Uncertainty_quantification.ipynb

{
  "nbformat": 4,
  "nbformat_minor": 0,
  "metadata": {
    "colab": {
      "name": "Uncertainly_quantification.ipynb",
      "provenance": [],
      "machine_shape": "hm"
    },
    "kernelspec": {
      "name": "python3",
      "display_name": "Python 3"
    },
    "accelerator": "GPU"
  },
  "cells": [
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "7FoUvoLy6UTD"
      },
      "source": [
        "# Uncertainty Quantification\n",
        "\n",
        "In this notebook we consider two sources of uncertainty:\n",
        "\n",
        "- Uncertainty stemming from the different way we split the dataset into training-validation-test sets.\n",
        "\n",
        "- Uncertainties due to label noise (mislabeling)\n",
        "\n",
        "\n",
        "### Import packages"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "MO87kPYJ6Uy3",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "a4313d1b-517e-44cb-f308-c7f388618a99"
      },
      "source": [
        "# Import basic packages\n",
        "import numpy as np\n",
        "import scipy as sp\n",
        "import pandas as pd\n",
        "from scipy import interpolate\n",
        "import random\n",
        "\n",
        "# Colab in order to download files\n",
        "from google.colab import files\n",
        "\n",
        "# =========================================\n",
        "# =========================================\n",
        "# scikit-learn for Machine Learn\n",
        "\n",
        "# Preprocessing\n",
        "from sklearn.preprocessing import StandardScaler #Standar scaler for standardization\n",
        "from sklearn.manifold import TSNE # Dimensionality reduction technique\n",
        "from sklearn.model_selection import GridSearchCV #Cross-Validation\n",
        "from sklearn.utils import resample #Resampling\n",
        "from sklearn.model_selection import train_test_split #Train test split\n",
        "# Classifiers\n",
        "from sklearn.svm import SVC #Support Vector Classifier\n",
        "from sklearn.ensemble import RandomForestClassifier as RF #Random Forests\n",
        "# Metrics\n",
        "from sklearn.metrics import accuracy_score \n",
        "from sklearn.metrics import confusion_matrix\n",
        "from sklearn.metrics import precision_score\n",
        "from sklearn.metrics import recall_score\n",
        "from sklearn.metrics import roc_auc_score\n",
        "from sklearn.metrics import roc_curve\n",
        "# Balanced metrics\n",
        "from sklearn.metrics import balanced_accuracy_score\n",
        "\n",
        "\n",
        "# =======================================\n",
        "# =======================================\n",
        "# Tensorflow and Keras\n",
        "import tensorflow as tf\n",
        "from tensorflow import keras\n",
        "from tensorflow.keras import regularizers\n",
        "from tensorflow.keras import optimizers\n",
        "# Keras Layers\n",
        "from keras.layers import InputLayer\n",
        "from keras.layers import Conv2D\n",
        "from keras.layers import BatchNormalization\n",
        "from keras.layers import Dropout\n",
        "from keras.layers import MaxPool2D\n",
        "from keras.layers import Flatten\n",
        "from keras.layers import Dense\n",
        "\n",
        "# ==============================================\n",
        "# ==============================================\n",
        "# Matplotlib, seaborn and plot pretty \n",
        "import matplotlib.pyplot as plt\n",
        "import matplotlib.gridspec as gridspec\n",
        "from mpl_toolkits.axes_grid.inset_locator import (inset_axes, InsetPosition,\n",
        "                                                  mark_inset)\n",
        "\n",
        "import seaborn as sns\n",
        "%matplotlib inline\n",
        "from matplotlib import rcParams\n",
        "rcParams['font.family'] = 'serif'\n",
        "\n",
        "# Adjust rc parameters to make plots pretty\n",
        "def plot_pretty(dpi=200, fontsize=9):\n",
        "    \n",
        "\n",
        "    plt.rc(\"savefig\", dpi=dpi)       # dpi resolution of saved image files\n",
        "    plt.rc('text', usetex=False)      # use LaTeX to process labels\n",
        "    plt.rc('font', size=fontsize)    # fontsize\n",
        "    plt.rc('xtick', direction='in')  # make axes ticks point inward\n",
        "    plt.rc('ytick', direction='in')\n",
        "    plt.rc('xtick.major', pad=10) \n",
        "    plt.rc('xtick.minor', pad=5)\n",
        "    plt.rc('ytick.major', pad=10) \n",
        "    plt.rc('ytick.minor', pad=5)\n",
        "    plt.rc('lines', dotted_pattern = [0.5, 1.1]) # fix dotted lines\n",
        "\n",
        "    return\n",
        "\n",
        "plot_pretty()\n",
        "\n",
        "# Pillow\n",
        "from PIL import Image"
      ],
      "execution_count": 1,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "/usr/local/lib/python3.6/dist-packages/ipykernel_launcher.py:56: MatplotlibDeprecationWarning: \n",
            "The mpl_toolkits.axes_grid module was deprecated in Matplotlib 2.1 and will be removed two minor releases later. Use mpl_toolkits.axes_grid1 and mpl_toolkits.axisartist, which provide the same functionality instead.\n"
          ],
          "name": "stderr"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "ZpjEI_TE6U-D"
      },
      "source": [
        "## Part I: Different random seeds\n",
        "\n",
        "We consider 6 different random seeds\n",
        "\n",
        "\n",
        "**Import dataset**"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "JtP9ePeU6VIl",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "272a7291-457d-448c-acad-c889eedd9cb6"
      },
      "source": [
        "# Mount drive to upload the datasets\n",
        "from google.colab import drive\n",
        "drive.mount('/content/drive')"
      ],
      "execution_count": 2,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Mounted at /content/drive\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "Tco0VmAF6VLW"
      },
      "source": [
        "# The LSBG image dataset\n",
        "LSBG_data_1 = np.float64(np.load(\"/content/drive/My Drive/Galaxies_Data/LSBG_sample_1.npy\"))/255.\n",
        "LSBG_data_2 = np.float64(np.load(\"/content/drive/My Drive/Galaxies_Data/LSBG_sample_2.npy\"))/255.\n",
        "LSBG_data_3 = np.float64(np.load(\"/content/drive/My Drive/Galaxies_Data/LSBG_sample_3.npy\"))/255.\n",
        "LSBG_data_4 = np.float64(np.load(\"/content/drive/My Drive/Galaxies_Data/LSBG_sample_4.npy\"))/255.\n",
        "# ===============================================================================================\n",
        "# ===============================================================================================\n",
        "# The artefacts image dataset\n",
        "Negative_data_2_1 = np.float64(np.load(\"/content/drive/My Drive/Galaxies_Data/Negative_sample_2_1.npy\"))/255.\n",
        "Negative_data_2_2 = np.float64(np.load(\"/content/drive/My Drive/Galaxies_Data/Negative_sample_2_2.npy\"))/255.\n",
        "Negative_data_2_3 = np.float64(np.load(\"/content/drive/My Drive/Galaxies_Data/Negative_sample_2_3.npy\"))/255.\n",
        "Negative_data_2_4 = np.float64(np.load(\"/content/drive/My Drive/Galaxies_Data/Negative_sample_2_4.npy\"))/255.\n",
        "# ===============================================================================================\n",
        "# ===============================================================================================\n",
        "# Import the HSC dataset - positives and negatives\n",
        "HSC_data = np.float64(np.load(\"/content/drive/My Drive/Galaxies_Data/Greco_sample_good.npy\"))/255.\n",
        "HSC_negatives = np.float64(np.load(\"/content/drive/My Drive/Galaxies_Data/Greco_sample_junk.npy\"))/255."
      ],
      "execution_count": 3,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "YV8S0Kcq6VdS"
      },
      "source": [
        "Combine LSBG and artefacts data into one dataset"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "kGl48DodTm6m"
      },
      "source": [
        "X_mat = np.concatenate((LSBG_data_1,LSBG_data_2,LSBG_data_3,LSBG_data_4,\n",
        "                        Negative_data_2_1,Negative_data_2_2,Negative_data_2_3,Negative_data_2_4)) # Features matrix\n",
        "# ==============================================================================\n",
        "# ==============================================================================\n",
        "y_label = np.concatenate((np.ones(20000),np.zeros(20000))) # Labels "
      ],
      "execution_count": 4,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "r2gcVIWdTm-M"
      },
      "source": [
        "seeds = [1,3,18,33,42,96] #Random seeds\n",
        "n_loops = 6 #Number of loops/random seeds\n",
        "\n",
        "\n",
        "# ============================================\n",
        "# Create empty arrays for the evaluation metrics of each loop\n",
        "\n",
        "Accuracies = np.zeros(n_loops)\n",
        "Precisions = np.zeros(n_loops)\n",
        "Recalls = np.zeros(n_loops)\n",
        "tprs = np.zeros((n_loops,200))\n",
        "fprs = np.linspace(0,1,200)\n",
        "AUCs = np.zeros(n_loops)"
      ],
      "execution_count": 5,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "A3cTbRjgUy_V"
      },
      "source": [
        "Calculate over loops"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "JRBak9uVSQUg",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "62f60fed-192b-47ae-c873-a9b6c5f4a5fc"
      },
      "source": [
        "for i in range(n_loops):\n",
        "  # Clear the backend session\n",
        "  tf.keras.backend.clear_session()\n",
        "  # ============================================================================\n",
        "  # ============================================================================\n",
        "  # Get local seed\n",
        "  loc_seed = seeds[i]\n",
        "  # Now split the DES dataset in training, validation and test sets\n",
        "  # First in training and a set for validation and test \n",
        "  X_tr, X_valt, y_tr, y_valt = train_test_split(X_mat, y_label, test_size=0.25, random_state=loc_seed)\n",
        "  # Now split in validation and test sets \n",
        "  X_val, X_test, y_val, y_test = train_test_split(X_valt, y_valt, test_size=0.50, random_state=loc_seed)\n",
        "  # ============================================================================ \n",
        "  # ============================================================================\n",
        "  # Now build the model\n",
        "  model = keras.models.Sequential()\n",
        "  model.add(InputLayer((64,64,3)))\n",
        "  model.add(Conv2D(filters=16, kernel_size=(3,3), padding='same', activation='relu',kernel_regularizer=regularizers.l2(0.13)))\n",
        "  model.add(BatchNormalization())\n",
        "  model.add(MaxPool2D(pool_size=(2, 2), strides=None, padding='valid'))\n",
        "  model.add(Dropout(rate=0.4))\n",
        "  model.add(Conv2D(filters=2*16, kernel_size=(3,3), padding='same', activation='relu',kernel_regularizer=regularizers.l2(0.13)))\n",
        "  model.add(BatchNormalization())\n",
        "  model.add(MaxPool2D(pool_size=(2, 2), strides=None, padding='valid'))\n",
        "  model.add(Dropout(rate=0.4))\n",
        "  model.add(Conv2D(filters=2*32, kernel_size=(3,3), padding='same', activation='relu',kernel_regularizer=regularizers.l2(0.13)))\n",
        "  model.add(BatchNormalization())\n",
        "  model.add(MaxPool2D(pool_size=(2, 2), strides=None, padding='valid'))\n",
        "  model.add(Dropout(rate=0.4))\n",
        "  model.add(Flatten())\n",
        "  model.add(Dense(units=1024, activation='relu',kernel_regularizer=regularizers.l2(0.12)))\n",
        "  model.add(Dense(units=1, activation='sigmoid'))\n",
        "  model.compile(optimizer=optimizers.Adadelta(0.1),\n",
        "                loss= 'binary_crossentropy',\n",
        "                metrics=['accuracy'])\n",
        "  # ============================================================================\n",
        "  # Now fit the model\n",
        "  model.fit(x=X_tr, y=y_tr,\n",
        "          epochs=100, batch_size=64,\n",
        "          shuffle=True,\n",
        "          validation_data=(X_val,y_val))\n",
        "  # ============================================================================\n",
        "  # ============================================================================\n",
        "  # ============================================================================\n",
        "  # Now predict on the test set\n",
        "  y_pred = model.predict_classes(X_test)\n",
        "  # Predict probability scores\n",
        "  scores = model.predict(X_test)\n",
        "  # ============================================================================\n",
        "  # ============================================================================\n",
        "  # Evaluation metrics \n",
        "\n",
        "  Accuracies[i] = accuracy_score(y_test, y_pred)\n",
        "  Precisions[i] = precision_score(y_test, y_pred)\n",
        "  Recalls[i] = recall_score(y_test, y_pred)\n",
        "  AUCs[i] = roc_auc_score(y_test, scores[:,0])\n",
        "  # ============================================================================\n",
        "  # Get now false positives and true positives\n",
        "  fpr, tpr, thresholds = roc_curve(y_test, scores[:,0], pos_label=1.0)\n",
        "  fpr_unique, ind_unique = np.unique(fpr,return_index=True)\n",
        "  tpr_unique = tpr[ind_unique]\n",
        "\n",
        "  # Interpolate the false positives - true positives\n",
        "  interp = interpolate.interp1d(fpr_unique, tpr_unique, kind='linear')\n",
        "  tprs[i,:] = interp(fprs)"
      ],
      "execution_count": 6,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Epoch 1/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 62.0233 - accuracy: 0.8035 - val_loss: 7.8454 - val_accuracy: 0.6938\n",
            "Epoch 2/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 2.8132 - accuracy: 0.8732 - val_loss: 1.1885 - val_accuracy: 0.8110\n",
            "Epoch 3/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.7556 - accuracy: 0.8920 - val_loss: 0.6220 - val_accuracy: 0.8640\n",
            "Epoch 4/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.4937 - accuracy: 0.9001 - val_loss: 0.4565 - val_accuracy: 0.8886\n",
            "Epoch 5/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.4062 - accuracy: 0.9050 - val_loss: 0.3945 - val_accuracy: 0.8958\n",
            "Epoch 6/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3632 - accuracy: 0.9054 - val_loss: 0.3463 - val_accuracy: 0.9116\n",
            "Epoch 7/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3419 - accuracy: 0.9053 - val_loss: 0.3166 - val_accuracy: 0.9078\n",
            "Epoch 8/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3231 - accuracy: 0.9072 - val_loss: 0.3109 - val_accuracy: 0.9094\n",
            "Epoch 9/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3136 - accuracy: 0.9085 - val_loss: 0.3032 - val_accuracy: 0.9142\n",
            "Epoch 10/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3064 - accuracy: 0.9089 - val_loss: 0.2914 - val_accuracy: 0.9086\n",
            "Epoch 11/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3004 - accuracy: 0.9103 - val_loss: 0.2897 - val_accuracy: 0.9118\n",
            "Epoch 12/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2978 - accuracy: 0.9091 - val_loss: 0.2932 - val_accuracy: 0.9040\n",
            "Epoch 13/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2922 - accuracy: 0.9109 - val_loss: 0.2796 - val_accuracy: 0.9192\n",
            "Epoch 14/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2897 - accuracy: 0.9108 - val_loss: 0.2664 - val_accuracy: 0.9176\n",
            "Epoch 15/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2885 - accuracy: 0.9097 - val_loss: 0.3070 - val_accuracy: 0.9124\n",
            "Epoch 16/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2871 - accuracy: 0.9105 - val_loss: 0.3021 - val_accuracy: 0.9140\n",
            "Epoch 17/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2830 - accuracy: 0.9123 - val_loss: 0.2721 - val_accuracy: 0.9192\n",
            "Epoch 18/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2829 - accuracy: 0.9119 - val_loss: 0.3011 - val_accuracy: 0.9016\n",
            "Epoch 19/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2810 - accuracy: 0.9117 - val_loss: 0.2646 - val_accuracy: 0.9148\n",
            "Epoch 20/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2795 - accuracy: 0.9132 - val_loss: 0.2701 - val_accuracy: 0.9142\n",
            "Epoch 21/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2778 - accuracy: 0.9129 - val_loss: 0.3054 - val_accuracy: 0.8970\n",
            "Epoch 22/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2771 - accuracy: 0.9132 - val_loss: 0.3062 - val_accuracy: 0.9108\n",
            "Epoch 23/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2751 - accuracy: 0.9127 - val_loss: 0.2898 - val_accuracy: 0.9072\n",
            "Epoch 24/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2748 - accuracy: 0.9137 - val_loss: 0.2883 - val_accuracy: 0.9062\n",
            "Epoch 25/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2743 - accuracy: 0.9131 - val_loss: 0.2558 - val_accuracy: 0.9202\n",
            "Epoch 26/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2748 - accuracy: 0.9124 - val_loss: 0.2664 - val_accuracy: 0.9130\n",
            "Epoch 27/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2733 - accuracy: 0.9136 - val_loss: 0.2735 - val_accuracy: 0.9110\n",
            "Epoch 28/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2724 - accuracy: 0.9118 - val_loss: 0.2830 - val_accuracy: 0.9078\n",
            "Epoch 29/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2696 - accuracy: 0.9138 - val_loss: 0.2600 - val_accuracy: 0.9170\n",
            "Epoch 30/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2713 - accuracy: 0.9119 - val_loss: 0.2655 - val_accuracy: 0.9116\n",
            "Epoch 31/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2695 - accuracy: 0.9141 - val_loss: 0.2800 - val_accuracy: 0.9084\n",
            "Epoch 32/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2682 - accuracy: 0.9151 - val_loss: 0.2590 - val_accuracy: 0.9168\n",
            "Epoch 33/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2691 - accuracy: 0.9150 - val_loss: 0.2956 - val_accuracy: 0.9022\n",
            "Epoch 34/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2699 - accuracy: 0.9131 - val_loss: 0.2529 - val_accuracy: 0.9168\n",
            "Epoch 35/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2665 - accuracy: 0.9128 - val_loss: 0.2533 - val_accuracy: 0.9160\n",
            "Epoch 36/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2673 - accuracy: 0.9143 - val_loss: 0.2647 - val_accuracy: 0.9200\n",
            "Epoch 37/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2650 - accuracy: 0.9157 - val_loss: 0.2633 - val_accuracy: 0.9102\n",
            "Epoch 38/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2655 - accuracy: 0.9134 - val_loss: 0.2586 - val_accuracy: 0.9152\n",
            "Epoch 39/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2655 - accuracy: 0.9149 - val_loss: 0.2559 - val_accuracy: 0.9136\n",
            "Epoch 40/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2651 - accuracy: 0.9142 - val_loss: 0.2570 - val_accuracy: 0.9144\n",
            "Epoch 41/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2653 - accuracy: 0.9127 - val_loss: 0.2927 - val_accuracy: 0.9032\n",
            "Epoch 42/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2642 - accuracy: 0.9147 - val_loss: 0.2643 - val_accuracy: 0.9106\n",
            "Epoch 43/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2635 - accuracy: 0.9149 - val_loss: 0.2563 - val_accuracy: 0.9180\n",
            "Epoch 44/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2637 - accuracy: 0.9153 - val_loss: 0.2661 - val_accuracy: 0.9134\n",
            "Epoch 45/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2634 - accuracy: 0.9143 - val_loss: 0.2648 - val_accuracy: 0.9088\n",
            "Epoch 46/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2631 - accuracy: 0.9151 - val_loss: 0.2741 - val_accuracy: 0.9086\n",
            "Epoch 47/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2625 - accuracy: 0.9147 - val_loss: 0.2454 - val_accuracy: 0.9202\n",
            "Epoch 48/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2619 - accuracy: 0.9156 - val_loss: 0.2751 - val_accuracy: 0.9094\n",
            "Epoch 49/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2617 - accuracy: 0.9139 - val_loss: 0.2536 - val_accuracy: 0.9194\n",
            "Epoch 50/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2613 - accuracy: 0.9146 - val_loss: 0.2538 - val_accuracy: 0.9134\n",
            "Epoch 51/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2617 - accuracy: 0.9143 - val_loss: 0.2620 - val_accuracy: 0.9130\n",
            "Epoch 52/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2610 - accuracy: 0.9134 - val_loss: 0.2588 - val_accuracy: 0.9134\n",
            "Epoch 53/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2587 - accuracy: 0.9157 - val_loss: 0.2484 - val_accuracy: 0.9170\n",
            "Epoch 54/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2615 - accuracy: 0.9160 - val_loss: 0.2555 - val_accuracy: 0.9148\n",
            "Epoch 55/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2593 - accuracy: 0.9144 - val_loss: 0.2746 - val_accuracy: 0.9050\n",
            "Epoch 56/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2591 - accuracy: 0.9163 - val_loss: 0.2584 - val_accuracy: 0.9150\n",
            "Epoch 57/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2609 - accuracy: 0.9150 - val_loss: 0.2532 - val_accuracy: 0.9128\n",
            "Epoch 58/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2597 - accuracy: 0.9147 - val_loss: 0.2456 - val_accuracy: 0.9170\n",
            "Epoch 59/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2601 - accuracy: 0.9149 - val_loss: 0.2512 - val_accuracy: 0.9208\n",
            "Epoch 60/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2584 - accuracy: 0.9153 - val_loss: 0.2495 - val_accuracy: 0.9210\n",
            "Epoch 61/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2599 - accuracy: 0.9144 - val_loss: 0.2518 - val_accuracy: 0.9146\n",
            "Epoch 62/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2586 - accuracy: 0.9168 - val_loss: 0.2561 - val_accuracy: 0.9150\n",
            "Epoch 63/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2572 - accuracy: 0.9173 - val_loss: 0.2524 - val_accuracy: 0.9164\n",
            "Epoch 64/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2590 - accuracy: 0.9151 - val_loss: 0.2482 - val_accuracy: 0.9182\n",
            "Epoch 65/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2568 - accuracy: 0.9169 - val_loss: 0.2669 - val_accuracy: 0.9104\n",
            "Epoch 66/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2571 - accuracy: 0.9158 - val_loss: 0.2474 - val_accuracy: 0.9148\n",
            "Epoch 67/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2581 - accuracy: 0.9153 - val_loss: 0.2580 - val_accuracy: 0.9138\n",
            "Epoch 68/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2582 - accuracy: 0.9154 - val_loss: 0.2462 - val_accuracy: 0.9210\n",
            "Epoch 69/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2578 - accuracy: 0.9157 - val_loss: 0.2537 - val_accuracy: 0.9158\n",
            "Epoch 70/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2562 - accuracy: 0.9162 - val_loss: 0.2537 - val_accuracy: 0.9142\n",
            "Epoch 71/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2577 - accuracy: 0.9158 - val_loss: 0.2541 - val_accuracy: 0.9130\n",
            "Epoch 72/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2568 - accuracy: 0.9150 - val_loss: 0.2684 - val_accuracy: 0.9098\n",
            "Epoch 73/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2572 - accuracy: 0.9160 - val_loss: 0.2522 - val_accuracy: 0.9142\n",
            "Epoch 74/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2573 - accuracy: 0.9159 - val_loss: 0.2501 - val_accuracy: 0.9136\n",
            "Epoch 75/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2563 - accuracy: 0.9136 - val_loss: 0.2438 - val_accuracy: 0.9198\n",
