoch 157/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0432 - val_loss: 0.0436\n",
            "Epoch 158/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0432 - val_loss: 0.0436\n",
            "Epoch 159/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0431 - val_loss: 0.0436\n",
            "Epoch 160/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0432 - val_loss: 0.0436\n",
            "Epoch 161/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0430 - val_loss: 0.0436\n",
            "Epoch 162/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0432 - val_loss: 0.0436\n",
            "Epoch 163/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0432 - val_loss: 0.0436\n",
            "Epoch 164/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0432 - val_loss: 0.0436\n",
            "Epoch 165/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0433 - val_loss: 0.0436\n",
            "Epoch 166/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0430 - val_loss: 0.0436\n",
            "Epoch 167/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0433 - val_loss: 0.0436\n",
            "Epoch 168/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0431 - val_loss: 0.0436\n",
            "Epoch 169/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0431 - val_loss: 0.0436\n",
            "Epoch 170/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0431 - val_loss: 0.0436\n",
            "Epoch 171/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0431 - val_loss: 0.0436\n",
            "Epoch 172/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0431 - val_loss: 0.0436\n",
            "Epoch 173/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0432 - val_loss: 0.0436\n",
            "Epoch 174/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0432 - val_loss: 0.0436\n",
            "Epoch 175/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0433 - val_loss: 0.0436\n",
            "Epoch 176/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0430 - val_loss: 0.0436\n",
            "Epoch 177/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0432 - val_loss: 0.0436\n",
            "Epoch 178/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0432 - val_loss: 0.0436\n",
            "Epoch 179/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0432 - val_loss: 0.0436\n",
            "Epoch 180/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0432 - val_loss: 0.0436\n",
            "Epoch 181/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0432 - val_loss: 0.0436\n",
            "Epoch 182/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0432 - val_loss: 0.0436\n",
            "Epoch 183/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0431 - val_loss: 0.0436\n",
            "Epoch 184/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0433 - val_loss: 0.0436\n",
            "Epoch 185/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0429 - val_loss: 0.0436\n",
            "Epoch 186/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0433 - val_loss: 0.0436\n",
            "Epoch 187/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0431 - val_loss: 0.0436\n",
            "Epoch 188/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0432 - val_loss: 0.0436\n",
            "Epoch 189/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0431 - val_loss: 0.0436\n",
            "Epoch 190/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0433 - val_loss: 0.0436\n",
            "Epoch 191/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0432 - val_loss: 0.0436\n",
            "Epoch 192/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0432 - val_loss: 0.0436\n",
            "Epoch 193/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0432 - val_loss: 0.0436\n",
            "Epoch 194/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0433 - val_loss: 0.0436\n",
            "Epoch 195/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0432 - val_loss: 0.0436\n",
            "Epoch 196/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0431 - val_loss: 0.0436\n",
            "Epoch 197/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0433 - val_loss: 0.0436\n",
            "Epoch 198/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0432 - val_loss: 0.0436\n",
            "Epoch 199/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0431 - val_loss: 0.0436\n",
            "Epoch 200/200\n",
            "313/313 [==============================] - 1s 2ms/step - loss: 0.0432 - val_loss: 0.0436\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "cWPHrlfuNbIJ",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 316
        },
        "outputId": "bb19f1b4-ee5b-43a4-d477-8849ab95f91b"
      },
      "source": [
        "print (\"Result: \\nTrain loss:%f, Val loss:%f\" % (history.history[\"loss\"][epochs-1], history.history[\"val_loss\"][epochs-1]))\n",
        "# Make Train Loss and Validation Loss versus Epochs plot \n",
        "plot_loss(history)"
      ],
      "execution_count": 18,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Result: \n",
            "Train loss:0.043190, Val loss:0.043592\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": [],
            "needs_background": "light"
          }
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "HPnSUYTSNbIL"
      },
      "source": [
        "### Threshold Selection"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "r6qrMobQNbIL"
      },
      "source": [
        "autoencoder = load_model('ae_model.h5')"
      ],
      "execution_count": 19,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "53fVrJtXNbIN",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "efb13bc8-32a5-4272-c41f-ff8b1884ce97"
      },
      "source": [
        "# compute reconstruction errors on train set\n",
        "predictions = autoencoder.predict(scaled_train)\n",
        "mse = np.mean(np.power(scaled_train - predictions, 2), axis=1)\n",
        "threshold = np.quantile(mse, (1-10/50000))\n",
        "print('MSE threshold:', threshold)"
      ],
      "execution_count": 20,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "MSE threshold: 0.0911897969373238\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "4xqgKr0bNbIQ"
      },
      "source": [
        "### Prediction"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "8oM1YukNNbIR",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "aa298b5a-45a8-4fe7-ebb3-d01c7d59daec"
      },
      "source": [
        "# predict on test set\n",
        "predictions = autoencoder.predict(scaled_test)\n",
        "mse = np.mean(np.power(scaled_test - predictions, 2), axis=1)\n",
        "a = np.array(mse)\n",
        "anoms_idx = np.argwhere(a > threshold)\n",
        "#print(\"Indices:\",anoms_idx.ravel())\n",
        "anomalies = X_test[anoms_idx]\n",
        "for anom in anomalies:\n",
        "    print(decode_sequence_list(anom)[0])"
      ],
      "execution_count": 21,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "['X', 'Y', 'D', 'C', '2', 'D', 'C', 'A']\n",
            "['T', 'X', 'S', 'X', '1', 'A', 'B', 'C']\n",
            "['X', 'T', 'S', 'X', '1', 'A', 'B', 'C']\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "lfVx77q4NbIV"
      },
      "source": [
        "### Visualizing Latent Variables"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "7kpV-wYoNbIV",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 265
        },
        "outputId": "cc1a27d8-93a8-4604-e9eb-4fdd472d3f16"
      },
      "source": [
        "# Plot the latent varaibles\n",
        "hidden_layers = keras.backend.function(\n",
        "[autoencoder.layers[0].input],\n",
        "[autoencoder.layers[2].output]\n",
        ")\n",
        "\n",
        "normal_idx = np.argwhere(a <= threshold)\n",
        "latent_normals = hidden_layers([scaled_test[normal_idx.flatten()]])[0]\n",
        "latent_anomalies = hidden_layers([scaled_test[anoms_idx.flatten()]])[0]\n",
        "\n",
        "plt.scatter(latent_normals[:,0], latent_normals[:,1], alpha=0.5, label='Normal')\n",
        "plt.scatter(latent_anomalies[:,0], latent_anomalies[:,1], alpha=0.5, label='Anomaly')\n",
        "plt.legend()\n",
        "plt.show()"
      ],
      "execution_count": 22,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": [],
            "needs_background": "light"
          }
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "dVYZsn2FNbIZ"
      },
      "source": [
        "### TensorBoard"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "UkOauAkKNbIZ"
      },
      "source": [
        "# Load the TensorBoard notebook extension\n",
        "#%load_ext tensorboard"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "18i-QOWsNbIb"
      },
      "source": [
        "#%tensorboard --logdir logs"
      ],
      "execution_count": null,
      "outputs": []
    }
  ]
}