            "Epoch 76/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2550 - accuracy: 0.9169 - val_loss: 0.2532 - val_accuracy: 0.9148\n",
            "Epoch 77/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2565 - accuracy: 0.9174 - val_loss: 0.2460 - val_accuracy: 0.9188\n",
            "Epoch 78/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2552 - accuracy: 0.9173 - val_loss: 0.2560 - val_accuracy: 0.9194\n",
            "Epoch 79/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2556 - accuracy: 0.9145 - val_loss: 0.2599 - val_accuracy: 0.9178\n",
            "Epoch 80/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2554 - accuracy: 0.9160 - val_loss: 0.2442 - val_accuracy: 0.9170\n",
            "Epoch 81/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2560 - accuracy: 0.9167 - val_loss: 0.2494 - val_accuracy: 0.9180\n",
            "Epoch 82/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2548 - accuracy: 0.9156 - val_loss: 0.2633 - val_accuracy: 0.9090\n",
            "Epoch 83/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2553 - accuracy: 0.9165 - val_loss: 0.2673 - val_accuracy: 0.9104\n",
            "Epoch 84/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2558 - accuracy: 0.9153 - val_loss: 0.2592 - val_accuracy: 0.9126\n",
            "Epoch 85/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2553 - accuracy: 0.9159 - val_loss: 0.2712 - val_accuracy: 0.9074\n",
            "Epoch 86/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2548 - accuracy: 0.9169 - val_loss: 0.2520 - val_accuracy: 0.9158\n",
            "Epoch 87/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2550 - accuracy: 0.9157 - val_loss: 0.2469 - val_accuracy: 0.9162\n",
            "Epoch 88/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2541 - accuracy: 0.9169 - val_loss: 0.2492 - val_accuracy: 0.9120\n",
            "Epoch 89/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2560 - accuracy: 0.9158 - val_loss: 0.2431 - val_accuracy: 0.9202\n",
            "Epoch 90/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2524 - accuracy: 0.9177 - val_loss: 0.2714 - val_accuracy: 0.9078\n",
            "Epoch 91/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2538 - accuracy: 0.9163 - val_loss: 0.2385 - val_accuracy: 0.9186\n",
            "Epoch 92/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2542 - accuracy: 0.9152 - val_loss: 0.2467 - val_accuracy: 0.9142\n",
            "Epoch 93/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2542 - accuracy: 0.9153 - val_loss: 0.2431 - val_accuracy: 0.9184\n",
            "Epoch 94/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2556 - accuracy: 0.9145 - val_loss: 0.2433 - val_accuracy: 0.9168\n",
            "Epoch 95/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2540 - accuracy: 0.9163 - val_loss: 0.2538 - val_accuracy: 0.9128\n",
            "Epoch 96/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2549 - accuracy: 0.9153 - val_loss: 0.2496 - val_accuracy: 0.9132\n",
            "Epoch 97/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2542 - accuracy: 0.9165 - val_loss: 0.2670 - val_accuracy: 0.9094\n",
            "Epoch 98/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2519 - accuracy: 0.9155 - val_loss: 0.2521 - val_accuracy: 0.9214\n",
            "Epoch 99/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2537 - accuracy: 0.9155 - val_loss: 0.2459 - val_accuracy: 0.9148\n",
            "Epoch 100/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2531 - accuracy: 0.9160 - val_loss: 0.2645 - val_accuracy: 0.9090\n",
            "WARNING:tensorflow:From <ipython-input-6-0e6c059f7597>:46: Sequential.predict_classes (from tensorflow.python.keras.engine.sequential) is deprecated and will be removed after 2021-01-01.\n",
            "Instructions for updating:\n",
            "Please use instead:* `np.argmax(model.predict(x), axis=-1)`,   if your model does multi-class classification   (e.g. if it uses a `softmax` last-layer activation).* `(model.predict(x) > 0.5).astype(\"int32\")`,   if your model does binary classification   (e.g. if it uses a `sigmoid` last-layer activation).\n",
            "Epoch 1/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 57.7632 - accuracy: 0.7988 - val_loss: 6.9717 - val_accuracy: 0.5574\n",
            "Epoch 2/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 2.6317 - accuracy: 0.8722 - val_loss: 1.2534 - val_accuracy: 0.8000\n",
            "Epoch 3/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.7881 - accuracy: 0.8916 - val_loss: 0.6480 - val_accuracy: 0.8634\n",
            "Epoch 4/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.5081 - accuracy: 0.8993 - val_loss: 0.4732 - val_accuracy: 0.8896\n",
            "Epoch 5/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.4151 - accuracy: 0.9027 - val_loss: 0.4015 - val_accuracy: 0.9016\n",
            "Epoch 6/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3738 - accuracy: 0.9033 - val_loss: 0.3455 - val_accuracy: 0.9112\n",
            "Epoch 7/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3454 - accuracy: 0.9070 - val_loss: 0.3380 - val_accuracy: 0.9078\n",
            "Epoch 8/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3304 - accuracy: 0.9069 - val_loss: 0.3273 - val_accuracy: 0.9072\n",
            "Epoch 9/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3192 - accuracy: 0.9065 - val_loss: 0.3294 - val_accuracy: 0.9068\n",
            "Epoch 10/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3104 - accuracy: 0.9087 - val_loss: 0.2959 - val_accuracy: 0.9124\n",
            "Epoch 11/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3028 - accuracy: 0.9083 - val_loss: 0.3019 - val_accuracy: 0.9084\n",
            "Epoch 12/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2985 - accuracy: 0.9105 - val_loss: 0.2845 - val_accuracy: 0.9152\n",
            "Epoch 13/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2946 - accuracy: 0.9103 - val_loss: 0.2876 - val_accuracy: 0.9180\n",
            "Epoch 14/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2923 - accuracy: 0.9097 - val_loss: 0.2760 - val_accuracy: 0.9182\n",
            "Epoch 15/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2896 - accuracy: 0.9097 - val_loss: 0.2951 - val_accuracy: 0.9094\n",
            "Epoch 16/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2862 - accuracy: 0.9108 - val_loss: 0.2869 - val_accuracy: 0.9106\n",
            "Epoch 17/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2860 - accuracy: 0.9121 - val_loss: 0.2830 - val_accuracy: 0.9114\n",
            "Epoch 18/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2835 - accuracy: 0.9108 - val_loss: 0.2950 - val_accuracy: 0.9128\n",
            "Epoch 19/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2816 - accuracy: 0.9115 - val_loss: 0.3035 - val_accuracy: 0.9022\n",
            "Epoch 20/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2814 - accuracy: 0.9099 - val_loss: 0.2737 - val_accuracy: 0.9158\n",
            "Epoch 21/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2818 - accuracy: 0.9102 - val_loss: 0.2668 - val_accuracy: 0.9150\n",
            "Epoch 22/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2788 - accuracy: 0.9117 - val_loss: 0.2729 - val_accuracy: 0.9148\n",
            "Epoch 23/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2768 - accuracy: 0.9134 - val_loss: 0.2771 - val_accuracy: 0.9112\n",
            "Epoch 24/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2784 - accuracy: 0.9106 - val_loss: 0.2654 - val_accuracy: 0.9150\n",
            "Epoch 25/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2768 - accuracy: 0.9113 - val_loss: 0.3085 - val_accuracy: 0.9018\n",
            "Epoch 26/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2754 - accuracy: 0.9118 - val_loss: 0.2688 - val_accuracy: 0.9122\n",
            "Epoch 27/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2763 - accuracy: 0.9120 - val_loss: 0.2733 - val_accuracy: 0.9156\n",
            "Epoch 28/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2738 - accuracy: 0.9125 - val_loss: 0.2572 - val_accuracy: 0.9196\n",
            "Epoch 29/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2736 - accuracy: 0.9116 - val_loss: 0.2809 - val_accuracy: 0.9060\n",
            "Epoch 30/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2718 - accuracy: 0.9134 - val_loss: 0.2770 - val_accuracy: 0.9134\n",
            "Epoch 31/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2702 - accuracy: 0.9132 - val_loss: 0.2734 - val_accuracy: 0.9120\n",
            "Epoch 32/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2701 - accuracy: 0.9132 - val_loss: 0.2859 - val_accuracy: 0.9056\n",
            "Epoch 33/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2696 - accuracy: 0.9132 - val_loss: 0.2570 - val_accuracy: 0.9192\n",
            "Epoch 34/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2682 - accuracy: 0.9142 - val_loss: 0.2716 - val_accuracy: 0.9100\n",
            "Epoch 35/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2679 - accuracy: 0.9143 - val_loss: 0.2798 - val_accuracy: 0.9090\n",
            "Epoch 36/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2697 - accuracy: 0.9140 - val_loss: 0.2993 - val_accuracy: 0.9052\n",
            "Epoch 37/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2656 - accuracy: 0.9134 - val_loss: 0.2835 - val_accuracy: 0.9068\n",
            "Epoch 38/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2690 - accuracy: 0.9114 - val_loss: 0.2567 - val_accuracy: 0.9190\n",
            "Epoch 39/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2677 - accuracy: 0.9140 - val_loss: 0.2633 - val_accuracy: 0.9158\n",
            "Epoch 40/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2696 - accuracy: 0.9121 - val_loss: 0.2666 - val_accuracy: 0.9168\n",
            "Epoch 41/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2665 - accuracy: 0.9141 - val_loss: 0.2632 - val_accuracy: 0.9142\n",
            "Epoch 42/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2670 - accuracy: 0.9130 - val_loss: 0.2627 - val_accuracy: 0.9146\n",
            "Epoch 43/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2678 - accuracy: 0.9127 - val_loss: 0.2595 - val_accuracy: 0.9146\n",
            "Epoch 44/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2665 - accuracy: 0.9127 - val_loss: 0.2564 - val_accuracy: 0.9178\n",
            "Epoch 45/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2668 - accuracy: 0.9136 - val_loss: 0.2603 - val_accuracy: 0.9152\n",
            "Epoch 46/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2639 - accuracy: 0.9133 - val_loss: 0.2586 - val_accuracy: 0.9170\n",
            "Epoch 47/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2660 - accuracy: 0.9128 - val_loss: 0.2665 - val_accuracy: 0.9162\n",
            "Epoch 48/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2641 - accuracy: 0.9150 - val_loss: 0.2584 - val_accuracy: 0.9152\n",
            "Epoch 49/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2632 - accuracy: 0.9139 - val_loss: 0.2592 - val_accuracy: 0.9150\n",
            "Epoch 50/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2638 - accuracy: 0.9127 - val_loss: 0.2533 - val_accuracy: 0.9198\n",
            "Epoch 51/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2623 - accuracy: 0.9144 - val_loss: 0.2585 - val_accuracy: 0.9174\n",
            "Epoch 52/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2626 - accuracy: 0.9136 - val_loss: 0.2706 - val_accuracy: 0.9094\n",
            "Epoch 53/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2627 - accuracy: 0.9139 - val_loss: 0.2575 - val_accuracy: 0.9158\n",
            "Epoch 54/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2630 - accuracy: 0.9136 - val_loss: 0.2606 - val_accuracy: 0.9154\n",
            "Epoch 55/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2609 - accuracy: 0.9148 - val_loss: 0.2759 - val_accuracy: 0.9128\n",
            "Epoch 56/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2616 - accuracy: 0.9133 - val_loss: 0.2543 - val_accuracy: 0.9202\n",
            "Epoch 57/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2610 - accuracy: 0.9150 - val_loss: 0.2658 - val_accuracy: 0.9138\n",
            "Epoch 58/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2619 - accuracy: 0.9150 - val_loss: 0.2595 - val_accuracy: 0.9156\n",
            "Epoch 59/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2594 - accuracy: 0.9155 - val_loss: 0.2633 - val_accuracy: 0.9144\n",
            "Epoch 60/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2612 - accuracy: 0.9133 - val_loss: 0.2521 - val_accuracy: 0.9162\n",
            "Epoch 61/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2598 - accuracy: 0.9144 - val_loss: 0.2528 - val_accuracy: 0.9160\n",
            "Epoch 62/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2596 - accuracy: 0.9141 - val_loss: 0.2534 - val_accuracy: 0.9184\n",
            "Epoch 63/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2592 - accuracy: 0.9144 - val_loss: 0.2548 - val_accuracy: 0.9174\n",
            "Epoch 64/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2592 - accuracy: 0.9157 - val_loss: 0.2535 - val_accuracy: 0.9176\n",
            "Epoch 65/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2609 - accuracy: 0.9132 - val_loss: 0.2464 - val_accuracy: 0.9230\n",
            "Epoch 66/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2572 - accuracy: 0.9157 - val_loss: 0.2936 - val_accuracy: 0.8924\n",
            "Epoch 67/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2605 - accuracy: 0.9141 - val_loss: 0.2662 - val_accuracy: 0.9124\n",
            "Epoch 68/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2593 - accuracy: 0.9150 - val_loss: 0.2625 - val_accuracy: 0.9142\n",
            "Epoch 69/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2591 - accuracy: 0.9141 - val_loss: 0.3015 - val_accuracy: 0.8976\n",
            "Epoch 70/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2577 - accuracy: 0.9152 - val_loss: 0.2562 - val_accuracy: 0.9164\n",
            "Epoch 71/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2596 - accuracy: 0.9155 - val_loss: 0.2542 - val_accuracy: 0.9186\n",
            "Epoch 72/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2568 - accuracy: 0.9148 - val_loss: 0.2692 - val_accuracy: 0.9138\n",
            "Epoch 73/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2581 - accuracy: 0.9128 - val_loss: 0.2530 - val_accuracy: 0.9194\n",
            "Epoch 74/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2566 - accuracy: 0.9161 - val_loss: 0.2584 - val_accuracy: 0.9158\n",
            "Epoch 75/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2578 - accuracy: 0.9158 - val_loss: 0.2657 - val_accuracy: 0.9124\n",
            "Epoch 76/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2572 - accuracy: 0.9153 - val_loss: 0.2471 - val_accuracy: 0.9176\n",
            "Epoch 77/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2568 - accuracy: 0.9150 - val_loss: 0.2452 - val_accuracy: 0.9214\n",
            "Epoch 78/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2574 - accuracy: 0.9160 - val_loss: 0.2579 - val_accuracy: 0.9148\n",
            "Epoch 79/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2582 - accuracy: 0.9152 - val_loss: 0.2519 - val_accuracy: 0.9182\n",
            "Epoch 80/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2552 - accuracy: 0.9160 - val_loss: 0.2630 - val_accuracy: 0.9148\n",
            "Epoch 81/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2565 - accuracy: 0.9154 - val_loss: 0.2594 - val_accuracy: 0.9148\n",
            "Epoch 82/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2543 - accuracy: 0.9163 - val_loss: 0.2675 - val_accuracy: 0.9122\n",
            "Epoch 83/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2572 - accuracy: 0.9141 - val_loss: 0.2532 - val_accuracy: 0.9184\n",
            "Epoch 84/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2587 - accuracy: 0.9133 - val_loss: 0.2539 - val_accuracy: 0.9190\n",
            "Epoch 85/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2564 - accuracy: 0.9155 - val_loss: 0.2495 - val_accuracy: 0.9192\n",
            "Epoch 86/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2557 - accuracy: 0.9138 - val_loss: 0.2459 - val_accuracy: 0.9220\n",
            "Epoch 87/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2561 - accuracy: 0.9146 - val_loss: 0.2511 - val_accuracy: 0.9186\n",
            "Epoch 88/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2555 - accuracy: 0.9152 - val_loss: 0.2655 - val_accuracy: 0.9086\n",
            "Epoch 89/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2569 - accuracy: 0.9147 - val_loss: 0.2487 - val_accuracy: 0.9196\n",
            "Epoch 90/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2556 - accuracy: 0.9139 - val_loss: 0.2697 - val_accuracy: 0.9106\n",
            "Epoch 91/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2559 - accuracy: 0.9143 - val_loss: 0.2576 - val_accuracy: 0.9156\n",
            "Epoch 92/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2558 - accuracy: 0.9167 - val_loss: 0.2596 - val_accuracy: 0.9126\n",
            "Epoch 93/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2551 - accuracy: 0.9160 - val_loss: 0.2583 - val_accuracy: 0.9148\n",
            "Epoch 94/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2552 - accuracy: 0.9158 - val_loss: 0.2706 - val_accuracy: 0.9080\n",
            "Epoch 95/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2566 - accuracy: 0.9148 - val_loss: 0.2591 - val_accuracy: 0.9146\n",
            "Epoch 96/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2534 - accuracy: 0.9161 - val_loss: 0.2720 - val_accuracy: 0.9112\n",
            "Epoch 97/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2559 - accuracy: 0.9135 - val_loss: 0.2596 - val_accuracy: 0.9148\n",
            "Epoch 98/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2550 - accuracy: 0.9147 - val_loss: 0.2577 - val_accuracy: 0.9148\n",
            "Epoch 99/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2533 - accuracy: 0.9157 - val_loss: 0.2495 - val_accuracy: 0.9180\n",
            "Epoch 100/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2535 - accuracy: 0.9160 - val_loss: 0.2595 - val_accuracy: 0.9138\n",
            "Epoch 1/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 57.4002 - accuracy: 0.8047 - val_loss: 6.6890 - val_accuracy: 0.6956\n",
            "Epoch 2/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 2.5722 - accuracy: 0.8740 - val_loss: 1.1406 - val_accuracy: 0.8348\n",
            "Epoch 3/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.7367 - accuracy: 0.8900 - val_loss: 0.6273 - val_accuracy: 0.8568\n",
            "Epoch 4/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.4793 - accuracy: 0.8985 - val_loss: 0.4497 - val_accuracy: 0.8848\n",
            "Epoch 5/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.3900 - accuracy: 0.9030 - val_loss: 0.3725 - val_accuracy: 0.9000\n",
            "Epoch 6/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.3561 - accuracy: 0.9031 - val_loss: 0.3513 - val_accuracy: 0.8984\n",
            "Epoch 7/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3344 - accuracy: 0.9060 - val_loss: 0.3159 - val_accuracy: 0.9144\n",
            "Epoch 8/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.3205 - accuracy: 0.9074 - val_loss: 0.3294 - val_accuracy: 0.8994\n",
            "Epoch 9/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3099 - accuracy: 0.9075 - val_loss: 0.2969 - val_accuracy: 0.9130\n",
            "Epoch 10/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3052 - accuracy: 0.9087 - val_loss: 0.3171 - val_accuracy: 0.9028\n",
            "Epoch 11/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3002 - accuracy: 0.9081 - val_loss: 0.3102 - val_accuracy: 0.9028\n",
            "Epoch 12/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2966 - accuracy: 0.9082 - val_loss: 0.3090 - val_accuracy: 0.8994\n",
            "Epoch 13/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2917 - accuracy: 0.9101 - val_loss: 0.2817 - val_accuracy: 0.9108\n",
            "Epoch 14/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2909 - accuracy: 0.9106 - val_loss: 0.2757 - val_accuracy: 0.9136\n",
            "Epoch 15/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2864 - accuracy: 0.9101 - val_loss: 0.2825 - val_accuracy: 0.9098\n",
            "Epoch 16/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2846 - accuracy: 0.9112 - val_loss: 0.2903 - val_accuracy: 0.9062\n",
            "Epoch 17/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2852 - accuracy: 0.9107 - val_loss: 0.2807 - val_accuracy: 0.9094\n",
            "Epoch 18/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2806 - accuracy: 0.9118 - val_loss: 0.2848 - val_accuracy: 0.9054\n",
            "Epoch 19/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2823 - accuracy: 0.9100 - val_loss: 0.2957 - val_accuracy: 0.9068\n",
            "Epoch 20/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2803 - accuracy: 0.9103 - val_loss: 0.2943 - val_accuracy: 0.9012\n",
            "Epoch 21/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2784 - accuracy: 0.9118 - val_loss: 0.2741 - val_accuracy: 0.9128\n",
            "Epoch 22/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2761 - accuracy: 0.9123 - val_loss: 0.2932 - val_accuracy: 0.9020\n",
            "Epoch 23/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2769 - accuracy: 0.9110 - val_loss: 0.3416 - val_accuracy: 0.8888\n",
            "Epoch 24/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2759 - accuracy: 0.9138 - val_loss: 0.2690 - val_accuracy: 0.9120\n",
            "Epoch 25/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2736 - accuracy: 0.9134 - val_loss: 0.2636 - val_accuracy: 0.9180\n",
            "Epoch 26/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2748 - accuracy: 0.9117 - val_loss: 0.2725 - val_accuracy: 0.9102\n",
            "Epoch 27/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2733 - accuracy: 0.9136 - val_loss: 0.3053 - val_accuracy: 0.8916\n",
            "Epoch 28/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2733 - accuracy: 0.9114 - val_loss: 0.2847 - val_accuracy: 0.9034\n",
            "Epoch 29/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2712 - accuracy: 0.9123 - val_loss: 0.2761 - val_accuracy: 0.9100\n",
            "Epoch 30/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2730 - accuracy: 0.9117 - val_loss: 0.2700 - val_accuracy: 0.9090\n",
            "Epoch 31/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2699 - accuracy: 0.9129 - val_loss: 0.2699 - val_accuracy: 0.9110\n",
            "Epoch 32/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2696 - accuracy: 0.9145 - val_loss: 0.2606 - val_accuracy: 0.9108\n",
            "Epoch 33/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2725 - accuracy: 0.9115 - val_loss: 0.2603 - val_accuracy: 0.9112\n",
            "Epoch 34/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2677 - accuracy: 0.9139 - val_loss: 0.2655 - val_accuracy: 0.9116\n",
            "Epoch 35/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2670 - accuracy: 0.9143 - val_loss: 0.2601 - val_accuracy: 0.9184\n",
            "Epoch 36/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2675 - accuracy: 0.9131 - val_loss: 0.2767 - val_accuracy: 0.9066\n",
            "Epoch 37/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2685 - accuracy: 0.9127 - val_loss: 0.2591 - val_accuracy: 0.9154\n",
            "Epoch 38/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2663 - accuracy: 0.9151 - val_loss: 0.2717 - val_accuracy: 0.9052\n",
            "Epoch 39/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2666 - accuracy: 0.9137 - val_loss: 0.2527 - val_accuracy: 0.9206\n",
            "Epoch 40/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2637 - accuracy: 0.9154 - val_loss: 0.2629 - val_accuracy: 0.9114\n",
            "Epoch 41/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2665 - accuracy: 0.9156 - val_loss: 0.2611 - val_accuracy: 0.9130\n",
            "Epoch 42/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2651 - accuracy: 0.9148 - val_loss: 0.3588 - val_accuracy: 0.8762\n",
            "Epoch 43/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2647 - accuracy: 0.9135 - val_loss: 0.2826 - val_accuracy: 0.9024\n",
            "Epoch 44/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2656 - accuracy: 0.9142 - val_loss: 0.2761 - val_accuracy: 0.9008\n",
            "Epoch 45/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2613 - accuracy: 0.9145 - val_loss: 0.2534 - val_accuracy: 0.9178\n",
            "Epoch 46/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2638 - accuracy: 0.9150 - val_loss: 0.2584 - val_accuracy: 0.9160\n",
            "Epoch 47/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2632 - accuracy: 0.9137 - val_loss: 0.2709 - val_accuracy: 0.9050\n",
            "Epoch 48/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2631 - accuracy: 0.9143 - val_loss: 0.2556 - val_accuracy: 0.9126\n",
            "Epoch 49/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2628 - accuracy: 0.9155 - val_loss: 0.2555 - val_accuracy: 0.9156\n",
            "Epoch 50/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2627 - accuracy: 0.9131 - val_loss: 0.2596 - val_accuracy: 0.9162\n",
            "Epoch 51/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2615 - accuracy: 0.9147 - val_loss: 0.2570 - val_accuracy: 0.9150\n",
            "Epoch 52/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2628 - accuracy: 0.9159 - val_loss: 0.2540 - val_accuracy: 0.9168\n",
            "Epoch 53/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2625 - accuracy: 0.9130 - val_loss: 0.2488 - val_accuracy: 0.9166\n",
            "Epoch 54/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2595 - accuracy: 0.9164 - val_loss: 0.2971 - val_accuracy: 0.9010\n",
            "Epoch 55/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2594 - accuracy: 0.9151 - val_loss: 0.2912 - val_accuracy: 0.9008\n",
            "Epoch 56/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2611 - accuracy: 0.9143 - val_loss: 0.2652 - val_accuracy: 0.9104\n",
            "Epoch 57/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2593 - accuracy: 0.9136 - val_loss: 0.2460 - val_accuracy: 0.9196\n",
            "Epoch 58/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2603 - accuracy: 0.9155 - val_loss: 0.2660 - val_accuracy: 0.9092\n",
            "Epoch 59/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2602 - accuracy: 0.9146 - val_loss: 0.2658 - val_accuracy: 0.9114\n",
            "Epoch 60/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2617 - accuracy: 0.9147 - val_loss: 0.2813 - val_accuracy: 0.9066\n",
            "Epoch 61/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2592 - accuracy: 0.9138 - val_loss: 0.2479 - val_accuracy: 0.9194\n",
            "Epoch 62/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2610 - accuracy: 0.9163 - val_loss: 0.2623 - val_accuracy: 0.9098\n",
            "Epoch 63/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2603 - accuracy: 0.9150 - val_loss: 0.2443 - val_accuracy: 0.9184\n",
            "Epoch 64/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2601 - accuracy: 0.9150 - val_loss: 0.2604 - val_accuracy: 0.9128\n",
            "Epoch 65/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2592 - accuracy: 0.9144 - val_loss: 0.2458 - val_accuracy: 0.9234\n",
            "Epoch 66/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2586 - accuracy: 0.9140 - val_loss: 0.2470 - val_accuracy: 0.9168\n",
            "Epoch 67/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2583 - accuracy: 0.9156 - val_loss: 0.2602 - val_accuracy: 0.9130\n",
            "Epoch 68/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2602 - accuracy: 0.9141 - val_loss: 0.2654 - val_accuracy: 0.9068\n",
            "Epoch 69/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2579 - accuracy: 0.9146 - val_loss: 0.2558 - val_accuracy: 0.9136\n",
            "Epoch 70/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2598 - accuracy: 0.9149 - val_loss: 0.2500 - val_accuracy: 0.9176\n",
            "Epoch 71/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2577 - accuracy: 0.9155 - val_loss: 0.2559 - val_accuracy: 0.9132\n",
            "Epoch 72/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2589 - accuracy: 0.9142 - val_loss: 0.2603 - val_accuracy: 0.9126\n",
            "Epoch 73/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2588 - accuracy: 0.9146 - val_loss: 0.2508 - val_accuracy: 0.9186\n",
            "Epoch 74/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2574 - accuracy: 0.9155 - val_loss: 0.2468 - val_accuracy: 0.9184\n",
            "Epoch 75/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2562 - accuracy: 0.9167 - val_loss: 0.2733 - val_accuracy: 0.9042\n",
            "Epoch 76/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2588 - accuracy: 0.9147 - val_loss: 0.2750 - val_accuracy: 0.9046\n",
            "Epoch 77/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2567 - accuracy: 0.9146 - val_loss: 0.2495 - val_accuracy: 0.9170\n",
            "Epoch 78/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2569 - accuracy: 0.9155 - val_loss: 0.2488 - val_accuracy: 0.9166\n",
            "Epoch 79/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2575 - accuracy: 0.9148 - val_loss: 0.2618 - val_accuracy: 0.9130\n",
            "Epoch 80/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2569 - accuracy: 0.9149 - val_loss: 0.2399 - val_accuracy: 0.9194\n",
            "Epoch 81/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2562 - accuracy: 0.9159 - val_loss: 0.2577 - val_accuracy: 0.9126\n",
            "Epoch 82/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2560 - accuracy: 0.9161 - val_loss: 0.2677 - val_accuracy: 0.9058\n",
            "Epoch 83/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2560 - accuracy: 0.9162 - val_loss: 0.2469 - val_accuracy: 0.9192\n",
            "Epoch 84/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2547 - accuracy: 0.9166 - val_loss: 0.2587 - val_accuracy: 0.9180\n",
            "Epoch 85/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2560 - accuracy: 0.9155 - val_loss: 0.2522 - val_accuracy: 0.9168\n",
            "Epoch 86/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2565 - accuracy: 0.9157 - val_loss: 0.2526 - val_accuracy: 0.9136\n",
            "Epoch 87/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2555 - accuracy: 0.9160 - val_loss: 0.2485 - val_accuracy: 0.9168\n",
            "Epoch 88/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2559 - accuracy: 0.9155 - val_loss: 0.2624 - val_accuracy: 0.9130\n",
            "Epoch 89/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2543 - accuracy: 0.9161 - val_loss: 0.2461 - val_accuracy: 0.9168\n",
            "Epoch 90/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2532 - accuracy: 0.9153 - val_loss: 0.2473 - val_accuracy: 0.9194\n",
            "Epoch 91/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2535 - accuracy: 0.9165 - val_loss: 0.2563 - val_accuracy: 0.9128\n",
            "Epoch 92/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2550 - accuracy: 0.9141 - val_loss: 0.2521 - val_accuracy: 0.9106\n",
            "Epoch 93/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2541 - accuracy: 0.9162 - val_loss: 0.2563 - val_accuracy: 0.9118\n",
            "Epoch 94/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2544 - accuracy: 0.9160 - val_loss: 0.2556 - val_accuracy: 0.9138\n",
            "Epoch 95/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2535 - accuracy: 0.9159 - val_loss: 0.2583 - val_accuracy: 0.9136\n",
            "Epoch 96/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2543 - accuracy: 0.9141 - val_loss: 0.2445 - val_accuracy: 0.9214\n",
            "Epoch 97/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2535 - accuracy: 0.9161 - val_loss: 0.2520 - val_accuracy: 0.9168\n",
            "Epoch 98/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2535 - accuracy: 0.9165 - val_loss: 0.2484 - val_accuracy: 0.9172\n",
            "Epoch 99/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2539 - accuracy: 0.9156 - val_loss: 0.2531 - val_accuracy: 0.9166\n",
            "Epoch 100/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2521 - accuracy: 0.9165 - val_loss: 0.2503 - val_accuracy: 0.9148\n",
            "Epoch 1/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 54.1381 - accuracy: 0.8256 - val_loss: 6.0122 - val_accuracy: 0.8126\n",
            "Epoch 2/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 2.3897 - accuracy: 0.8772 - val_loss: 1.1292 - val_accuracy: 0.8056\n",
            "Epoch 3/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.6954 - accuracy: 0.8931 - val_loss: 0.5630 - val_accuracy: 0.8778\n",
            "Epoch 4/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.4566 - accuracy: 0.8999 - val_loss: 0.4455 - val_accuracy: 0.8862\n",
            "Epoch 5/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3808 - accuracy: 0.9042 - val_loss: 0.3918 - val_accuracy: 0.8866\n",
            "Epoch 6/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3467 - accuracy: 0.9052 - val_loss: 0.3530 - val_accuracy: 0.8974\n",
            "Epoch 7/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3263 - accuracy: 0.9058 - val_loss: 0.3340 - val_accuracy: 0.8972\n",
            "Epoch 8/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.3156 - accuracy: 0.9077 - val_loss: 0.3155 - val_accuracy: 0.9048\n",
            "Epoch 9/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3054 - accuracy: 0.9093 - val_loss: 0.3178 - val_accuracy: 0.9022\n",
            "Epoch 10/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.3017 - accuracy: 0.9085 - val_loss: 0.3111 - val_accuracy: 0.9010\n",
            "Epoch 11/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2961 - accuracy: 0.9110 - val_loss: 0.3384 - val_accuracy: 0.8932\n",
            "Epoch 12/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2949 - accuracy: 0.9091 - val_loss: 0.3068 - val_accuracy: 0.9026\n",
            "Epoch 13/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2913 - accuracy: 0.9093 - val_loss: 0.3168 - val_accuracy: 0.9004\n",
            "Epoch 14/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2901 - accuracy: 0.9099 - val_loss: 0.2922 - val_accuracy: 0.9074\n",
            "Epoch 15/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2867 - accuracy: 0.9113 - val_loss: 0.2857 - val_accuracy: 0.9086\n",
            "Epoch 16/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2843 - accuracy: 0.9111 - val_loss: 0.3139 - val_accuracy: 0.9010\n",
            "Epoch 17/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2835 - accuracy: 0.9122 - val_loss: 0.2962 - val_accuracy: 0.8994\n",
            "Epoch 18/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2811 - accuracy: 0.9122 - val_loss: 0.3001 - val_accuracy: 0.9034\n",
            "Epoch 19/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2808 - accuracy: 0.9104 - val_loss: 0.2783 - val_accuracy: 0.9100\n",
            "Epoch 20/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2783 - accuracy: 0.9121 - val_loss: 0.2879 - val_accuracy: 0.9108\n",
            "Epoch 21/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2761 - accuracy: 0.9127 - val_loss: 0.3086 - val_accuracy: 0.8980\n",
            "Epoch 22/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2762 - accuracy: 0.9119 - val_loss: 0.2902 - val_accuracy: 0.9064\n",
            "Epoch 23/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2758 - accuracy: 0.9134 - val_loss: 0.2804 - val_accuracy: 0.9102\n",
            "Epoch 24/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2739 - accuracy: 0.9135 - val_loss: 0.2777 - val_accuracy: 0.9098\n",
            "Epoch 25/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2726 - accuracy: 0.9127 - val_loss: 0.2876 - val_accuracy: 0.9038\n",
            "Epoch 26/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2721 - accuracy: 0.9133 - val_loss: 0.2850 - val_accuracy: 0.9140\n",
            "Epoch 27/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2703 - accuracy: 0.9141 - val_loss: 0.2855 - val_accuracy: 0.9084\n",
            "Epoch 28/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2692 - accuracy: 0.9143 - val_loss: 0.2868 - val_accuracy: 0.9022\n",
            "Epoch 29/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2685 - accuracy: 0.9136 - val_loss: 0.2759 - val_accuracy: 0.9114\n",
            "Epoch 30/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2695 - accuracy: 0.9139 - val_loss: 0.2732 - val_accuracy: 0.9134\n",
            "Epoch 31/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2696 - accuracy: 0.9132 - val_loss: 0.2775 - val_accuracy: 0.9056\n",
            "Epoch 32/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2660 - accuracy: 0.9154 - val_loss: 0.2693 - val_accuracy: 0.9108\n",
            "Epoch 33/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2674 - accuracy: 0.9146 - val_loss: 0.2813 - val_accuracy: 0.9038\n",
            "Epoch 34/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2658 - accuracy: 0.9140 - val_loss: 0.2787 - val_accuracy: 0.9104\n",
            "Epoch 35/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2641 - accuracy: 0.9151 - val_loss: 0.2787 - val_accuracy: 0.9106\n",
            "Epoch 36/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2659 - accuracy: 0.9139 - val_loss: 0.3099 - val_accuracy: 0.8990\n",
            "Epoch 37/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2659 - accuracy: 0.9134 - val_loss: 0.2786 - val_accuracy: 0.9082\n",
            "Epoch 38/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2639 - accuracy: 0.9142 - val_loss: 0.2778 - val_accuracy: 0.9096\n",
            "Epoch 39/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2637 - accuracy: 0.9139 - val_loss: 0.2958 - val_accuracy: 0.9020\n",
            "Epoch 40/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2619 - accuracy: 0.9151 - val_loss: 0.2933 - val_accuracy: 0.8992\n",
            "Epoch 41/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2634 - accuracy: 0.9157 - val_loss: 0.2689 - val_accuracy: 0.9142\n",
            "Epoch 42/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2619 - accuracy: 0.9153 - val_loss: 0.3036 - val_accuracy: 0.8984\n",
            "Epoch 43/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2627 - accuracy: 0.9146 - val_loss: 0.2705 - val_accuracy: 0.9086\n",
            "Epoch 44/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2629 - accuracy: 0.9148 - val_loss: 0.2644 - val_accuracy: 0.9164\n",
            "Epoch 45/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2623 - accuracy: 0.9145 - val_loss: 0.2911 - val_accuracy: 0.9050\n",
            "Epoch 46/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2621 - accuracy: 0.9146 - val_loss: 0.2729 - val_accuracy: 0.9070\n",
            "Epoch 47/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2594 - accuracy: 0.9158 - val_loss: 0.2761 - val_accuracy: 0.9096\n",
            "Epoch 48/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2597 - accuracy: 0.9154 - val_loss: 0.2780 - val_accuracy: 0.9088\n",
            "Epoch 49/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2604 - accuracy: 0.9146 - val_loss: 0.2656 - val_accuracy: 0.9092\n",
            "Epoch 50/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2603 - accuracy: 0.9152 - val_loss: 0.2662 - val_accuracy: 0.9080\n",
            "Epoch 51/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2565 - accuracy: 0.9167 - val_loss: 0.3040 - val_accuracy: 0.8996\n",
            "Epoch 52/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2600 - accuracy: 0.9161 - val_loss: 0.2741 - val_accuracy: 0.9070\n",
            "Epoch 53/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2588 - accuracy: 0.9151 - val_loss: 0.2788 - val_accuracy: 0.9102\n",
            "Epoch 54/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2598 - accuracy: 0.9154 - val_loss: 0.2723 - val_accuracy: 0.9138\n",
            "Epoch 55/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2600 - accuracy: 0.9156 - val_loss: 0.2877 - val_accuracy: 0.9048\n",
            "Epoch 56/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2573 - accuracy: 0.9168 - val_loss: 0.2665 - val_accuracy: 0.9096\n",
            "Epoch 57/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2582 - accuracy: 0.9154 - val_loss: 0.2787 - val_accuracy: 0.9050\n",
            "Epoch 58/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2580 - accuracy: 0.9160 - val_loss: 0.2689 - val_accuracy: 0.9102\n",
            "Epoch 59/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2581 - accuracy: 0.9145 - val_loss: 0.2684 - val_accuracy: 0.9122\n",
            "Epoch 60/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2589 - accuracy: 0.9147 - val_loss: 0.2608 - val_accuracy: 0.9128\n",
            "Epoch 61/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2575 - accuracy: 0.9157 - val_loss: 0.2683 - val_accuracy: 0.9080\n",
            "Epoch 62/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2564 - accuracy: 0.9162 - val_loss: 0.2688 - val_accuracy: 0.9102\n",
            "Epoch 63/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2581 - accuracy: 0.9153 - val_loss: 0.2795 - val_accuracy: 0.9072\n",
            "Epoch 64/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2565 - accuracy: 0.9158 - val_loss: 0.2711 - val_accuracy: 0.9094\n",
            "Epoch 65/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2576 - accuracy: 0.9163 - val_loss: 0.2742 - val_accuracy: 0.9106\n",
            "Epoch 66/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2575 - accuracy: 0.9154 - val_loss: 0.2659 - val_accuracy: 0.9130\n",
            "Epoch 67/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2556 - accuracy: 0.9158 - val_loss: 0.2682 - val_accuracy: 0.9098\n",
            "Epoch 68/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2575 - accuracy: 0.9157 - val_loss: 0.2741 - val_accuracy: 0.9052\n",
            "Epoch 69/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2562 - accuracy: 0.9157 - val_loss: 0.2975 - val_accuracy: 0.9002\n",
            "Epoch 70/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2554 - accuracy: 0.9160 - val_loss: 0.2679 - val_accuracy: 0.9130\n",
            "Epoch 71/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2549 - accuracy: 0.9165 - val_loss: 0.2806 - val_accuracy: 0.9036\n",
            "Epoch 72/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2564 - accuracy: 0.9162 - val_loss: 0.2727 - val_accuracy: 0.9074\n",
            "Epoch 73/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2563 - accuracy: 0.9156 - val_loss: 0.2780 - val_accuracy: 0.9040\n",
            "Epoch 74/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2539 - accuracy: 0.9166 - val_loss: 0.2730 - val_accuracy: 0.9060\n",
            "Epoch 75/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2554 - accuracy: 0.9160 - val_loss: 0.2641 - val_accuracy: 0.9114\n",
            "Epoch 76/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2547 - accuracy: 0.9156 - val_loss: 0.2770 - val_accuracy: 0.9042\n",
            "Epoch 77/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2550 - accuracy: 0.9156 - val_loss: 0.2722 - val_accuracy: 0.9130\n",
            "Epoch 78/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2544 - accuracy: 0.9181 - val_loss: 0.2684 - val_accuracy: 0.9116\n",
            "Epoch 79/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2538 - accuracy: 0.9167 - val_loss: 0.2673 - val_accuracy: 0.9116\n",
            "Epoch 80/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2533 - accuracy: 0.9157 - val_loss: 0.2548 - val_accuracy: 0.9174\n",
            "Epoch 81/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2548 - accuracy: 0.9169 - val_loss: 0.2655 - val_accuracy: 0.9078\n",
            "Epoch 82/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2548 - accuracy: 0.9164 - val_loss: 0.3014 - val_accuracy: 0.8998\n",
            "Epoch 83/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2543 - accuracy: 0.9164 - val_loss: 0.2714 - val_accuracy: 0.9096\n",
            "Epoch 84/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2544 - accuracy: 0.9152 - val_loss: 0.2618 - val_accuracy: 0.9112\n",
            "Epoch 85/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2528 - accuracy: 0.9179 - val_loss: 0.2625 - val_accuracy: 0.9116\n",
            "Epoch 86/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2529 - accuracy: 0.9161 - val_loss: 0.2787 - val_accuracy: 0.9048\n",
            "Epoch 87/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2544 - accuracy: 0.9162 - val_loss: 0.2689 - val_accuracy: 0.9118\n",
            "Epoch 88/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2527 - accuracy: 0.9159 - val_loss: 0.2894 - val_accuracy: 0.8996\n",
            "Epoch 89/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2527 - accuracy: 0.9170 - val_loss: 0.2724 - val_accuracy: 0.9048\n",
            "Epoch 90/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2552 - accuracy: 0.9181 - val_loss: 0.2610 - val_accuracy: 0.9106\n",
            "Epoch 91/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2514 - accuracy: 0.9173 - val_loss: 0.2543 - val_accuracy: 0.9146\n",
            "Epoch 92/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2514 - accuracy: 0.9165 - val_loss: 0.2674 - val_accuracy: 0.9114\n",
            "Epoch 93/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2515 - accuracy: 0.9181 - val_loss: 0.2578 - val_accuracy: 0.9138\n",
            "Epoch 94/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2535 - accuracy: 0.9161 - val_loss: 0.2737 - val_accuracy: 0.9128\n",
            "Epoch 95/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2510 - accuracy: 0.9177 - val_loss: 0.2664 - val_accuracy: 0.9088\n",
            "Epoch 96/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2518 - accuracy: 0.9177 - val_loss: 0.2851 - val_accuracy: 0.9032\n",
            "Epoch 97/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2528 - accuracy: 0.9165 - val_loss: 0.2878 - val_accuracy: 0.9028\n",
            "Epoch 98/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2522 - accuracy: 0.9162 - val_loss: 0.2574 - val_accuracy: 0.9140\n",
            "Epoch 99/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2513 - accuracy: 0.9154 - val_loss: 0.2841 - val_accuracy: 0.9078\n",
            "Epoch 100/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2520 - accuracy: 0.9174 - val_loss: 0.2667 - val_accuracy: 0.9098\n",
            "Epoch 1/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 59.5405 - accuracy: 0.8117 - val_loss: 7.5592 - val_accuracy: 0.8218\n",
            "Epoch 2/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 2.8509 - accuracy: 0.8721 - val_loss: 1.1681 - val_accuracy: 0.8352\n",
            "Epoch 3/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.7479 - accuracy: 0.8916 - val_loss: 0.5920 - val_accuracy: 0.8888\n",
            "Epoch 4/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.4787 - accuracy: 0.9001 - val_loss: 0.4354 - val_accuracy: 0.8960\n",
            "Epoch 5/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3912 - accuracy: 0.9037 - val_loss: 0.4289 - val_accuracy: 0.8848\n",
            "Epoch 6/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3524 - accuracy: 0.9052 - val_loss: 0.3308 - val_accuracy: 0.9088\n",
            "Epoch 7/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3297 - accuracy: 0.9056 - val_loss: 0.3308 - val_accuracy: 0.9028\n",
            "Epoch 8/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3155 - accuracy: 0.9078 - val_loss: 0.2987 - val_accuracy: 0.9160\n",
            "Epoch 9/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3072 - accuracy: 0.9076 - val_loss: 0.3109 - val_accuracy: 0.9046\n",
            "Epoch 10/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3018 - accuracy: 0.9087 - val_loss: 0.2968 - val_accuracy: 0.9120\n",
            "Epoch 11/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2969 - accuracy: 0.9090 - val_loss: 0.3018 - val_accuracy: 0.9054\n",
            "Epoch 12/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2937 - accuracy: 0.9090 - val_loss: 0.3074 - val_accuracy: 0.9040\n",
            "Epoch 13/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2897 - accuracy: 0.9107 - val_loss: 0.2816 - val_accuracy: 0.9152\n",
            "Epoch 14/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2895 - accuracy: 0.9095 - val_loss: 0.3141 - val_accuracy: 0.9034\n",
            "Epoch 15/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2870 - accuracy: 0.9096 - val_loss: 0.3158 - val_accuracy: 0.8914\n",
            "Epoch 16/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2833 - accuracy: 0.9110 - val_loss: 0.3026 - val_accuracy: 0.9082\n",
            "Epoch 17/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2841 - accuracy: 0.9100 - val_loss: 0.2952 - val_accuracy: 0.9090\n",
            "Epoch 18/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2822 - accuracy: 0.9112 - val_loss: 0.2733 - val_accuracy: 0.9146\n",
            "Epoch 19/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2796 - accuracy: 0.9134 - val_loss: 0.3111 - val_accuracy: 0.9010\n",
            "Epoch 20/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2791 - accuracy: 0.9118 - val_loss: 0.2667 - val_accuracy: 0.9184\n",
            "Epoch 21/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2773 - accuracy: 0.9116 - val_loss: 0.2826 - val_accuracy: 0.9108\n",
            "Epoch 22/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2774 - accuracy: 0.9120 - val_loss: 0.2730 - val_accuracy: 0.9164\n",
            "Epoch 23/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2768 - accuracy: 0.9118 - val_loss: 0.2744 - val_accuracy: 0.9130\n",
            "Epoch 24/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2740 - accuracy: 0.9125 - val_loss: 0.2916 - val_accuracy: 0.9068\n",
            "Epoch 25/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2748 - accuracy: 0.9119 - val_loss: 0.2968 - val_accuracy: 0.9090\n",
            "Epoch 26/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2719 - accuracy: 0.9141 - val_loss: 0.2854 - val_accuracy: 0.9092\n",
            "Epoch 27/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2739 - accuracy: 0.9112 - val_loss: 0.2613 - val_accuracy: 0.9196\n",
            "Epoch 28/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2734 - accuracy: 0.9130 - val_loss: 0.2700 - val_accuracy: 0.9178\n",
            "Epoch 29/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2700 - accuracy: 0.9133 - val_loss: 0.2720 - val_accuracy: 0.9120\n",
            "Epoch 30/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2698 - accuracy: 0.9124 - val_loss: 0.2686 - val_accuracy: 0.9154\n",
            "Epoch 31/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2694 - accuracy: 0.9137 - val_loss: 0.2684 - val_accuracy: 0.9136\n",
            "Epoch 32/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2692 - accuracy: 0.9126 - val_loss: 0.2655 - val_accuracy: 0.9146\n",
            "Epoch 33/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2701 - accuracy: 0.9133 - val_loss: 0.4239 - val_accuracy: 0.8324\n",
            "Epoch 34/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2694 - accuracy: 0.9130 - val_loss: 0.2689 - val_accuracy: 0.9148\n",
            "Epoch 35/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2686 - accuracy: 0.9135 - val_loss: 0.2974 - val_accuracy: 0.9046\n",
            "Epoch 36/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2685 - accuracy: 0.9136 - val_loss: 0.2672 - val_accuracy: 0.9176\n",
            "Epoch 37/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2661 - accuracy: 0.9144 - val_loss: 0.2568 - val_accuracy: 0.9212\n",
            "Epoch 38/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2681 - accuracy: 0.9142 - val_loss: 0.2844 - val_accuracy: 0.9104\n",
            "Epoch 39/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2655 - accuracy: 0.9137 - val_loss: 0.2862 - val_accuracy: 0.9092\n",
            "Epoch 40/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2663 - accuracy: 0.9136 - val_loss: 0.2637 - val_accuracy: 0.9160\n",
            "Epoch 41/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2663 - accuracy: 0.9130 - val_loss: 0.2661 - val_accuracy: 0.9184\n",
            "Epoch 42/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2667 - accuracy: 0.9143 - val_loss: 0.2693 - val_accuracy: 0.9142\n",
            "Epoch 43/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2657 - accuracy: 0.9127 - val_loss: 0.2650 - val_accuracy: 0.9160\n",
            "Epoch 44/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2651 - accuracy: 0.9137 - val_loss: 0.2731 - val_accuracy: 0.9136\n",
            "Epoch 45/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2646 - accuracy: 0.9137 - val_loss: 0.2550 - val_accuracy: 0.9186\n",
            "Epoch 46/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2644 - accuracy: 0.9137 - val_loss: 0.2710 - val_accuracy: 0.9154\n",
            "Epoch 47/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2651 - accuracy: 0.9135 - val_loss: 0.2613 - val_accuracy: 0.9154\n",
            "Epoch 48/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2637 - accuracy: 0.9137 - val_loss: 0.2639 - val_accuracy: 0.9160\n",
            "Epoch 49/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2617 - accuracy: 0.9149 - val_loss: 0.2562 - val_accuracy: 0.9182\n",
            "Epoch 50/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2641 - accuracy: 0.9148 - val_loss: 0.2668 - val_accuracy: 0.9146\n",
            "Epoch 51/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2614 - accuracy: 0.9148 - val_loss: 0.2612 - val_accuracy: 0.9134\n",
            "Epoch 52/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2617 - accuracy: 0.9147 - val_loss: 0.2682 - val_accuracy: 0.9112\n",
            "Epoch 53/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2629 - accuracy: 0.9145 - val_loss: 0.2700 - val_accuracy: 0.9154\n",
            "Epoch 54/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2624 - accuracy: 0.9149 - val_loss: 0.2739 - val_accuracy: 0.9110\n",
            "Epoch 55/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2603 - accuracy: 0.9154 - val_loss: 0.2749 - val_accuracy: 0.9136\n",
            "Epoch 56/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2593 - accuracy: 0.9146 - val_loss: 0.2641 - val_accuracy: 0.9104\n",
            "Epoch 57/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2612 - accuracy: 0.9144 - val_loss: 0.2711 - val_accuracy: 0.9124\n",
            "Epoch 58/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2601 - accuracy: 0.9137 - val_loss: 0.2611 - val_accuracy: 0.9146\n",
            "Epoch 59/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2611 - accuracy: 0.9123 - val_loss: 0.2520 - val_accuracy: 0.9208\n",
            "Epoch 60/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2602 - accuracy: 0.9153 - val_loss: 0.2742 - val_accuracy: 0.9110\n",
            "Epoch 61/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2615 - accuracy: 0.9137 - val_loss: 0.2599 - val_accuracy: 0.9164\n",
            "Epoch 62/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2598 - accuracy: 0.9146 - val_loss: 0.2684 - val_accuracy: 0.9126\n",
            "Epoch 63/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2574 - accuracy: 0.9163 - val_loss: 0.2578 - val_accuracy: 0.9142\n",
            "Epoch 64/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2579 - accuracy: 0.9143 - val_loss: 0.2797 - val_accuracy: 0.9106\n",
            "Epoch 65/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2591 - accuracy: 0.9152 - val_loss: 0.2658 - val_accuracy: 0.9170\n",
            "Epoch 66/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2605 - accuracy: 0.9138 - val_loss: 0.2694 - val_accuracy: 0.9142\n",
            "Epoch 67/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2589 - accuracy: 0.9142 - val_loss: 0.2708 - val_accuracy: 0.9094\n",
            "Epoch 68/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2570 - accuracy: 0.9155 - val_loss: 0.2506 - val_accuracy: 0.9202\n",
            "Epoch 69/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2590 - accuracy: 0.9139 - val_loss: 0.2571 - val_accuracy: 0.9160\n",
            "Epoch 70/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2569 - accuracy: 0.9161 - val_loss: 0.2546 - val_accuracy: 0.9172\n",
            "Epoch 71/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2606 - accuracy: 0.9144 - val_loss: 0.2674 - val_accuracy: 0.9142\n",
            "Epoch 72/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2573 - accuracy: 0.9152 - val_loss: 0.2524 - val_accuracy: 0.9208\n",
            "Epoch 73/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2568 - accuracy: 0.9159 - val_loss: 0.2757 - val_accuracy: 0.9086\n",
            "Epoch 74/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2574 - accuracy: 0.9144 - val_loss: 0.2542 - val_accuracy: 0.9212\n",
            "Epoch 75/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2571 - accuracy: 0.9149 - val_loss: 0.2783 - val_accuracy: 0.9060\n",
            "Epoch 76/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2554 - accuracy: 0.9166 - val_loss: 0.2578 - val_accuracy: 0.9160\n",
            "Epoch 77/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2573 - accuracy: 0.9143 - val_loss: 0.2545 - val_accuracy: 0.9206\n",
            "Epoch 78/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2571 - accuracy: 0.9144 - val_loss: 0.2551 - val_accuracy: 0.9138\n",
            "Epoch 79/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2581 - accuracy: 0.9144 - val_loss: 0.2686 - val_accuracy: 0.9110\n",
            "Epoch 80/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2564 - accuracy: 0.9162 - val_loss: 0.2544 - val_accuracy: 0.9174\n",
            "Epoch 81/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2552 - accuracy: 0.9148 - val_loss: 0.2631 - val_accuracy: 0.9132\n",
            "Epoch 82/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2567 - accuracy: 0.9161 - val_loss: 0.2523 - val_accuracy: 0.9204\n",
            "Epoch 83/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2556 - accuracy: 0.9157 - val_loss: 0.2545 - val_accuracy: 0.9186\n",
            "Epoch 84/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2544 - accuracy: 0.9167 - val_loss: 0.2510 - val_accuracy: 0.9228\n",
            "Epoch 85/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2570 - accuracy: 0.9149 - val_loss: 0.2656 - val_accuracy: 0.9140\n",
            "Epoch 86/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2539 - accuracy: 0.9162 - val_loss: 0.2650 - val_accuracy: 0.9148\n",
            "Epoch 87/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2549 - accuracy: 0.9156 - val_loss: 0.2802 - val_accuracy: 0.9020\n",
            "Epoch 88/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2551 - accuracy: 0.9149 - val_loss: 0.2623 - val_accuracy: 0.9160\n",
            "Epoch 89/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2525 - accuracy: 0.9168 - val_loss: 0.2608 - val_accuracy: 0.9110\n",
            "Epoch 90/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2537 - accuracy: 0.9157 - val_loss: 0.2467 - val_accuracy: 0.9208\n",
            "Epoch 91/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2547 - accuracy: 0.9150 - val_loss: 0.2623 - val_accuracy: 0.9168\n",
            "Epoch 92/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2557 - accuracy: 0.9150 - val_loss: 0.2525 - val_accuracy: 0.9168\n",
            "Epoch 93/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2540 - accuracy: 0.9156 - val_loss: 0.2554 - val_accuracy: 0.9162\n",
            "Epoch 94/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2544 - accuracy: 0.9158 - val_loss: 0.2612 - val_accuracy: 0.9140\n",
            "Epoch 95/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2533 - accuracy: 0.9162 - val_loss: 0.2774 - val_accuracy: 0.9124\n",
            "Epoch 96/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2549 - accuracy: 0.9136 - val_loss: 0.2577 - val_accuracy: 0.9142\n",
            "Epoch 97/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2533 - accuracy: 0.9158 - val_loss: 0.2636 - val_accuracy: 0.9128\n",
            "Epoch 98/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2536 - accuracy: 0.9172 - val_loss: 0.2511 - val_accuracy: 0.9176\n",
            "Epoch 99/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2544 - accuracy: 0.9157 - val_loss: 0.2787 - val_accuracy: 0.9074\n",
            "Epoch 100/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2536 - accuracy: 0.9159 - val_loss: 0.2551 - val_accuracy: 0.9158\n",
            "Epoch 1/100\n",
            "469/469 [==============================] - 4s 8ms/step - loss: 55.2221 - accuracy: 0.8176 - val_loss: 6.3071 - val_accuracy: 0.8414\n",
            "Epoch 2/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 2.5048 - accuracy: 0.8741 - val_loss: 1.1013 - val_accuracy: 0.8502\n",
            "Epoch 3/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.7132 - accuracy: 0.8957 - val_loss: 0.5503 - val_accuracy: 0.8950\n",
            "Epoch 4/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.4678 - accuracy: 0.9015 - val_loss: 0.4582 - val_accuracy: 0.8758\n",
            "Epoch 5/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.3884 - accuracy: 0.9053 - val_loss: 0.3587 - val_accuracy: 0.9092\n",
            "Epoch 6/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.3510 - accuracy: 0.9069 - val_loss: 0.3362 - val_accuracy: 0.9028\n",
            "Epoch 7/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3308 - accuracy: 0.9071 - val_loss: 0.3298 - val_accuracy: 0.9002\n",
            "Epoch 8/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3157 - accuracy: 0.9081 - val_loss: 0.3131 - val_accuracy: 0.9058\n",
            "Epoch 9/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3078 - accuracy: 0.9101 - val_loss: 0.3124 - val_accuracy: 0.8948\n",
            "Epoch 10/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.3014 - accuracy: 0.9096 - val_loss: 0.2885 - val_accuracy: 0.9112\n",
            "Epoch 11/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2961 - accuracy: 0.9103 - val_loss: 0.3187 - val_accuracy: 0.8986\n",
            "Epoch 12/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2919 - accuracy: 0.9103 - val_loss: 0.2997 - val_accuracy: 0.9018\n",
            "Epoch 13/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2908 - accuracy: 0.9090 - val_loss: 0.2855 - val_accuracy: 0.9080\n",
            "Epoch 14/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2892 - accuracy: 0.9097 - val_loss: 0.2870 - val_accuracy: 0.9048\n",
            "Epoch 15/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2842 - accuracy: 0.9127 - val_loss: 0.2746 - val_accuracy: 0.9140\n",
            "Epoch 16/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2850 - accuracy: 0.9121 - val_loss: 0.3761 - val_accuracy: 0.8682\n",
            "Epoch 17/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2827 - accuracy: 0.9111 - val_loss: 0.2683 - val_accuracy: 0.9152\n",
            "Epoch 18/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2806 - accuracy: 0.9115 - val_loss: 0.2701 - val_accuracy: 0.9138\n",
            "Epoch 19/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2789 - accuracy: 0.9129 - val_loss: 0.2742 - val_accuracy: 0.9130\n",
            "Epoch 20/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2788 - accuracy: 0.9137 - val_loss: 0.2642 - val_accuracy: 0.9138\n",
            "Epoch 21/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2783 - accuracy: 0.9119 - val_loss: 0.2658 - val_accuracy: 0.9146\n",
            "Epoch 22/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2761 - accuracy: 0.9125 - val_loss: 0.2665 - val_accuracy: 0.9146\n",
            "Epoch 23/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2739 - accuracy: 0.9147 - val_loss: 0.2843 - val_accuracy: 0.9146\n",
            "Epoch 24/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2740 - accuracy: 0.9126 - val_loss: 0.2913 - val_accuracy: 0.9076\n",
            "Epoch 25/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2747 - accuracy: 0.9117 - val_loss: 0.2613 - val_accuracy: 0.9158\n",
            "Epoch 26/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2735 - accuracy: 0.9137 - val_loss: 0.2804 - val_accuracy: 0.9130\n",
            "Epoch 27/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2733 - accuracy: 0.9126 - val_loss: 0.2614 - val_accuracy: 0.9168\n",
            "Epoch 28/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2714 - accuracy: 0.9138 - val_loss: 0.3049 - val_accuracy: 0.8984\n",
            "Epoch 29/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2724 - accuracy: 0.9143 - val_loss: 0.2701 - val_accuracy: 0.9068\n",
            "Epoch 30/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2705 - accuracy: 0.9128 - val_loss: 0.2686 - val_accuracy: 0.9088\n",
            "Epoch 31/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2723 - accuracy: 0.9141 - val_loss: 0.2842 - val_accuracy: 0.9070\n",
            "Epoch 32/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2713 - accuracy: 0.9136 - val_loss: 0.2671 - val_accuracy: 0.9096\n",
            "Epoch 33/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2690 - accuracy: 0.9144 - val_loss: 0.2644 - val_accuracy: 0.9112\n",
            "Epoch 34/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2700 - accuracy: 0.9130 - val_loss: 0.2703 - val_accuracy: 0.9126\n",
            "Epoch 35/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2688 - accuracy: 0.9135 - val_loss: 0.2599 - val_accuracy: 0.9126\n",
            "Epoch 36/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2678 - accuracy: 0.9152 - val_loss: 0.2527 - val_accuracy: 0.9188\n",
            "Epoch 37/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2665 - accuracy: 0.9145 - val_loss: 0.2609 - val_accuracy: 0.9124\n",
            "Epoch 38/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2661 - accuracy: 0.9152 - val_loss: 0.2675 - val_accuracy: 0.9100\n",
            "Epoch 39/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2654 - accuracy: 0.9154 - val_loss: 0.2832 - val_accuracy: 0.9018\n",
            "Epoch 40/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2656 - accuracy: 0.9144 - val_loss: 0.2587 - val_accuracy: 0.9214\n",
            "Epoch 41/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2658 - accuracy: 0.9146 - val_loss: 0.2902 - val_accuracy: 0.8994\n",
            "Epoch 42/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2650 - accuracy: 0.9144 - val_loss: 0.3098 - val_accuracy: 0.8958\n",
            "Epoch 43/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2649 - accuracy: 0.9138 - val_loss: 0.2573 - val_accuracy: 0.9098\n",
            "Epoch 44/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2646 - accuracy: 0.9133 - val_loss: 0.2460 - val_accuracy: 0.9226\n",
            "Epoch 45/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2636 - accuracy: 0.9149 - val_loss: 0.2577 - val_accuracy: 0.9112\n",
            "Epoch 46/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2630 - accuracy: 0.9149 - val_loss: 0.2511 - val_accuracy: 0.9174\n",
            "Epoch 47/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2640 - accuracy: 0.9146 - val_loss: 0.2520 - val_accuracy: 0.9182\n",
            "Epoch 48/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2626 - accuracy: 0.9141 - val_loss: 0.2723 - val_accuracy: 0.9106\n",
            "Epoch 49/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2637 - accuracy: 0.9153 - val_loss: 0.2713 - val_accuracy: 0.9086\n",
            "Epoch 50/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2630 - accuracy: 0.9134 - val_loss: 0.2592 - val_accuracy: 0.9110\n",
            "Epoch 51/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2638 - accuracy: 0.9135 - val_loss: 0.2547 - val_accuracy: 0.9108\n",
            "Epoch 52/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2638 - accuracy: 0.9143 - val_loss: 0.2617 - val_accuracy: 0.9106\n",
            "Epoch 53/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2636 - accuracy: 0.9136 - val_loss: 0.2804 - val_accuracy: 0.9030\n",
            "Epoch 54/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2626 - accuracy: 0.9150 - val_loss: 0.2807 - val_accuracy: 0.9036\n",
            "Epoch 55/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2615 - accuracy: 0.9155 - val_loss: 0.2528 - val_accuracy: 0.9146\n",
            "Epoch 56/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2616 - accuracy: 0.9139 - val_loss: 0.2575 - val_accuracy: 0.9142\n",
            "Epoch 57/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2633 - accuracy: 0.9140 - val_loss: 0.2536 - val_accuracy: 0.9132\n",
            "Epoch 58/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2610 - accuracy: 0.9155 - val_loss: 0.2611 - val_accuracy: 0.9118\n",
            "Epoch 59/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2617 - accuracy: 0.9145 - val_loss: 0.2561 - val_accuracy: 0.9116\n",
            "Epoch 60/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2617 - accuracy: 0.9141 - val_loss: 0.2554 - val_accuracy: 0.9148\n",
            "Epoch 61/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2634 - accuracy: 0.9163 - val_loss: 0.2558 - val_accuracy: 0.9192\n",
            "Epoch 62/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2591 - accuracy: 0.9147 - val_loss: 0.2612 - val_accuracy: 0.9122\n",
            "Epoch 63/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2597 - accuracy: 0.9141 - val_loss: 0.2594 - val_accuracy: 0.9120\n",
            "Epoch 64/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2594 - accuracy: 0.9149 - val_loss: 0.2525 - val_accuracy: 0.9168\n",
            "Epoch 65/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2601 - accuracy: 0.9160 - val_loss: 0.2527 - val_accuracy: 0.9158\n",
            "Epoch 66/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2616 - accuracy: 0.9150 - val_loss: 0.2462 - val_accuracy: 0.9188\n",
            "Epoch 67/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2601 - accuracy: 0.9144 - val_loss: 0.2468 - val_accuracy: 0.9196\n",
            "Epoch 68/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2594 - accuracy: 0.9150 - val_loss: 0.2474 - val_accuracy: 0.9186\n",
            "Epoch 69/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2582 - accuracy: 0.9145 - val_loss: 0.2470 - val_accuracy: 0.9182\n",
            "Epoch 70/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2578 - accuracy: 0.9153 - val_loss: 0.2560 - val_accuracy: 0.9172\n",
            "Epoch 71/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2585 - accuracy: 0.9165 - val_loss: 0.2941 - val_accuracy: 0.8966\n",
            "Epoch 72/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2580 - accuracy: 0.9157 - val_loss: 0.2512 - val_accuracy: 0.9182\n",
            "Epoch 73/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2579 - accuracy: 0.9156 - val_loss: 0.2642 - val_accuracy: 0.9086\n",
            "Epoch 74/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2572 - accuracy: 0.9160 - val_loss: 0.2559 - val_accuracy: 0.9148\n",
            "Epoch 75/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2594 - accuracy: 0.9162 - val_loss: 0.2442 - val_accuracy: 0.9198\n",
            "Epoch 76/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2551 - accuracy: 0.9158 - val_loss: 0.2564 - val_accuracy: 0.9160\n",
            "Epoch 77/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2577 - accuracy: 0.9151 - val_loss: 0.2516 - val_accuracy: 0.9180\n",
            "Epoch 78/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2575 - accuracy: 0.9153 - val_loss: 0.2436 - val_accuracy: 0.9196\n",
            "Epoch 79/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2567 - accuracy: 0.9164 - val_loss: 0.2586 - val_accuracy: 0.9102\n",
            "Epoch 80/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2582 - accuracy: 0.9151 - val_loss: 0.2600 - val_accuracy: 0.9082\n",
            "Epoch 81/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2571 - accuracy: 0.9158 - val_loss: 0.2638 - val_accuracy: 0.9130\n",
            "Epoch 82/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2568 - accuracy: 0.9153 - val_loss: 0.2560 - val_accuracy: 0.9140\n",
            "Epoch 83/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2555 - accuracy: 0.9166 - val_loss: 0.2444 - val_accuracy: 0.9176\n",
            "Epoch 84/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2561 - accuracy: 0.9166 - val_loss: 0.2620 - val_accuracy: 0.9138\n",
            "Epoch 85/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2555 - accuracy: 0.9160 - val_loss: 0.2557 - val_accuracy: 0.9166\n",
            "Epoch 86/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2556 - accuracy: 0.9157 - val_loss: 0.2525 - val_accuracy: 0.9188\n",
            "Epoch 87/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2573 - accuracy: 0.9152 - val_loss: 0.2534 - val_accuracy: 0.9190\n",
            "Epoch 88/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2571 - accuracy: 0.9158 - val_loss: 0.2599 - val_accuracy: 0.9172\n",
            "Epoch 89/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2558 - accuracy: 0.9158 - val_loss: 0.2504 - val_accuracy: 0.9160\n",
            "Epoch 90/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2572 - accuracy: 0.9151 - val_loss: 0.2457 - val_accuracy: 0.9216\n",
            "Epoch 91/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2546 - accuracy: 0.9166 - val_loss: 0.2505 - val_accuracy: 0.9154\n",
            "Epoch 92/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2550 - accuracy: 0.9156 - val_loss: 0.2497 - val_accuracy: 0.9160\n",
            "Epoch 93/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2544 - accuracy: 0.9163 - val_loss: 0.2463 - val_accuracy: 0.9218\n",
            "Epoch 94/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2545 - accuracy: 0.9164 - val_loss: 0.2458 - val_accuracy: 0.9196\n",
            "Epoch 95/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2559 - accuracy: 0.9152 - val_loss: 0.2448 - val_accuracy: 0.9186\n",
            "Epoch 96/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2543 - accuracy: 0.9169 - val_loss: 0.2473 - val_accuracy: 0.9156\n",
            "Epoch 97/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2561 - accuracy: 0.9156 - val_loss: 0.2585 - val_accuracy: 0.9068\n",
            "Epoch 98/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2544 - accuracy: 0.9166 - val_loss: 0.2480 - val_accuracy: 0.9118\n",
            "Epoch 99/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2537 - accuracy: 0.9164 - val_loss: 0.2485 - val_accuracy: 0.9178\n",
            "Epoch 100/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2528 - accuracy: 0.9169 - val_loss: 0.2496 - val_accuracy: 0.9240\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "APjiCN3gSQXw"
      },
      "source": [
        "**Print evaluation metrics**"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "gyEpFzBVSQcu",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "a4ddf761-74a1-4cbc-d021-bd7c53da21d9"
      },
      "source": [
        "print('Accuracies are:',Accuracies)\n",
        "print('Precisions are:',Precisions)\n",
        "print('Recalls are:',Recalls)\n",
        "print('AUCs are:', AUCs)"
      ],
      "execution_count": 7,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Accuracies are: [0.9116 0.9148 0.917  0.9144 0.9198 0.915 ]\n",
            "Precisions are: [0.86808957 0.87000726 0.88956127 0.89617908 0.92396497 0.89766194]\n",
            "Recalls are: [0.97446976 0.9724026  0.95030303 0.93591294 0.91702884 0.9368    ]\n",
            "AUCs are: [0.96670569 0.97138279 0.96837564 0.96762429 0.97234695 0.97126432]\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "5UP80n4E8Jd2"
      },
      "source": [
        "### Import data in the original split"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "VlSR07Mu8PZY"
      },
      "source": [
        "# For image-based classification\n",
        "X_tr = np.float64(np.load(\"/content/drive/My Drive/Galaxies_Data/X_train.npy\")) #Training\n",
        "X_val = np.float64(np.load(\"/content/drive/My Drive/Galaxies_Data/X_val.npy\")) #Validation\n",
        "X_test = np.float64(np.load(\"/content/drive/My Drive/Galaxies_Data/X_test.npy\")) #Test\n",
        "\n",
        "# Labels\n",
        "y_tr = np.float64(np.load(\"/content/drive/My Drive/Galaxies_Data/y_train.npy\")) # Training\n",
        "y_val = np.float64(np.load(\"/content/drive/My Drive/Galaxies_Data/y_val.npy\")) #Validation\n",
        "y_test = np.float64(np.load(\"/content/drive/My Drive/Galaxies_Data/y_test.npy\")) #Test"
      ],
      "execution_count": 11,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "YomCP8so8TvU"
      },
      "source": [
        "# Clear the backend session\n",
        "\n",
        "tf.keras.backend.clear_session()\n",
        "# build the  original model\n",
        "model = keras.models.Sequential()\n",
        "model.add(InputLayer((64,64,3)))\n",
        "model.add(Conv2D(filters=16, kernel_size=(3,3), padding='same', activation='relu',kernel_regularizer=regularizers.l2(0.13)))\n",
        "model.add(BatchNormalization())\n",
        "model.add(MaxPool2D(pool_size=(2, 2), strides=None, padding='valid'))\n",
        "model.add(Dropout(rate=0.4))\n",
        "model.add(Conv2D(filters=2*16, kernel_size=(3,3), padding='same', activation='relu',kernel_regularizer=regularizers.l2(0.13)))\n",
        "model.add(BatchNormalization())\n",
        "model.add(MaxPool2D(pool_size=(2, 2), strides=None, padding='valid'))\n",
        "model.add(Dropout(rate=0.4))\n",
        "model.add(Conv2D(filters=2*32, kernel_size=(3,3), padding='same', activation='relu',kernel_regularizer=regularizers.l2(0.13)))\n",
        "model.add(BatchNormalization())\n",
        "model.add(MaxPool2D(pool_size=(2, 2), strides=None, padding='valid'))\n",
        "model.add(Dropout(rate=0.4))\n",
        "model.add(Flatten())\n",
        "model.add(Dense(units=1024, activation='relu',kernel_regularizer=regularizers.l2(0.12)))\n",
        "model.add(Dense(units=1, activation='sigmoid'))\n",
        "model.compile(optimizer=optimizers.Adadelta(0.1),\n",
        "              loss= 'binary_crossentropy',\n",
        "              metrics=['accuracy'])"
      ],
      "execution_count": 12,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "SEsJIpqX8eyS",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "333591f2-3b30-4f8e-e3d7-5c1933aa8719"
      },
      "source": [
        "model.fit(x=X_tr, y=y_tr,\n",
        "          epochs=100, batch_size=64,\n",
        "          shuffle=True,\n",
        "          validation_data=(X_val,y_val))"
      ],
      "execution_count": 13,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Epoch 1/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 60.2342 - accuracy: 0.8029 - val_loss: 7.7306 - val_accuracy: 0.8620\n",
            "Epoch 2/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 2.9623 - accuracy: 0.8705 - val_loss: 1.2512 - val_accuracy: 0.8216\n",
            "Epoch 3/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.7781 - accuracy: 0.8911 - val_loss: 0.6225 - val_accuracy: 0.8716\n",
            "Epoch 4/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.4846 - accuracy: 0.8998 - val_loss: 0.4217 - val_accuracy: 0.9032\n",
            "Epoch 5/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3935 - accuracy: 0.9045 - val_loss: 0.3782 - val_accuracy: 0.9008\n",
            "Epoch 6/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3524 - accuracy: 0.9068 - val_loss: 0.3590 - val_accuracy: 0.8968\n",
            "Epoch 7/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3316 - accuracy: 0.9044 - val_loss: 0.3139 - val_accuracy: 0.9102\n",
            "Epoch 8/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3166 - accuracy: 0.9090 - val_loss: 0.3119 - val_accuracy: 0.9030\n",
            "Epoch 9/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3079 - accuracy: 0.9077 - val_loss: 0.3336 - val_accuracy: 0.8958\n",
            "Epoch 10/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3024 - accuracy: 0.9091 - val_loss: 0.3150 - val_accuracy: 0.9068\n",
            "Epoch 11/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2975 - accuracy: 0.9102 - val_loss: 0.3110 - val_accuracy: 0.9056\n",
            "Epoch 12/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2933 - accuracy: 0.9092 - val_loss: 0.3044 - val_accuracy: 0.9052\n",
            "Epoch 13/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2919 - accuracy: 0.9100 - val_loss: 0.3224 - val_accuracy: 0.8992\n",
            "Epoch 14/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2899 - accuracy: 0.9102 - val_loss: 0.2914 - val_accuracy: 0.9068\n",
            "Epoch 15/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2871 - accuracy: 0.9113 - val_loss: 0.2849 - val_accuracy: 0.9098\n",
            "Epoch 16/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2858 - accuracy: 0.9118 - val_loss: 0.2902 - val_accuracy: 0.9116\n",
            "Epoch 17/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2827 - accuracy: 0.9128 - val_loss: 0.2876 - val_accuracy: 0.9086\n",
            "Epoch 18/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2837 - accuracy: 0.9118 - val_loss: 0.3019 - val_accuracy: 0.9038\n",
            "Epoch 19/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2822 - accuracy: 0.9135 - val_loss: 0.2801 - val_accuracy: 0.9118\n",
            "Epoch 20/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2821 - accuracy: 0.9113 - val_loss: 0.2769 - val_accuracy: 0.9108\n",
            "Epoch 21/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2785 - accuracy: 0.9133 - val_loss: 0.2738 - val_accuracy: 0.9120\n",
            "Epoch 22/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2781 - accuracy: 0.9120 - val_loss: 0.2903 - val_accuracy: 0.9052\n",
            "Epoch 23/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2786 - accuracy: 0.9128 - val_loss: 0.3045 - val_accuracy: 0.9020\n",
            "Epoch 24/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2763 - accuracy: 0.9121 - val_loss: 0.2956 - val_accuracy: 0.9072\n",
            "Epoch 25/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2770 - accuracy: 0.9127 - val_loss: 0.2722 - val_accuracy: 0.9114\n",
            "Epoch 26/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2744 - accuracy: 0.9139 - val_loss: 0.2669 - val_accuracy: 0.9112\n",
            "Epoch 27/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2723 - accuracy: 0.9131 - val_loss: 0.2695 - val_accuracy: 0.9110\n",
            "Epoch 28/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2747 - accuracy: 0.9121 - val_loss: 0.2703 - val_accuracy: 0.9114\n",
            "Epoch 29/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2726 - accuracy: 0.9135 - val_loss: 0.2620 - val_accuracy: 0.9154\n",
            "Epoch 30/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2721 - accuracy: 0.9137 - val_loss: 0.2927 - val_accuracy: 0.9064\n",
            "Epoch 31/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2709 - accuracy: 0.9138 - val_loss: 0.2792 - val_accuracy: 0.9066\n",
            "Epoch 32/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2709 - accuracy: 0.9139 - val_loss: 0.2749 - val_accuracy: 0.9122\n",
            "Epoch 33/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2712 - accuracy: 0.9130 - val_loss: 0.2627 - val_accuracy: 0.9142\n",
            "Epoch 34/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2701 - accuracy: 0.9140 - val_loss: 0.2770 - val_accuracy: 0.9124\n",
            "Epoch 35/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2700 - accuracy: 0.9137 - val_loss: 0.2716 - val_accuracy: 0.9086\n",
            "Epoch 36/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2700 - accuracy: 0.9136 - val_loss: 0.2723 - val_accuracy: 0.9102\n",
            "Epoch 37/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2690 - accuracy: 0.9143 - val_loss: 0.2675 - val_accuracy: 0.9122\n",
            "Epoch 38/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2674 - accuracy: 0.9150 - val_loss: 0.2686 - val_accuracy: 0.9126\n",
            "Epoch 39/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2694 - accuracy: 0.9142 - val_loss: 0.2715 - val_accuracy: 0.9086\n",
            "Epoch 40/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2682 - accuracy: 0.9133 - val_loss: 0.2953 - val_accuracy: 0.9058\n",
            "Epoch 41/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2679 - accuracy: 0.9145 - val_loss: 0.2674 - val_accuracy: 0.9120\n",
            "Epoch 42/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2665 - accuracy: 0.9137 - val_loss: 0.2660 - val_accuracy: 0.9118\n",
            "Epoch 43/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2670 - accuracy: 0.9141 - val_loss: 0.2659 - val_accuracy: 0.9118\n",
            "Epoch 44/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2662 - accuracy: 0.9133 - val_loss: 0.2594 - val_accuracy: 0.9144\n",
            "Epoch 45/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2663 - accuracy: 0.9146 - val_loss: 0.2644 - val_accuracy: 0.9094\n",
            "Epoch 46/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2645 - accuracy: 0.9144 - val_loss: 0.2736 - val_accuracy: 0.9130\n",
            "Epoch 47/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2651 - accuracy: 0.9145 - val_loss: 0.3526 - val_accuracy: 0.8778\n",
            "Epoch 48/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2646 - accuracy: 0.9145 - val_loss: 0.2686 - val_accuracy: 0.9120\n",
            "Epoch 49/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2623 - accuracy: 0.9160 - val_loss: 0.2632 - val_accuracy: 0.9102\n",
            "Epoch 50/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2663 - accuracy: 0.9136 - val_loss: 0.2695 - val_accuracy: 0.9102\n",
            "Epoch 51/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2635 - accuracy: 0.9139 - val_loss: 0.2625 - val_accuracy: 0.9116\n",
            "Epoch 52/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2620 - accuracy: 0.9153 - val_loss: 0.2603 - val_accuracy: 0.9144\n",
            "Epoch 53/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2638 - accuracy: 0.9154 - val_loss: 0.2886 - val_accuracy: 0.9052\n",
            "Epoch 54/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2635 - accuracy: 0.9149 - val_loss: 0.2622 - val_accuracy: 0.9142\n",
            "Epoch 55/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2633 - accuracy: 0.9150 - val_loss: 0.3525 - val_accuracy: 0.8788\n",
            "Epoch 56/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2651 - accuracy: 0.9158 - val_loss: 0.2669 - val_accuracy: 0.9122\n",
            "Epoch 57/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2626 - accuracy: 0.9152 - val_loss: 0.2660 - val_accuracy: 0.9086\n",
            "Epoch 58/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2615 - accuracy: 0.9161 - val_loss: 0.2640 - val_accuracy: 0.9112\n",
            "Epoch 59/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2644 - accuracy: 0.9140 - val_loss: 0.2843 - val_accuracy: 0.9046\n",
            "Epoch 60/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2650 - accuracy: 0.9143 - val_loss: 0.2631 - val_accuracy: 0.9128\n",
            "Epoch 61/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2618 - accuracy: 0.9158 - val_loss: 0.2728 - val_accuracy: 0.9140\n",
            "Epoch 62/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2629 - accuracy: 0.9149 - val_loss: 0.2628 - val_accuracy: 0.9126\n",
            "Epoch 63/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2612 - accuracy: 0.9157 - val_loss: 0.2575 - val_accuracy: 0.9122\n",
            "Epoch 64/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2609 - accuracy: 0.9151 - val_loss: 0.2760 - val_accuracy: 0.9088\n",
            "Epoch 65/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2611 - accuracy: 0.9159 - val_loss: 0.2651 - val_accuracy: 0.9138\n",
            "Epoch 66/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2615 - accuracy: 0.9149 - val_loss: 0.2622 - val_accuracy: 0.9118\n",
            "Epoch 67/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2623 - accuracy: 0.9147 - val_loss: 0.2839 - val_accuracy: 0.9092\n",
            "Epoch 68/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2621 - accuracy: 0.9138 - val_loss: 0.2601 - val_accuracy: 0.9122\n",
            "Epoch 69/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2605 - accuracy: 0.9145 - val_loss: 0.2793 - val_accuracy: 0.9104\n",
            "Epoch 70/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2595 - accuracy: 0.9173 - val_loss: 0.2586 - val_accuracy: 0.9130\n",
            "Epoch 71/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2604 - accuracy: 0.9161 - val_loss: 0.2618 - val_accuracy: 0.9116\n",
            "Epoch 72/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2612 - accuracy: 0.9153 - val_loss: 0.2566 - val_accuracy: 0.9156\n",
            "Epoch 73/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2616 - accuracy: 0.9143 - val_loss: 0.2736 - val_accuracy: 0.9070\n",
            "Epoch 74/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2613 - accuracy: 0.9145 - val_loss: 0.2564 - val_accuracy: 0.9144\n",
            "Epoch 75/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2605 - accuracy: 0.9144 - val_loss: 0.2560 - val_accuracy: 0.9138\n",
            "Epoch 76/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2605 - accuracy: 0.9142 - val_loss: 0.2551 - val_accuracy: 0.9182\n",
            "Epoch 77/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2597 - accuracy: 0.9149 - val_loss: 0.2616 - val_accuracy: 0.9084\n",
            "Epoch 78/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2578 - accuracy: 0.9168 - val_loss: 0.2631 - val_accuracy: 0.9100\n",
            "Epoch 79/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2577 - accuracy: 0.9133 - val_loss: 0.2760 - val_accuracy: 0.9098\n",
            "Epoch 80/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2609 - accuracy: 0.9137 - val_loss: 0.2570 - val_accuracy: 0.9180\n",
            "Epoch 81/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2576 - accuracy: 0.9142 - val_loss: 0.2632 - val_accuracy: 0.9118\n",
            "Epoch 82/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2588 - accuracy: 0.9160 - val_loss: 0.2774 - val_accuracy: 0.9074\n",
            "Epoch 83/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2567 - accuracy: 0.9155 - val_loss: 0.2765 - val_accuracy: 0.9046\n",
            "Epoch 84/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2580 - accuracy: 0.9154 - val_loss: 0.2657 - val_accuracy: 0.9120\n",
            "Epoch 85/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2585 - accuracy: 0.9163 - val_loss: 0.2495 - val_accuracy: 0.9154\n",
            "Epoch 86/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2556 - accuracy: 0.9166 - val_loss: 0.2711 - val_accuracy: 0.9086\n",
            "Epoch 87/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2559 - accuracy: 0.9165 - val_loss: 0.2755 - val_accuracy: 0.9048\n",
            "Epoch 88/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2572 - accuracy: 0.9153 - val_loss: 0.2586 - val_accuracy: 0.9094\n",
            "Epoch 89/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2570 - accuracy: 0.9164 - val_loss: 0.2584 - val_accuracy: 0.9096\n",
            "Epoch 90/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2583 - accuracy: 0.9157 - val_loss: 0.2526 - val_accuracy: 0.9158\n",
            "Epoch 91/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2547 - accuracy: 0.9165 - val_loss: 0.2605 - val_accuracy: 0.9128\n",
            "Epoch 92/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2555 - accuracy: 0.9162 - val_loss: 0.2824 - val_accuracy: 0.9026\n",
            "Epoch 93/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2541 - accuracy: 0.9159 - val_loss: 0.2620 - val_accuracy: 0.9066\n",
            "Epoch 94/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2569 - accuracy: 0.9154 - val_loss: 0.2748 - val_accuracy: 0.9092\n",
            "Epoch 95/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2583 - accuracy: 0.9163 - val_loss: 0.2876 - val_accuracy: 0.9038\n",
            "Epoch 96/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2537 - accuracy: 0.9153 - val_loss: 0.2572 - val_accuracy: 0.9132\n",
            "Epoch 97/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2549 - accuracy: 0.9154 - val_loss: 0.2603 - val_accuracy: 0.9126\n",
            "Epoch 98/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2556 - accuracy: 0.9157 - val_loss: 0.2575 - val_accuracy: 0.9104\n",
            "Epoch 99/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2551 - accuracy: 0.9170 - val_loss: 0.2605 - val_accuracy: 0.9128\n",
            "Epoch 100/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2546 - accuracy: 0.9171 - val_loss: 0.2582 - val_accuracy: 0.9136\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "<tensorflow.python.keras.callbacks.History at 0x7f766c2afba8>"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 13
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "7n9X2lQS8jEN"
      },
      "source": [
        "Predict now and calculate ROC"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "oo6Wc8sG8jK5"
      },
      "source": [
        "# Predict classes\n",
        "y_pred = model.predict_classes(X_test)\n",
        "# Predict probability scores\n",
        "scores = model.predict(X_test)"
      ],
      "execution_count": 14,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "yyFU_pCi8jPr"
      },
      "source": [
        "# fpr, tpr for roc curve plot\n",
        "fpr, tpr, thresholds = roc_curve(y_test, scores[:,0], pos_label=1.0)\n",
        "\n",
        "fpr_unique, ind_unique = np.unique(fpr,return_index=True)\n",
        "tpr_unique = tpr[ind_unique]\n",
        "\n",
        "# Interpolate the false positives - true positives\n",
        "interp = interpolate.interp1d(fpr_unique, tpr_unique, kind='linear')\n",
        "tpr_baseline = interp(fprs)"
      ],
      "execution_count": 15,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "q06eWFP26VDj",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 514
        },
        "outputId": "4f27960a-6ad4-4f8b-f084-0ac025dc5d10"
      },
      "source": [
        "custom_blues = [\"#66CCFF\", \"#33BBFF\", \"#00AAFF\", \"#0088CC\", \"#006699\", \"#004466\"]\n",
        "\n",
        "x = np.linspace(0,1,100)\n",
        "\n",
        "\n",
        "fig, ax1 = plt.subplots(figsize = (6.5,6.5))\n",
        "# =============================================================\n",
        "# =============================================================\n",
        "# Main figure\n",
        "\n",
        "ax1.plot(fprs,tpr_baseline,c='coral',linewidth=2.1, label='Baseline split')\n",
        "ax1.plot(fprs,tprs[0],c=custom_blues[1],linewidth=0.8)\n",
        "ax1.plot(fprs,tprs[1],c=custom_blues[1],linewidth=0.8)\n",
        "ax1.plot(fprs,tprs[2],c=custom_blues[1],linewidth=0.8,label='Other random seeds')\n",
        "ax1.plot(fprs,tprs[3],c=custom_blues[1],linewidth=0.8)\n",
        "ax1.plot(fprs,tprs[4],c=custom_blues[1],linewidth=0.8)\n",
        "ax1.plot(fprs,tprs[5],c=custom_blues[1],linewidth=0.8)\n",
        "\n",
        "\n",
        "\n",
        "#Plot diagonal\n",
        "ax1.plot(x,x, c='k',ls='--',linewidth=0.8)\n",
        "ax1.grid(ls='--', alpha=0.6)\n",
        "\n",
        "ax1.set_xlabel('False Positive Rate',fontsize=14);ax1.set_ylabel('True Positive Rate',fontsize=14)\n",
        "ax1.tick_params(axis='both', labelsize=13)\n",
        "\n",
        "\n",
        "ax1.legend(frameon=True, loc='lower right', fontsize=14)\n",
        "# =============================================================\n",
        "# =============================================================\n",
        "# Inset figure\n",
        "left, bottom, width, height = [0.53, 0.33, 0.35, 0.35]\n",
        "ax2 = fig.add_axes([left, bottom, width, height])\n",
        "\n",
        "ax2.plot(fprs,tpr_baseline,c='coral',linewidth=2.2)\n",
        "ax2.plot(fprs,tprs[0],c=custom_blues[1],linewidth=0.9)\n",
        "ax2.plot(fprs,tprs[1],c=custom_blues[1],linewidth=0.9)\n",
        "ax2.plot(fprs,tprs[2],c=custom_blues[1],linewidth=0.9)\n",
        "ax2.plot(fprs,tprs[3],c=custom_blues[1],linewidth=0.9)\n",
        "ax2.plot(fprs,tprs[4],c=custom_blues[1],linewidth=0.9)\n",
        "ax2.plot(fprs,tprs[5],c=custom_blues[1],linewidth=0.9)\n",
        "\n",
        "\n",
        "ax2.set_xlim(0.0,0.2)\n",
        "ax2.set_ylim(0.8,1.0)\n",
        "ax2.set_xticks([0.0,0.1,0.2])\n",
        "ax2.set_yticks([0.8,0.9,1.0])\n",
        "ax2.tick_params(axis='both', labelsize=12)\n",
        "ax2.grid(ls='--', alpha=0.6)\n",
        "# =============================================================\n",
        "# =============================================================\n",
        "# Format the whole figure\n",
        "# Mark the region corresponding to the inset axes on ax1 and draw lines\n",
        "# in grey linking the two axes.\n",
        "mark_inset(ax1, ax2, loc1=1, loc2=3, fc=\"none\",ls='--', ec='0.5')\n",
        "ax1.set_title('ROC curves - Different random seeds',fontsize=14)\n",
        "plt.tight_layout()\n",
        "plt.savefig(\"/content/drive/My Drive/Plots/ROC_other_seeds.pdf\")\n",
        "#files.download(\"/content/drive/My Drive/Plots/ROC_other_seeds.pdf\")\n",
        "plt.show()"
      ],
      "execution_count": 23,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "/usr/local/lib/python3.6/dist-packages/ipykernel_launcher.py:58: UserWarning: This figure includes Axes that are not compatible with tight_layout, so results might be incorrect.\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "display_data",
          "data": {
            "application/javascript": [
              "\n",
              "    async function download(id, filename, size) {\n",
              "      if (!google.colab.kernel.accessAllowed) {\n",
              "        return;\n",
              "      }\n",
              "      const div = document.createElement('div');\n",
              "      const label = document.createElement('label');\n",
              "      label.textContent = `Downloading \"${filename}\": `;\n",
              "      div.appendChild(label);\n",
              "      const progress = document.createElement('progress');\n",
              "      progress.max = size;\n",
              "      div.appendChild(progress);\n",
              "      document.body.appendChild(div);\n",
              "\n",
              "      const buffers = [];\n",
              "      let downloaded = 0;\n",
              "\n",
              "      const channel = await google.colab.kernel.comms.open(id);\n",
              "      // Send a message to notify the kernel that we're ready.\n",
              "      channel.send({})\n",
              "\n",
              "      for await (const message of channel.messages) {\n",
              "        // Send a message to notify the kernel that we're ready.\n",
              "        channel.send({})\n",
              "        if (message.buffers) {\n",
              "          for (const buffer of message.buffers) {\n",
              "            buffers.push(buffer);\n",
              "            downloaded += buffer.byteLength;\n",
              "            progress.value = downloaded;\n",
              "          }\n",
              "        }\n",
              "      }\n",
              "      const blob = new Blob(buffers, {type: 'application/binary'});\n",
              "      const a = document.createElement('a');\n",
              "      a.href = window.URL.createObjectURL(blob);\n",
              "      a.download = filename;\n",
              "      div.appendChild(a);\n",
              "      a.click();\n",
              "      div.remove();\n",
              "    }\n",
              "  "
            ],
            "text/plain": [
              "<IPython.core.display.Javascript object>"
            ]
          },
          "metadata": {
            "tags": []
          }
        },
        {
          "output_type": "display_data",
          "data": {
            "application/javascript": [
              "download(\"download_73df84b3-c077-4a1f-9173-a8d707359ee3\", \"ROC_other_seeds.pdf\", 21905)"
            ],
            "text/plain": [
              "<IPython.core.display.Javascript object>"
            ]
          },
          "metadata": {
            "tags": []
          }
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 468x468 with 2 Axes>"
            ]
          },
          "metadata": {
            "tags": [],
            "needs_background": "light"
          }
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "XdBs5R-1vfB6"
      },
      "source": [
        "## Part II: Effect of label noise\n",
        "\n",
        "**In this part we explore the effect of the label noise.**\n",
        "\n",
        "The training set has 30000 labeled examples. To estimate the effect of label noise (random mislabeling) we randomly change the labels of i) $1\\%$, ii) $5\\%$ and iii) $10\\%$ iv) $1/3$ of the total sample. (In number of examples, then these number correspond to 300, 1500, 3000 and 10000 examples, respectively).\n",
        "\n",
        "\n",
        "### IIa) Random noise\n",
        "\n",
        "In this case the noise does not have any preference, so when the initial label is 0 it becomes 1, while when the initial lable is 1 it becomes 0."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "c2jcW4HDvfH8"
      },
      "source": [
        "# Create empty arrays for the evaluation metrics of each loop\n",
        "Accuracies_noise = np.zeros(4)\n",
        "Precisions_noise = np.zeros(4)\n",
        "Recalls_noise = np.zeros(4)\n",
        "tprs_noise = np.zeros((4,200))\n",
        "fprs_noise = np.linspace(0,1,200)\n",
        "AUCs_noise = np.zeros(4)\n",
        "\n",
        "# =====================================================\n",
        "num_examples = [300,1500,3000,10000] #Number of examples to change the labels"
      ],
      "execution_count": 24,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "7lpaQyPGn2sh",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "6b26346e-2e46-4d48-9757-a1014f5de169"
      },
      "source": [
        "for i in range(4):\n",
        "  # Clear the backend session\n",
        "  tf.keras.backend.clear_session()\n",
        "  # ============================================================================\n",
        "  # ============================================================================\n",
        "  # Number of examples to change the labels\n",
        "  n_examples = num_examples[i]\n",
        "  \n",
        "  #Get n_examples random indices from 0 to 30000\n",
        "  rands_loc = random.sample(range(0, 30000),n_examples)\n",
        "\n",
        "  y_train_loc = np.copy(y_tr) #Make a copy of y_train \n",
        "  # Change the indices \n",
        "  y_train_loc[rands_loc] = y_train_loc[rands_loc] + 1 \n",
        "  y_train_loc[y_train_loc>1] = 0\n",
        "\n",
        "  # ============================================================================\n",
        "  # ============================================================================ \n",
        "  # ============================================================================\n",
        "  # Now build the model\n",
        "  model = keras.models.Sequential()\n",
        "  model.add(InputLayer((64,64,3)))\n",
        "  model.add(Conv2D(filters=16, kernel_size=(3,3), padding='same', activation='relu',kernel_regularizer=regularizers.l2(0.13)))\n",
        "  model.add(BatchNormalization())\n",
        "  model.add(MaxPool2D(pool_size=(2, 2), strides=None, padding='valid'))\n",
        "  model.add(Dropout(rate=0.4))\n",
        "  model.add(Conv2D(filters=2*16, kernel_size=(3,3), padding='same', activation='relu',kernel_regularizer=regularizers.l2(0.13)))\n",
        "  model.add(BatchNormalization())\n",
        "  model.add(MaxPool2D(pool_size=(2, 2), strides=None, padding='valid'))\n",
        "  model.add(Dropout(rate=0.4))\n",
        "  model.add(Conv2D(filters=2*32, kernel_size=(3,3), padding='same', activation='relu',kernel_regularizer=regularizers.l2(0.13)))\n",
        "  model.add(BatchNormalization())\n",
        "  model.add(MaxPool2D(pool_size=(2, 2), strides=None, padding='valid'))\n",
        "  model.add(Dropout(rate=0.4))\n",
        "  model.add(Flatten())\n",
        "  model.add(Dense(units=1024, activation='relu',kernel_regularizer=regularizers.l2(0.12)))\n",
        "  model.add(Dense(units=1, activation='sigmoid'))\n",
        "  model.compile(optimizer=optimizers.Adadelta(0.1),\n",
        "                loss= 'binary_crossentropy',\n",
        "                metrics=['accuracy'])\n",
        "  # ============================================================================\n",
        "  # Now fit the model\n",
        "  model.fit(x=X_tr, y=y_train_loc,\n",
        "          epochs=100, batch_size=64,\n",
        "          shuffle=True,\n",
        "          validation_data=(X_val,y_val))\n",
        "  # ============================================================================\n",
        "  # ============================================================================\n",
        "  # ============================================================================\n",
        "  # Now predict on the test set\n",
        "  y_pred_loc = model.predict_classes(X_test)\n",
        "  # Predict probability scores\n",
        "  scores_loc = model.predict(X_test)\n",
        "  # ============================================================================\n",
        "  # ============================================================================\n",
        "  # ============================================================================\n",
        "  # Evaluation metrics \n",
        "\n",
        "  Accuracies_noise[i] = accuracy_score(y_test, y_pred_loc)\n",
        "  Precisions_noise[i] = precision_score(y_test, y_pred_loc)\n",
        "  Recalls_noise[i] = recall_score(y_test, y_pred_loc)\n",
        "  AUCs_noise[i] = roc_auc_score(y_test, scores_loc[:,0])\n",
        "  # ============================================================================\n",
        "  # Get now false positives and true positives\n",
        "  fpr, tpr, thresholds = roc_curve(y_test, scores_loc[:,0], pos_label=1.0)\n",
        "  fpr_unique, ind_unique = np.unique(fpr,return_index=True)\n",
        "  tpr_unique = tpr[ind_unique]\n",
        "\n",
        "  # Interpolate the false positives - true positives\n",
        "  interp = interpolate.interp1d(fpr_unique, tpr_unique, kind='linear')\n",
        "  tprs_noise[i,:] = interp(fprs_noise)"
      ],
      "execution_count": 25,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Epoch 1/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 57.9916 - accuracy: 0.7981 - val_loss: 6.9900 - val_accuracy: 0.6750\n",
            "Epoch 2/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 2.6419 - accuracy: 0.8638 - val_loss: 1.1316 - val_accuracy: 0.8136\n",
            "Epoch 3/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.7313 - accuracy: 0.8819 - val_loss: 0.5717 - val_accuracy: 0.8746\n",
            "Epoch 4/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.4893 - accuracy: 0.8905 - val_loss: 0.4347 - val_accuracy: 0.8898\n",
            "Epoch 5/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.4153 - accuracy: 0.8935 - val_loss: 0.3893 - val_accuracy: 0.8886\n",
            "Epoch 6/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3782 - accuracy: 0.8960 - val_loss: 0.3527 - val_accuracy: 0.9108\n",
            "Epoch 7/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3573 - accuracy: 0.8989 - val_loss: 0.3228 - val_accuracy: 0.9036\n",
            "Epoch 8/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3459 - accuracy: 0.8998 - val_loss: 0.3225 - val_accuracy: 0.9072\n",
            "Epoch 9/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3377 - accuracy: 0.8992 - val_loss: 0.3066 - val_accuracy: 0.9052\n",
            "Epoch 10/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3308 - accuracy: 0.9009 - val_loss: 0.3951 - val_accuracy: 0.8626\n",
            "Epoch 11/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3284 - accuracy: 0.9007 - val_loss: 0.3091 - val_accuracy: 0.9052\n",
            "Epoch 12/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3244 - accuracy: 0.9008 - val_loss: 0.3066 - val_accuracy: 0.9068\n",
            "Epoch 13/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3215 - accuracy: 0.9013 - val_loss: 0.2858 - val_accuracy: 0.9082\n",
            "Epoch 14/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3170 - accuracy: 0.9040 - val_loss: 0.2917 - val_accuracy: 0.9054\n",
            "Epoch 15/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3170 - accuracy: 0.9028 - val_loss: 0.3105 - val_accuracy: 0.8998\n",
            "Epoch 16/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3137 - accuracy: 0.9030 - val_loss: 0.3002 - val_accuracy: 0.9022\n",
            "Epoch 17/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3128 - accuracy: 0.9009 - val_loss: 0.2747 - val_accuracy: 0.9088\n",
            "Epoch 18/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3120 - accuracy: 0.9031 - val_loss: 0.2845 - val_accuracy: 0.9052\n",
            "Epoch 19/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3094 - accuracy: 0.9033 - val_loss: 0.2800 - val_accuracy: 0.9086\n",
            "Epoch 20/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3083 - accuracy: 0.9040 - val_loss: 0.2981 - val_accuracy: 0.9022\n",
            "Epoch 21/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3058 - accuracy: 0.9043 - val_loss: 0.2889 - val_accuracy: 0.9056\n",
            "Epoch 22/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3054 - accuracy: 0.9026 - val_loss: 0.2764 - val_accuracy: 0.9134\n",
            "Epoch 23/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3056 - accuracy: 0.9027 - val_loss: 0.2949 - val_accuracy: 0.9012\n",
            "Epoch 24/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3050 - accuracy: 0.9027 - val_loss: 0.2692 - val_accuracy: 0.9110\n",
            "Epoch 25/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3026 - accuracy: 0.9044 - val_loss: 0.2931 - val_accuracy: 0.9000\n",
            "Epoch 26/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3004 - accuracy: 0.9052 - val_loss: 0.2928 - val_accuracy: 0.9124\n",
            "Epoch 27/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3018 - accuracy: 0.9046 - val_loss: 0.2743 - val_accuracy: 0.9106\n",
            "Epoch 28/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3011 - accuracy: 0.9041 - val_loss: 0.2677 - val_accuracy: 0.9118\n",
            "Epoch 29/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2991 - accuracy: 0.9042 - val_loss: 0.3130 - val_accuracy: 0.8938\n",
            "Epoch 30/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2989 - accuracy: 0.9050 - val_loss: 0.2661 - val_accuracy: 0.9114\n",
            "Epoch 31/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2985 - accuracy: 0.9056 - val_loss: 0.2880 - val_accuracy: 0.9000\n",
            "Epoch 32/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2985 - accuracy: 0.9055 - val_loss: 0.2756 - val_accuracy: 0.9070\n",
            "Epoch 33/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2964 - accuracy: 0.9048 - val_loss: 0.2706 - val_accuracy: 0.9094\n",
            "Epoch 34/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2966 - accuracy: 0.9050 - val_loss: 0.2649 - val_accuracy: 0.9148\n",
            "Epoch 35/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2967 - accuracy: 0.9043 - val_loss: 0.2948 - val_accuracy: 0.8980\n",
            "Epoch 36/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2958 - accuracy: 0.9054 - val_loss: 0.2836 - val_accuracy: 0.9000\n",
            "Epoch 37/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2948 - accuracy: 0.9052 - val_loss: 0.2860 - val_accuracy: 0.9076\n",
            "Epoch 38/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2975 - accuracy: 0.9032 - val_loss: 0.2701 - val_accuracy: 0.9072\n",
            "Epoch 39/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2966 - accuracy: 0.9049 - val_loss: 0.2679 - val_accuracy: 0.9128\n",
            "Epoch 40/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2941 - accuracy: 0.9062 - val_loss: 0.2705 - val_accuracy: 0.9102\n",
            "Epoch 41/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2954 - accuracy: 0.9040 - val_loss: 0.2791 - val_accuracy: 0.9046\n",
            "Epoch 42/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2929 - accuracy: 0.9082 - val_loss: 0.2900 - val_accuracy: 0.9018\n",
            "Epoch 43/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2949 - accuracy: 0.9033 - val_loss: 0.2789 - val_accuracy: 0.9072\n",
            "Epoch 44/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2930 - accuracy: 0.9062 - val_loss: 0.2831 - val_accuracy: 0.9044\n",
            "Epoch 45/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2909 - accuracy: 0.9060 - val_loss: 0.2620 - val_accuracy: 0.9116\n",
            "Epoch 46/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2907 - accuracy: 0.9064 - val_loss: 0.2699 - val_accuracy: 0.9048\n",
            "Epoch 47/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2929 - accuracy: 0.9055 - val_loss: 0.2764 - val_accuracy: 0.9086\n",
            "Epoch 48/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2916 - accuracy: 0.9062 - val_loss: 0.2619 - val_accuracy: 0.9140\n",
            "Epoch 49/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2915 - accuracy: 0.9048 - val_loss: 0.2749 - val_accuracy: 0.9048\n",
            "Epoch 50/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2920 - accuracy: 0.9052 - val_loss: 0.2716 - val_accuracy: 0.9078\n",
            "Epoch 51/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2904 - accuracy: 0.9061 - val_loss: 0.2773 - val_accuracy: 0.9060\n",
            "Epoch 52/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2921 - accuracy: 0.9034 - val_loss: 0.2664 - val_accuracy: 0.9112\n",
            "Epoch 53/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2902 - accuracy: 0.9065 - val_loss: 0.2564 - val_accuracy: 0.9184\n",
            "Epoch 54/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2912 - accuracy: 0.9055 - val_loss: 0.2762 - val_accuracy: 0.9078\n",
            "Epoch 55/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2883 - accuracy: 0.9067 - val_loss: 0.2622 - val_accuracy: 0.9122\n",
            "Epoch 56/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2897 - accuracy: 0.9070 - val_loss: 0.2678 - val_accuracy: 0.9098\n",
            "Epoch 57/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2894 - accuracy: 0.9057 - val_loss: 0.2699 - val_accuracy: 0.9130\n",
            "Epoch 58/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2897 - accuracy: 0.9078 - val_loss: 0.2759 - val_accuracy: 0.9056\n",
            "Epoch 59/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2878 - accuracy: 0.9059 - val_loss: 0.2968 - val_accuracy: 0.8972\n",
            "Epoch 60/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2901 - accuracy: 0.9067 - val_loss: 0.2554 - val_accuracy: 0.9140\n",
            "Epoch 61/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2898 - accuracy: 0.9064 - val_loss: 0.2658 - val_accuracy: 0.9130\n",
            "Epoch 62/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2878 - accuracy: 0.9070 - val_loss: 0.2543 - val_accuracy: 0.9160\n",
            "Epoch 63/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2899 - accuracy: 0.9060 - val_loss: 0.2747 - val_accuracy: 0.9050\n",
            "Epoch 64/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2899 - accuracy: 0.9047 - val_loss: 0.2641 - val_accuracy: 0.9116\n",
            "Epoch 65/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2873 - accuracy: 0.9065 - val_loss: 0.2662 - val_accuracy: 0.9090\n",
            "Epoch 66/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2875 - accuracy: 0.9075 - val_loss: 0.2798 - val_accuracy: 0.9018\n",
            "Epoch 67/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2879 - accuracy: 0.9054 - val_loss: 0.4325 - val_accuracy: 0.8172\n",
            "Epoch 68/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2865 - accuracy: 0.9070 - val_loss: 0.2763 - val_accuracy: 0.9048\n",
            "Epoch 69/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2845 - accuracy: 0.9069 - val_loss: 0.2597 - val_accuracy: 0.9126\n",
            "Epoch 70/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2876 - accuracy: 0.9064 - val_loss: 0.2627 - val_accuracy: 0.9140\n",
            "Epoch 71/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2880 - accuracy: 0.9055 - val_loss: 0.2553 - val_accuracy: 0.9162\n",
            "Epoch 72/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.2860 - accuracy: 0.9061 - val_loss: 0.2602 - val_accuracy: 0.9126\n",
            "Epoch 73/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2878 - accuracy: 0.9057 - val_loss: 0.2570 - val_accuracy: 0.9088\n",
            "Epoch 74/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2858 - accuracy: 0.9054 - val_loss: 0.2594 - val_accuracy: 0.9178\n",
            "Epoch 75/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2845 - accuracy: 0.9059 - val_loss: 0.2773 - val_accuracy: 0.9050\n",
            "Epoch 76/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2869 - accuracy: 0.9067 - val_loss: 0.2833 - val_accuracy: 0.8978\n",
            "Epoch 77/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2866 - accuracy: 0.9064 - val_loss: 0.2588 - val_accuracy: 0.9118\n",
            "Epoch 78/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2871 - accuracy: 0.9070 - val_loss: 0.2556 - val_accuracy: 0.9114\n",
            "Epoch 79/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2849 - accuracy: 0.9055 - val_loss: 0.2648 - val_accuracy: 0.9104\n",
            "Epoch 80/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2870 - accuracy: 0.9071 - val_loss: 0.2699 - val_accuracy: 0.9118\n",
            "Epoch 81/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2868 - accuracy: 0.9058 - val_loss: 0.2602 - val_accuracy: 0.9082\n",
            "Epoch 82/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2856 - accuracy: 0.9058 - val_loss: 0.2639 - val_accuracy: 0.9104\n",
            "Epoch 83/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2835 - accuracy: 0.9063 - val_loss: 0.2616 - val_accuracy: 0.9108\n",
            "Epoch 84/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2855 - accuracy: 0.9064 - val_loss: 0.2704 - val_accuracy: 0.9114\n",
            "Epoch 85/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2855 - accuracy: 0.9060 - val_loss: 0.2502 - val_accuracy: 0.9120\n",
            "Epoch 86/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2842 - accuracy: 0.9072 - val_loss: 0.2498 - val_accuracy: 0.9160\n",
            "Epoch 87/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2842 - accuracy: 0.9070 - val_loss: 0.2628 - val_accuracy: 0.9092\n",
            "Epoch 88/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2838 - accuracy: 0.9074 - val_loss: 0.2671 - val_accuracy: 0.9068\n",
            "Epoch 89/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2831 - accuracy: 0.9057 - val_loss: 0.2595 - val_accuracy: 0.9098\n",
            "Epoch 90/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2845 - accuracy: 0.9076 - val_loss: 0.2695 - val_accuracy: 0.9084\n",
            "Epoch 91/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2822 - accuracy: 0.9081 - val_loss: 0.2557 - val_accuracy: 0.9138\n",
            "Epoch 92/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2839 - accuracy: 0.9058 - val_loss: 0.2583 - val_accuracy: 0.9104\n",
            "Epoch 93/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2828 - accuracy: 0.9082 - val_loss: 0.2599 - val_accuracy: 0.9130\n",
            "Epoch 94/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2828 - accuracy: 0.9075 - val_loss: 0.2612 - val_accuracy: 0.9104\n",
            "Epoch 95/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2833 - accuracy: 0.9072 - val_loss: 0.2589 - val_accuracy: 0.9102\n",
            "Epoch 96/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2834 - accuracy: 0.9074 - val_loss: 0.2640 - val_accuracy: 0.9126\n",
            "Epoch 97/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2840 - accuracy: 0.9064 - val_loss: 0.2607 - val_accuracy: 0.9116\n",
            "Epoch 98/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2834 - accuracy: 0.9063 - val_loss: 0.2568 - val_accuracy: 0.9116\n",
            "Epoch 99/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2832 - accuracy: 0.9076 - val_loss: 0.2558 - val_accuracy: 0.9092\n",
            "Epoch 100/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.2833 - accuracy: 0.9064 - val_loss: 0.2516 - val_accuracy: 0.9136\n",
            "Epoch 1/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 60.0744 - accuracy: 0.7593 - val_loss: 7.6399 - val_accuracy: 0.6908\n",
            "Epoch 2/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 2.9609 - accuracy: 0.8281 - val_loss: 1.1826 - val_accuracy: 0.8544\n",
            "Epoch 3/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.8593 - accuracy: 0.8473 - val_loss: 0.6690 - val_accuracy: 0.8464\n",
            "Epoch 4/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.5932 - accuracy: 0.8531 - val_loss: 0.5063 - val_accuracy: 0.8848\n",
            "Epoch 5/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.5099 - accuracy: 0.8600 - val_loss: 0.4466 - val_accuracy: 0.8732\n",
            "Epoch 6/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.4698 - accuracy: 0.8597 - val_loss: 0.3802 - val_accuracy: 0.9016\n",
            "Epoch 7/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.4487 - accuracy: 0.8612 - val_loss: 0.3520 - val_accuracy: 0.9000\n",
            "Epoch 8/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.4342 - accuracy: 0.8652 - val_loss: 0.3678 - val_accuracy: 0.8762\n",
            "Epoch 9/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.4250 - accuracy: 0.8644 - val_loss: 0.3667 - val_accuracy: 0.8816\n",
            "Epoch 10/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.4213 - accuracy: 0.8632 - val_loss: 0.3117 - val_accuracy: 0.8994\n",
            "Epoch 11/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.4180 - accuracy: 0.8645 - val_loss: 0.3222 - val_accuracy: 0.9050\n",
            "Epoch 12/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.4127 - accuracy: 0.8628 - val_loss: 0.3224 - val_accuracy: 0.8950\n",
            "Epoch 13/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.4101 - accuracy: 0.8623 - val_loss: 0.3054 - val_accuracy: 0.9080\n",
            "Epoch 14/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.4091 - accuracy: 0.8637 - val_loss: 0.3430 - val_accuracy: 0.9012\n",
            "Epoch 15/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.4065 - accuracy: 0.8667 - val_loss: 0.2987 - val_accuracy: 0.8980\n",
            "Epoch 16/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.4037 - accuracy: 0.8652 - val_loss: 0.3146 - val_accuracy: 0.9076\n",
            "Epoch 17/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.4024 - accuracy: 0.8663 - val_loss: 0.2988 - val_accuracy: 0.8974\n",
            "Epoch 18/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.4017 - accuracy: 0.8662 - val_loss: 0.3131 - val_accuracy: 0.9092\n",
            "Epoch 19/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.4009 - accuracy: 0.8657 - val_loss: 0.3166 - val_accuracy: 0.9050\n",
            "Epoch 20/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3989 - accuracy: 0.8662 - val_loss: 0.3044 - val_accuracy: 0.9092\n",
            "Epoch 21/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3993 - accuracy: 0.8664 - val_loss: 0.2876 - val_accuracy: 0.9030\n",
            "Epoch 22/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3967 - accuracy: 0.8662 - val_loss: 0.3027 - val_accuracy: 0.8992\n",
            "Epoch 23/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3959 - accuracy: 0.8657 - val_loss: 0.2834 - val_accuracy: 0.9108\n",
            "Epoch 24/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3951 - accuracy: 0.8680 - val_loss: 0.3160 - val_accuracy: 0.8968\n",
            "Epoch 25/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3965 - accuracy: 0.8656 - val_loss: 0.3537 - val_accuracy: 0.8928\n",
            "Epoch 26/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3946 - accuracy: 0.8674 - val_loss: 0.3255 - val_accuracy: 0.9120\n",
            "Epoch 27/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3925 - accuracy: 0.8671 - val_loss: 0.2894 - val_accuracy: 0.9092\n",
            "Epoch 28/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3934 - accuracy: 0.8674 - val_loss: 0.2948 - val_accuracy: 0.9020\n",
            "Epoch 29/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3934 - accuracy: 0.8687 - val_loss: 0.2963 - val_accuracy: 0.9102\n",
            "Epoch 30/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3910 - accuracy: 0.8682 - val_loss: 0.3025 - val_accuracy: 0.9098\n",
            "Epoch 31/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3917 - accuracy: 0.8669 - val_loss: 0.2906 - val_accuracy: 0.9070\n",
            "Epoch 32/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3911 - accuracy: 0.8688 - val_loss: 0.3070 - val_accuracy: 0.8998\n",
            "Epoch 33/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3938 - accuracy: 0.8664 - val_loss: 0.2821 - val_accuracy: 0.9118\n",
            "Epoch 34/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3897 - accuracy: 0.8696 - val_loss: 0.2842 - val_accuracy: 0.9034\n",
            "Epoch 35/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3903 - accuracy: 0.8679 - val_loss: 0.2987 - val_accuracy: 0.9058\n",
            "Epoch 36/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3887 - accuracy: 0.8687 - val_loss: 0.3028 - val_accuracy: 0.9026\n",
            "Epoch 37/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3884 - accuracy: 0.8688 - val_loss: 0.3007 - val_accuracy: 0.9042\n",
            "Epoch 38/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3898 - accuracy: 0.8682 - val_loss: 0.3023 - val_accuracy: 0.9036\n",
            "Epoch 39/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3881 - accuracy: 0.8682 - val_loss: 0.2968 - val_accuracy: 0.9050\n",
            "Epoch 40/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3894 - accuracy: 0.8693 - val_loss: 0.2824 - val_accuracy: 0.9028\n",
            "Epoch 41/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3892 - accuracy: 0.8666 - val_loss: 0.2968 - val_accuracy: 0.9032\n",
            "Epoch 42/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3871 - accuracy: 0.8681 - val_loss: 0.2879 - val_accuracy: 0.9114\n",
            "Epoch 43/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3881 - accuracy: 0.8675 - val_loss: 0.2769 - val_accuracy: 0.9074\n",
            "Epoch 44/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3872 - accuracy: 0.8695 - val_loss: 0.2708 - val_accuracy: 0.9048\n",
            "Epoch 45/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3868 - accuracy: 0.8690 - val_loss: 0.3005 - val_accuracy: 0.9112\n",
            "Epoch 46/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3857 - accuracy: 0.8684 - val_loss: 0.2730 - val_accuracy: 0.9140\n",
            "Epoch 47/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3864 - accuracy: 0.8701 - val_loss: 0.2778 - val_accuracy: 0.9136\n",
            "Epoch 48/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3875 - accuracy: 0.8669 - val_loss: 0.2885 - val_accuracy: 0.9092\n",
            "Epoch 49/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3860 - accuracy: 0.8687 - val_loss: 0.2821 - val_accuracy: 0.9028\n",
            "Epoch 50/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3863 - accuracy: 0.8685 - val_loss: 0.2991 - val_accuracy: 0.9110\n",
            "Epoch 51/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3854 - accuracy: 0.8688 - val_loss: 0.2735 - val_accuracy: 0.9072\n",
            "Epoch 52/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3848 - accuracy: 0.8716 - val_loss: 0.2949 - val_accuracy: 0.9060\n",
            "Epoch 53/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3851 - accuracy: 0.8686 - val_loss: 0.2812 - val_accuracy: 0.9038\n",
            "Epoch 54/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3861 - accuracy: 0.8685 - val_loss: 0.2913 - val_accuracy: 0.9140\n",
            "Epoch 55/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3837 - accuracy: 0.8702 - val_loss: 0.2877 - val_accuracy: 0.9050\n",
            "Epoch 56/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3846 - accuracy: 0.8697 - val_loss: 0.3266 - val_accuracy: 0.9056\n",
            "Epoch 57/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3853 - accuracy: 0.8684 - val_loss: 0.2790 - val_accuracy: 0.9082\n",
            "Epoch 58/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3861 - accuracy: 0.8686 - val_loss: 0.2904 - val_accuracy: 0.9080\n",
            "Epoch 59/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3822 - accuracy: 0.8691 - val_loss: 0.2936 - val_accuracy: 0.9092\n",
            "Epoch 60/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3845 - accuracy: 0.8688 - val_loss: 0.3105 - val_accuracy: 0.9122\n",
            "Epoch 61/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3839 - accuracy: 0.8697 - val_loss: 0.2987 - val_accuracy: 0.9046\n",
            "Epoch 62/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3843 - accuracy: 0.8685 - val_loss: 0.2757 - val_accuracy: 0.9100\n",
            "Epoch 63/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3823 - accuracy: 0.8700 - val_loss: 0.2707 - val_accuracy: 0.9072\n",
            "Epoch 64/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.3844 - accuracy: 0.8700 - val_loss: 0.2710 - val_accuracy: 0.9096\n",
            "Epoch 65/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.3840 - accuracy: 0.8690 - val_loss: 0.2825 - val_accuracy: 0.9060\n",
            "Epoch 66/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3845 - accuracy: 0.8688 - val_loss: 0.2826 - val_accuracy: 0.9130\n",
            "Epoch 67/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3825 - accuracy: 0.8690 - val_loss: 0.2743 - val_accuracy: 0.9120\n",
            "Epoch 68/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3823 - accuracy: 0.8698 - val_loss: 0.2806 - val_accuracy: 0.9092\n",
            "Epoch 69/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3823 - accuracy: 0.8701 - val_loss: 0.2780 - val_accuracy: 0.9112\n",
            "Epoch 70/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3830 - accuracy: 0.8695 - val_loss: 0.2792 - val_accuracy: 0.9110\n",
            "Epoch 71/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.3811 - accuracy: 0.8692 - val_loss: 0.2854 - val_accuracy: 0.9124\n",
            "Epoch 72/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.3821 - accuracy: 0.8703 - val_loss: 0.3191 - val_accuracy: 0.9122\n",
            "Epoch 73/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.3817 - accuracy: 0.8710 - val_loss: 0.2930 - val_accuracy: 0.9106\n",
            "Epoch 74/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.3825 - accuracy: 0.8690 - val_loss: 0.2960 - val_accuracy: 0.9088\n",
            "Epoch 75/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3836 - accuracy: 0.8686 - val_loss: 0.2995 - val_accuracy: 0.9066\n",
            "Epoch 76/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3820 - accuracy: 0.8693 - val_loss: 0.2996 - val_accuracy: 0.9116\n",
            "Epoch 77/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3822 - accuracy: 0.8687 - val_loss: 0.2863 - val_accuracy: 0.9070\n",
            "Epoch 78/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3829 - accuracy: 0.8697 - val_loss: 0.3223 - val_accuracy: 0.9002\n",
            "Epoch 79/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3819 - accuracy: 0.8704 - val_loss: 0.3459 - val_accuracy: 0.8728\n",
            "Epoch 80/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3825 - accuracy: 0.8696 - val_loss: 0.2926 - val_accuracy: 0.9122\n",
            "Epoch 81/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3810 - accuracy: 0.8703 - val_loss: 0.2781 - val_accuracy: 0.9048\n",
            "Epoch 82/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3811 - accuracy: 0.8700 - val_loss: 0.2932 - val_accuracy: 0.9074\n",
            "Epoch 83/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3796 - accuracy: 0.8706 - val_loss: 0.3315 - val_accuracy: 0.8884\n",
            "Epoch 84/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3822 - accuracy: 0.8692 - val_loss: 0.2772 - val_accuracy: 0.9120\n",
            "Epoch 85/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3809 - accuracy: 0.8697 - val_loss: 0.2737 - val_accuracy: 0.9052\n",
            "Epoch 86/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3804 - accuracy: 0.8686 - val_loss: 0.2783 - val_accuracy: 0.9046\n",
            "Epoch 87/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3795 - accuracy: 0.8707 - val_loss: 0.2676 - val_accuracy: 0.9074\n",
            "Epoch 88/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3801 - accuracy: 0.8695 - val_loss: 0.2779 - val_accuracy: 0.9142\n",
            "Epoch 89/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3798 - accuracy: 0.8693 - val_loss: 0.2721 - val_accuracy: 0.9110\n",
            "Epoch 90/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3795 - accuracy: 0.8702 - val_loss: 0.2789 - val_accuracy: 0.9096\n",
            "Epoch 91/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3802 - accuracy: 0.8709 - val_loss: 0.2845 - val_accuracy: 0.9034\n",
            "Epoch 92/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3782 - accuracy: 0.8706 - val_loss: 0.3076 - val_accuracy: 0.9150\n",
            "Epoch 93/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3808 - accuracy: 0.8679 - val_loss: 0.3001 - val_accuracy: 0.9092\n",
            "Epoch 94/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3779 - accuracy: 0.8719 - val_loss: 0.2782 - val_accuracy: 0.9068\n",
            "Epoch 95/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3799 - accuracy: 0.8691 - val_loss: 0.2812 - val_accuracy: 0.9084\n",
            "Epoch 96/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3785 - accuracy: 0.8706 - val_loss: 0.2797 - val_accuracy: 0.9020\n",
            "Epoch 97/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3795 - accuracy: 0.8692 - val_loss: 0.2789 - val_accuracy: 0.9096\n",
            "Epoch 98/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3795 - accuracy: 0.8705 - val_loss: 0.2768 - val_accuracy: 0.9108\n",
            "Epoch 99/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3781 - accuracy: 0.8723 - val_loss: 0.2647 - val_accuracy: 0.9100\n",
            "Epoch 100/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3796 - accuracy: 0.8713 - val_loss: 0.2750 - val_accuracy: 0.9056\n",
            "Epoch 1/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 56.5366 - accuracy: 0.7209 - val_loss: 6.6952 - val_accuracy: 0.7106\n",
            "Epoch 2/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 2.6496 - accuracy: 0.7865 - val_loss: 1.0763 - val_accuracy: 0.8246\n",
            "Epoch 3/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.8137 - accuracy: 0.8061 - val_loss: 0.5683 - val_accuracy: 0.8730\n",
            "Epoch 4/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.6018 - accuracy: 0.8142 - val_loss: 0.4541 - val_accuracy: 0.8906\n",
            "Epoch 5/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.5445 - accuracy: 0.8197 - val_loss: 0.4658 - val_accuracy: 0.8908\n",
            "Epoch 6/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.5222 - accuracy: 0.8196 - val_loss: 0.3717 - val_accuracy: 0.8988\n",
            "Epoch 7/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.5107 - accuracy: 0.8191 - val_loss: 0.3599 - val_accuracy: 0.8960\n",
            "Epoch 8/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.5026 - accuracy: 0.8190 - val_loss: 0.3605 - val_accuracy: 0.8876\n",
            "Epoch 9/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.4982 - accuracy: 0.8206 - val_loss: 0.3650 - val_accuracy: 0.9014\n",
            "Epoch 10/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.4936 - accuracy: 0.8212 - val_loss: 0.3513 - val_accuracy: 0.8850\n",
            "Epoch 11/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.4918 - accuracy: 0.8224 - val_loss: 0.3358 - val_accuracy: 0.8912\n",
            "Epoch 12/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.4913 - accuracy: 0.8213 - val_loss: 0.3389 - val_accuracy: 0.8974\n",
            "Epoch 13/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.4886 - accuracy: 0.8213 - val_loss: 0.3493 - val_accuracy: 0.8888\n",
            "Epoch 14/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.4864 - accuracy: 0.8224 - val_loss: 0.3949 - val_accuracy: 0.8848\n",
            "Epoch 15/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.4849 - accuracy: 0.8225 - val_loss: 0.3408 - val_accuracy: 0.9004\n",
            "Epoch 16/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.4832 - accuracy: 0.8237 - val_loss: 0.3399 - val_accuracy: 0.8982\n",
            "Epoch 17/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.4820 - accuracy: 0.8233 - val_loss: 0.3095 - val_accuracy: 0.9024\n",
            "Epoch 18/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.4814 - accuracy: 0.8236 - val_loss: 0.3286 - val_accuracy: 0.8936\n",
            "Epoch 19/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.4804 - accuracy: 0.8236 - val_loss: 0.3587 - val_accuracy: 0.9040\n",
            "Epoch 20/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.4787 - accuracy: 0.8256 - val_loss: 0.3420 - val_accuracy: 0.9072\n",
            "Epoch 21/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.4798 - accuracy: 0.8239 - val_loss: 0.3198 - val_accuracy: 0.8944\n",
            "Epoch 22/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.4780 - accuracy: 0.8245 - val_loss: 0.3600 - val_accuracy: 0.8968\n",
            "Epoch 23/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.4780 - accuracy: 0.8246 - val_loss: 0.3382 - val_accuracy: 0.8954\n",
            "Epoch 24/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.4777 - accuracy: 0.8254 - val_loss: 0.3463 - val_accuracy: 0.9078\n",
            "Epoch 25/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.4770 - accuracy: 0.8239 - val_loss: 0.3467 - val_accuracy: 0.8982\n",
            "Epoch 26/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.4764 - accuracy: 0.8249 - val_loss: 0.3282 - val_accuracy: 0.8978\n",
            "Epoch 27/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.4759 - accuracy: 0.8257 - val_loss: 0.3295 - val_accuracy: 0.9050\n",
            "Epoch 28/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.4756 - accuracy: 0.8261 - val_loss: 0.3372 - val_accuracy: 0.8842\n",
            "Epoch 29/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.4765 - accuracy: 0.8253 - val_loss: 0.3554 - val_accuracy: 0.9060\n",
            "Epoch 30/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.4743 - accuracy: 0.8247 - val_loss: 0.3367 - val_accuracy: 0.8940\n",
            "Epoch 31/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.4744 - accuracy: 0.8251 - val_loss: 0.3419 - val_accuracy: 0.9028\n",
            "Epoch 32/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.4736 - accuracy: 0.8274 - val_loss: 0.3471 - val_accuracy: 0.9078\n",
            "Epoch 33/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.4733 - accuracy: 0.8256 - val_loss: 0.3458 - val_accuracy: 0.9056\n",
            "Epoch 34/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.4743 - accuracy: 0.8252 - val_loss: 0.3189 - val_accuracy: 0.9042\n",
            "Epoch 35/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.4730 - accuracy: 0.8268 - val_loss: 0.3232 - val_accuracy: 0.8992\n",
            "Epoch 36/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.4741 - accuracy: 0.8250 - val_loss: 0.3452 - val_accuracy: 0.9078\n",
            "Epoch 37/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.4727 - accuracy: 0.8260 - val_loss: 0.3386 - val_accuracy: 0.9044\n",
            "Epoch 38/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.4711 - accuracy: 0.8278 - val_loss: 0.3090 - val_accuracy: 0.8992\n",
            "Epoch 39/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.4724 - accuracy: 0.8253 - val_loss: 0.3370 - val_accuracy: 0.8974\n",
            "Epoch 40/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.4717 - accuracy: 0.8256 - val_loss: 0.3720 - val_accuracy: 0.9006\n",
            "Epoch 41/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.4715 - accuracy: 0.8272 - val_loss: 0.3205 - val_accuracy: 0.9012\n",
            "Epoch 42/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.4707 - accuracy: 0.8263 - val_loss: 0.3533 - val_accuracy: 0.9056\n",
            "Epoch 43/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.4708 - accuracy: 0.8273 - val_loss: 0.3273 - val_accuracy: 0.8974\n",
            "Epoch 44/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.4710 - accuracy: 0.8263 - val_loss: 0.3137 - val_accuracy: 0.9092\n",
            "Epoch 45/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.4707 - accuracy: 0.8283 - val_loss: 0.3308 - val_accuracy: 0.9068\n",
            "Epoch 46/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.4707 - accuracy: 0.8261 - val_loss: 0.3218 - val_accuracy: 0.9088\n",
            "Epoch 47/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.4701 - accuracy: 0.8272 - val_loss: 0.3124 - val_accuracy: 0.9032\n",
            "Epoch 48/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.4708 - accuracy: 0.8286 - val_loss: 0.3291 - val_accuracy: 0.9084\n",
            "Epoch 49/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.4712 - accuracy: 0.8259 - val_loss: 0.3082 - val_accuracy: 0.9040\n",
            "Epoch 50/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.4700 - accuracy: 0.8268 - val_loss: 0.3177 - val_accuracy: 0.9110\n",
            "Epoch 51/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.4691 - accuracy: 0.8273 - val_loss: 0.3163 - val_accuracy: 0.8994\n",
            "Epoch 52/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.4702 - accuracy: 0.8260 - val_loss: 0.3251 - val_accuracy: 0.9016\n",
            "Epoch 53/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.4692 - accuracy: 0.8279 - val_loss: 0.3064 - val_accuracy: 0.9016\n",
            "Epoch 54/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.4697 - accuracy: 0.8272 - val_loss: 0.3183 - val_accuracy: 0.9074\n",
            "Epoch 55/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.4698 - accuracy: 0.8260 - val_loss: 0.3267 - val_accuracy: 0.9044\n",
            "Epoch 56/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.4697 - accuracy: 0.8268 - val_loss: 0.3461 - val_accuracy: 0.9090\n",
            "Epoch 57/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.4688 - accuracy: 0.8275 - val_loss: 0.3575 - val_accuracy: 0.9012\n",
            "Epoch 58/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.4682 - accuracy: 0.8258 - val_loss: 0.3173 - val_accuracy: 0.9024\n",
            "Epoch 59/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.4699 - accuracy: 0.8275 - val_loss: 0.3095 - val_accuracy: 0.9084\n",
            "Epoch 60/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.4681 - accuracy: 0.8279 - val_loss: 0.3198 - val_accuracy: 0.9044\n",
            "Epoch 61/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.4676 - accuracy: 0.8261 - val_loss: 0.3057 - val_accuracy: 0.9058\n",
            "Epoch 62/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.4664 - accuracy: 0.8285 - val_loss: 0.3243 - val_accuracy: 0.8912\n",
            "Epoch 63/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.4671 - accuracy: 0.8277 - val_loss: 0.3547 - val_accuracy: 0.9080\n",
            "Epoch 64/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.4682 - accuracy: 0.8276 - val_loss: 0.3211 - val_accuracy: 0.9114\n",
            "Epoch 65/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.4674 - accuracy: 0.8273 - val_loss: 0.3217 - val_accuracy: 0.9032\n",
            "Epoch 66/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.4677 - accuracy: 0.8258 - val_loss: 0.3579 - val_accuracy: 0.9084\n",
            "Epoch 67/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.4663 - accuracy: 0.8296 - val_loss: 0.3194 - val_accuracy: 0.8936\n",
            "Epoch 68/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.4663 - accuracy: 0.8274 - val_loss: 0.3308 - val_accuracy: 0.9002\n",
            "Epoch 69/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.4665 - accuracy: 0.8284 - val_loss: 0.3364 - val_accuracy: 0.9004\n",
            "Epoch 70/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.4675 - accuracy: 0.8282 - val_loss: 0.3274 - val_accuracy: 0.9122\n",
            "Epoch 71/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.4660 - accuracy: 0.8278 - val_loss: 0.3184 - val_accuracy: 0.9028\n",
            "Epoch 72/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.4674 - accuracy: 0.8274 - val_loss: 0.3806 - val_accuracy: 0.8884\n",
            "Epoch 73/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.4651 - accuracy: 0.8269 - val_loss: 0.3423 - val_accuracy: 0.9102\n",
            "Epoch 74/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.4670 - accuracy: 0.8280 - val_loss: 0.3072 - val_accuracy: 0.9056\n",
            "Epoch 75/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.4659 - accuracy: 0.8267 - val_loss: 0.3375 - val_accuracy: 0.9054\n",
            "Epoch 76/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.4656 - accuracy: 0.8279 - val_loss: 0.3777 - val_accuracy: 0.8904\n",
            "Epoch 77/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.4641 - accuracy: 0.8287 - val_loss: 0.3426 - val_accuracy: 0.9004\n",
            "Epoch 78/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.4641 - accuracy: 0.8279 - val_loss: 0.3746 - val_accuracy: 0.8764\n",
            "Epoch 79/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.4651 - accuracy: 0.8281 - val_loss: 0.3377 - val_accuracy: 0.8996\n",
            "Epoch 80/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.4642 - accuracy: 0.8299 - val_loss: 0.3328 - val_accuracy: 0.9084\n",
            "Epoch 81/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.4651 - accuracy: 0.8275 - val_loss: 0.3210 - val_accuracy: 0.9094\n",
            "Epoch 82/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.4653 - accuracy: 0.8274 - val_loss: 0.3226 - val_accuracy: 0.8974\n",
            "Epoch 83/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.4648 - accuracy: 0.8288 - val_loss: 0.3249 - val_accuracy: 0.9100\n",
            "Epoch 84/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.4636 - accuracy: 0.8286 - val_loss: 0.3188 - val_accuracy: 0.9072\n",
            "Epoch 85/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.4644 - accuracy: 0.8273 - val_loss: 0.3186 - val_accuracy: 0.9066\n",
            "Epoch 86/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.4636 - accuracy: 0.8294 - val_loss: 0.3253 - val_accuracy: 0.8964\n",
            "Epoch 87/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.4635 - accuracy: 0.8285 - val_loss: 0.3337 - val_accuracy: 0.9070\n",
            "Epoch 88/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.4636 - accuracy: 0.8278 - val_loss: 0.3267 - val_accuracy: 0.9074\n",
            "Epoch 89/100\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.4634 - accuracy: 0.8287 - val_loss: 0.3365 - val_accuracy: 0.9128\n",
            "Epoch 90/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.4629 - accuracy: 0.8289 - val_loss: 0.3255 - val_accuracy: 0.9106\n",
            "Epoch 91/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.4623 - accuracy: 0.8280 - val_loss: 0.3037 - val_accuracy: 0.9112\n",
            "Epoch 92/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.4643 - accuracy: 0.8271 - val_loss: 0.3160 - val_accuracy: 0.9066\n",
            "Epoch 93/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.4621 - accuracy: 0.8297 - val_loss: 0.3208 - val_accuracy: 0.9062\n",
            "Epoch 94/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.4623 - accuracy: 0.8287 - val_loss: 0.3180 - val_accuracy: 0.9090\n",
            "Epoch 95/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.4623 - accuracy: 0.8297 - val_loss: 0.3156 - val_accuracy: 0.9072\n",
            "Epoch 96/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.4632 - accuracy: 0.8287 - val_loss: 0.3132 - val_accuracy: 0.9026\n",
            "Epoch 97/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.4610 - accuracy: 0.8301 - val_loss: 0.3230 - val_accuracy: 0.9078\n",
            "Epoch 98/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.4622 - accuracy: 0.8283 - val_loss: 0.3276 - val_accuracy: 0.9118\n",
            "Epoch 99/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.4635 - accuracy: 0.8284 - val_loss: 0.3213 - val_accuracy: 0.9024\n",
            "Epoch 100/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.4635 - accuracy: 0.8288 - val_loss: 0.3392 - val_accuracy: 0.9012\n",
            "Epoch 1/100\n",
            "469/469 [==============================] - 4s 8ms/step - loss: 59.9466 - accuracy: 0.5466 - val_loss: 7.9810 - val_accuracy: 0.4968\n",
            "Epoch 2/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 2.8862 - accuracy: 0.5924 - val_loss: 1.0623 - val_accuracy: 0.7724\n",
            "Epoch 3/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.8545 - accuracy: 0.6110 - val_loss: 0.6712 - val_accuracy: 0.7864\n",
            "Epoch 4/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.7014 - accuracy: 0.6204 - val_loss: 0.6252 - val_accuracy: 0.7360\n",
            "Epoch 5/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.6798 - accuracy: 0.6207 - val_loss: 0.5817 - val_accuracy: 0.8480\n",
            "Epoch 6/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.6747 - accuracy: 0.6226 - val_loss: 0.5183 - val_accuracy: 0.8550\n",
            "Epoch 7/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.6722 - accuracy: 0.6227 - val_loss: 0.5702 - val_accuracy: 0.8212\n",
            "Epoch 8/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.6712 - accuracy: 0.6229 - val_loss: 0.5630 - val_accuracy: 0.8596\n",
            "Epoch 9/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.6707 - accuracy: 0.6230 - val_loss: 0.5523 - val_accuracy: 0.8750\n",
            "Epoch 10/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.6697 - accuracy: 0.6249 - val_loss: 0.5986 - val_accuracy: 0.8608\n",
            "Epoch 11/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.6701 - accuracy: 0.6238 - val_loss: 0.5562 - val_accuracy: 0.8572\n",
            "Epoch 12/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.6697 - accuracy: 0.6230 - val_loss: 0.5687 - val_accuracy: 0.8700\n",
            "Epoch 13/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.6697 - accuracy: 0.6238 - val_loss: 0.5963 - val_accuracy: 0.7754\n",
            "Epoch 14/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.6694 - accuracy: 0.6241 - val_loss: 0.5387 - val_accuracy: 0.8646\n",
            "Epoch 15/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.6686 - accuracy: 0.6249 - val_loss: 0.5732 - val_accuracy: 0.8680\n",
            "Epoch 16/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.6689 - accuracy: 0.6235 - val_loss: 0.5713 - val_accuracy: 0.8604\n",
            "Epoch 17/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.6683 - accuracy: 0.6250 - val_loss: 0.5540 - val_accuracy: 0.8688\n",
            "Epoch 18/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.6684 - accuracy: 0.6253 - val_loss: 0.5818 - val_accuracy: 0.8768\n",
            "Epoch 19/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.6676 - accuracy: 0.6257 - val_loss: 0.5365 - val_accuracy: 0.8662\n",
            "Epoch 20/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.6672 - accuracy: 0.6255 - val_loss: 0.5509 - val_accuracy: 0.8712\n",
            "Epoch 21/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.6678 - accuracy: 0.6236 - val_loss: 0.5485 - val_accuracy: 0.8738\n",
            "Epoch 22/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.6674 - accuracy: 0.6263 - val_loss: 0.5224 - val_accuracy: 0.8666\n",
            "Epoch 23/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.6674 - accuracy: 0.6258 - val_loss: 0.4826 - val_accuracy: 0.8676\n",
            "Epoch 24/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.6669 - accuracy: 0.6258 - val_loss: 0.5259 - val_accuracy: 0.8590\n",
            "Epoch 25/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.6666 - accuracy: 0.6251 - val_loss: 0.5241 - val_accuracy: 0.8716\n",
            "Epoch 26/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.6669 - accuracy: 0.6271 - val_loss: 0.5141 - val_accuracy: 0.8572\n",
            "Epoch 27/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.6670 - accuracy: 0.6253 - val_loss: 0.5566 - val_accuracy: 0.8708\n",
            "Epoch 28/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.6671 - accuracy: 0.6280 - val_loss: 0.5610 - val_accuracy: 0.8658\n",
            "Epoch 29/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.6669 - accuracy: 0.6254 - val_loss: 0.5420 - val_accuracy: 0.8482\n",
            "Epoch 30/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.6667 - accuracy: 0.6260 - val_loss: 0.5407 - val_accuracy: 0.8554\n",
            "Epoch 31/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.6666 - accuracy: 0.6253 - val_loss: 0.5405 - val_accuracy: 0.8742\n",
            "Epoch 32/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.6663 - accuracy: 0.6270 - val_loss: 0.5264 - val_accuracy: 0.8684\n",
            "Epoch 33/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.6664 - accuracy: 0.6266 - val_loss: 0.5258 - val_accuracy: 0.8666\n",
            "Epoch 34/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.6663 - accuracy: 0.6245 - val_loss: 0.5265 - val_accuracy: 0.8588\n",
            "Epoch 35/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.6668 - accuracy: 0.6246 - val_loss: 0.5540 - val_accuracy: 0.8772\n",
            "Epoch 36/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.6660 - accuracy: 0.6243 - val_loss: 0.6074 - val_accuracy: 0.8214\n",
            "Epoch 37/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.6662 - accuracy: 0.6282 - val_loss: 0.5533 - val_accuracy: 0.8734\n",
            "Epoch 38/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.6662 - accuracy: 0.6259 - val_loss: 0.5284 - val_accuracy: 0.8710\n",
            "Epoch 39/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.6661 - accuracy: 0.6273 - val_loss: 0.5003 - val_accuracy: 0.8732\n",
            "Epoch 40/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.6661 - accuracy: 0.6269 - val_loss: 0.5229 - val_accuracy: 0.8706\n",
            "Epoch 41/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.6659 - accuracy: 0.6259 - val_loss: 0.5781 - val_accuracy: 0.8614\n",
            "Epoch 42/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.6658 - accuracy: 0.6263 - val_loss: 0.5802 - val_accuracy: 0.8452\n",
            "Epoch 43/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.6658 - accuracy: 0.6274 - val_loss: 0.5354 - val_accuracy: 0.8778\n",
            "Epoch 44/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.6660 - accuracy: 0.6264 - val_loss: 0.5264 - val_accuracy: 0.8708\n",
            "Epoch 45/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.6655 - accuracy: 0.6278 - val_loss: 0.5325 - val_accuracy: 0.8746\n",
            "Epoch 46/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.6654 - accuracy: 0.6261 - val_loss: 0.5673 - val_accuracy: 0.8782\n",
            "Epoch 47/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.6654 - accuracy: 0.6252 - val_loss: 0.5199 - val_accuracy: 0.8622\n",
            "Epoch 48/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.6656 - accuracy: 0.6260 - val_loss: 0.5404 - val_accuracy: 0.8742\n",
            "Epoch 49/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.6657 - accuracy: 0.6267 - val_loss: 0.5458 - val_accuracy: 0.8750\n",
            "Epoch 50/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.6652 - accuracy: 0.6269 - val_loss: 0.5337 - val_accuracy: 0.8764\n",
            "Epoch 51/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.6653 - accuracy: 0.6255 - val_loss: 0.5352 - val_accuracy: 0.8664\n",
            "Epoch 52/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.6653 - accuracy: 0.6273 - val_loss: 0.5117 - val_accuracy: 0.8694\n",
            "Epoch 53/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.6653 - accuracy: 0.6263 - val_loss: 0.5277 - val_accuracy: 0.8562\n",
            "Epoch 54/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.6654 - accuracy: 0.6252 - val_loss: 0.5434 - val_accuracy: 0.8708\n",
            "Epoch 55/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.6656 - accuracy: 0.6266 - val_loss: 0.5800 - val_accuracy: 0.8422\n",
            "Epoch 56/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.6653 - accuracy: 0.6277 - val_loss: 0.5410 - val_accuracy: 0.8786\n",
            "Epoch 57/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.6648 - accuracy: 0.6262 - val_loss: 0.5516 - val_accuracy: 0.8740\n",
            "Epoch 58/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.6648 - accuracy: 0.6247 - val_loss: 0.5543 - val_accuracy: 0.8680\n",
            "Epoch 59/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.6645 - accuracy: 0.6261 - val_loss: 0.5476 - val_accuracy: 0.8732\n",
            "Epoch 60/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.6648 - accuracy: 0.6261 - val_loss: 0.5383 - val_accuracy: 0.8732\n",
            "Epoch 61/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.6649 - accuracy: 0.6258 - val_loss: 0.5426 - val_accuracy: 0.8572\n",
            "Epoch 62/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.6647 - accuracy: 0.6282 - val_loss: 0.4963 - val_accuracy: 0.8636\n",
            "Epoch 63/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.6647 - accuracy: 0.6257 - val_loss: 0.5324 - val_accuracy: 0.8730\n",
            "Epoch 64/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.6645 - accuracy: 0.6258 - val_loss: 0.5224 - val_accuracy: 0.8680\n",
            "Epoch 65/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.6648 - accuracy: 0.6258 - val_loss: 0.5323 - val_accuracy: 0.8610\n",
            "Epoch 66/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.6648 - accuracy: 0.6261 - val_loss: 0.5356 - val_accuracy: 0.8534\n",
            "Epoch 67/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.6640 - accuracy: 0.6278 - val_loss: 0.5690 - val_accuracy: 0.8714\n",
            "Epoch 68/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.6648 - accuracy: 0.6278 - val_loss: 0.5433 - val_accuracy: 0.8704\n",
            "Epoch 69/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.6646 - accuracy: 0.6270 - val_loss: 0.4961 - val_accuracy: 0.8710\n",
            "Epoch 70/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.6643 - accuracy: 0.6264 - val_loss: 0.5188 - val_accuracy: 0.8716\n",
            "Epoch 71/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.6644 - accuracy: 0.6268 - val_loss: 0.5354 - val_accuracy: 0.8702\n",
            "Epoch 72/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.6644 - accuracy: 0.6277 - val_loss: 0.5206 - val_accuracy: 0.8770\n",
            "Epoch 73/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.6645 - accuracy: 0.6271 - val_loss: 0.5380 - val_accuracy: 0.8712\n",
            "Epoch 74/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.6640 - accuracy: 0.6260 - val_loss: 0.5282 - val_accuracy: 0.8696\n",
            "Epoch 75/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.6641 - accuracy: 0.6270 - val_loss: 0.5399 - val_accuracy: 0.8754\n",
            "Epoch 76/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.6643 - accuracy: 0.6277 - val_loss: 0.5606 - val_accuracy: 0.8710\n",
            "Epoch 77/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.6642 - accuracy: 0.6262 - val_loss: 0.5255 - val_accuracy: 0.8750\n",
            "Epoch 78/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.6645 - accuracy: 0.6257 - val_loss: 0.4961 - val_accuracy: 0.8744\n",
            "Epoch 79/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.6648 - accuracy: 0.6262 - val_loss: 0.5417 - val_accuracy: 0.8834\n",
            "Epoch 80/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.6636 - accuracy: 0.6272 - val_loss: 0.5332 - val_accuracy: 0.8762\n",
            "Epoch 81/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.6646 - accuracy: 0.6272 - val_loss: 0.5471 - val_accuracy: 0.8762\n",
            "Epoch 82/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.6644 - accuracy: 0.6262 - val_loss: 0.5022 - val_accuracy: 0.8698\n",
            "Epoch 83/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.6644 - accuracy: 0.6281 - val_loss: 0.5213 - val_accuracy: 0.8768\n",
            "Epoch 84/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.6643 - accuracy: 0.6270 - val_loss: 0.5347 - val_accuracy: 0.8602\n",
            "Epoch 85/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.6641 - accuracy: 0.6269 - val_loss: 0.5770 - val_accuracy: 0.8606\n",
            "Epoch 86/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.6639 - accuracy: 0.6275 - val_loss: 0.5292 - val_accuracy: 0.8674\n",
            "Epoch 87/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.6640 - accuracy: 0.6259 - val_loss: 0.5096 - val_accuracy: 0.8780\n",
            "Epoch 88/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.6641 - accuracy: 0.6282 - val_loss: 0.5163 - val_accuracy: 0.8776\n",
            "Epoch 89/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.6641 - accuracy: 0.6254 - val_loss: 0.4907 - val_accuracy: 0.8760\n",
            "Epoch 90/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.6637 - accuracy: 0.6273 - val_loss: 0.5592 - val_accuracy: 0.8718\n",
            "Epoch 91/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.6642 - accuracy: 0.6253 - val_loss: 0.5541 - val_accuracy: 0.8676\n",
            "Epoch 92/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.6645 - accuracy: 0.6252 - val_loss: 0.5130 - val_accuracy: 0.8778\n",
            "Epoch 93/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.6644 - accuracy: 0.6278 - val_loss: 0.5385 - val_accuracy: 0.8790\n",
            "Epoch 94/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.6642 - accuracy: 0.6260 - val_loss: 0.5744 - val_accuracy: 0.8744\n",
            "Epoch 95/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.6641 - accuracy: 0.6259 - val_loss: 0.5322 - val_accuracy: 0.8810\n",
            "Epoch 96/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.6640 - accuracy: 0.6276 - val_loss: 0.5293 - val_accuracy: 0.8726\n",
            "Epoch 97/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.6640 - accuracy: 0.6268 - val_loss: 0.5096 - val_accuracy: 0.8798\n",
            "Epoch 98/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.6637 - accuracy: 0.6265 - val_loss: 0.5282 - val_accuracy: 0.8726\n",
            "Epoch 99/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.6643 - accuracy: 0.6268 - val_loss: 0.5289 - val_accuracy: 0.8732\n",
            "Epoch 100/100\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.6639 - accuracy: 0.6268 - val_loss: 0.5220 - val_accuracy: 0.8730\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "sR-UrZNhn2-b",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "4db3706b-a593-47dc-92cc-6830913d139d"
      },
      "source": [
        "print(Accuracies_noise)\n",
        "print(Precisions_noise)\n",
        "print(Recalls_noise)\n",
        "print(AUCs_noise)"
      ],
      "execution_count": 26,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "[0.9184 0.9142 0.9088 0.883 ]\n",
            "[0.89947683 0.89988666 0.92171106 0.87989151]\n",
            "[0.94540456 0.93558523 0.89709348 0.89198743]\n",
            "[0.97451617 0.9700637  0.96984867 0.95018978]\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "yS-Ht0YwvfFw"
      },
      "source": [
        "### IIb) Biased noise\n",
        "\n",
        "**Part I**\n",
        "\n",
        "In this part we assume a systematic mislabeling of LSBGs as artefacts 1 $\\to$ 0. This is something that can happen if we are too conservative, favoring purity over completeness in our visual inspection of the training set.\n",
        "\n",
        "\n",
        "We randomly select 300, 1500 and 3000 LSBGs and we change their labels"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "MVqqg4b5jt-L"
      },
      "source": [
        "# Create empty arrays for the evaluation metrics of each loop\n",
        "Accuracies_biasedLSBG = np.zeros(3)\n",
        "Accuracies_balanced_biasedLSBG = np.zeros(3)\n",
        "Precisions_biasedLSBG = np.zeros(3)\n",
        "Recalls_biasedLSBG = np.zeros(3)\n",
        "tprs_biasedLSBG = np.zeros((3,200))\n",
        "fprs_biasedLSBG = np.linspace(0,1,200)\n",
        "AUCs_biasedLSBG = np.zeros(3)\n",
        "# =================================\n",
        "# =================================\n",
        "num_examples = [300,1500,3000] #Number of examples to change the labels"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "jMD9fARQjuDj",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 1000
        },
        "outputId": "5568b323-02e9-4268-95c1-145f45ff238d"
      },
      "source": [
        "for i in range(3):\n",
        "  # Clear the backend session\n",
        "  tf.keras.backend.clear_session()\n",
        "  # ============================================================================\n",
        "  # ============================================================================\n",
        "  # Number of examples to change the labels\n",
        "  n_examples = num_examples[i]\n",
        "\n",
        "  # Make now a copy of y_train\n",
        "  y_train_loc = np.copy(y_tr)\n",
        "\n",
        "  # Now get an array with the indices in the y_train array that \n",
        "  # point to the elements with y_train = 1 (LSBGs) \n",
        "  indices_LSBG = np.asarray(np.where(y_train_loc==1.0)[0])\n",
        "\n",
        "  # Now get n_examples random indices from 0 to len(indices_LSBG)\n",
        "  rands_loc = random.sample(range(0, len(indices_LSBG)),n_examples)\n",
        "\n",
        "  # Now get indices of the objects to change\n",
        "  indices_change = indices_LSBG[rands_loc]\n",
        "\n",
        "  # Now, change the indices \n",
        "  y_train_loc[indices_change] = 0.\n",
        "  # ============================================================================\n",
        "  # ============================================================================\n",
        "  # ============================================================================\n",
        "  # Now build the model\n",
        "  model = keras.models.Sequential()\n",
        "  model.add(InputLayer((64,64,3)))\n",
        "  model.add(Conv2D(filters=16, kernel_size=(3,3), padding='same', activation='relu',kernel_regularizer=regularizers.l2(0.13)))\n",
        "  model.add(BatchNormalization())\n",
        "  model.add(MaxPool2D(pool_size=(2, 2), strides=None, padding='valid'))\n",
        "  model.add(Dropout(rate=0.4))\n",
        "  model.add(Conv2D(filters=2*16, kernel_size=(3,3), padding='same', activation='relu',kernel_regularizer=regularizers.l2(0.13)))\n",
        "  model.add(BatchNormalization())\n",
        "  model.add(MaxPool2D(pool_size=(2, 2), strides=None, padding='valid'))\n",
        "  model.add(Dropout(rate=0.4))\n",
        "  model.add(Conv2D(filters=2*32, kernel_size=(3,3), padding='same', activation='relu',kernel_regularizer=regularizers.l2(0.13)))\n",
        "  model.add(BatchNormalization())\n",
        "  model.add(MaxPool2D(pool_size=(2, 2), strides=None, padding='valid'))\n",
        "  model.add(Dropout(rate=0.4))\n",
        "  model.add(Flatten())\n",
        "  model.add(Dense(units=1024, activation='relu',kernel_regularizer=regularizers.l2(0.12)))\n",
        "  model.add(Dense(units=1, activation='sigmoid'))\n",
        "  model.compile(optimizer=optimizers.Adadelta(0.1),\n",
        "                loss= 'binary_crossentropy',\n",
        "                metrics=['accuracy'])\n",
        "  # ============================================================================\n",
        "  # Now fit the model\n",
        "  model.fit(x=X_tr, y=y_train_loc,\n",
        "          epochs=20, batch_size=64,\n",
        "          shuffle=True,\n",
        "          validation_data=(X_val,y_val))\n",
        "  # ============================================================================\n",
        "  # ============================================================================\n",
        "  # ============================================================================\n",
        "  # Now predict on the test set\n",
        "  y_pred_loc = model.predict_classes(X_test)\n",
        "  # Predict probability scores\n",
        "  scores_loc = model.predict(X_test)\n",
        "  # ============================================================================\n",
        "  # ============================================================================\n",
        "  # ============================================================================\n",
        "  # Evaluation metrics \n",
        "\n",
        "  Accuracies_biasedLSBG[i] = accuracy_score(y_test, y_pred_loc)\n",
        "  Accuracies_balanced_biasedLSBG[i] = balanced_accuracy_score(y_test, y_pred_loc)\n",
        "  Precisions_biasedLSBG[i] = precision_score(y_test, y_pred_loc)\n",
        "  Recalls_biasedLSBG[i] = recall_score(y_test, y_pred_loc)\n",
        "  AUCs_biasedLSBG[i] = roc_auc_score(y_test, scores_loc[:,0])\n",
        "  # ============================================================================\n",
        "  # Get now false positives and true positives\n",
        "  fpr, tpr, thresholds = roc_curve(y_test, scores_loc[:,0], pos_label=1.0)\n",
        "  fpr_unique, ind_unique = np.unique(fpr,return_index=True)\n",
        "  tpr_unique = tpr[ind_unique]\n",
        "\n",
        "  # Interpolate the false positives - true positives\n",
        "  interp = interpolate.interp1d(fpr_unique, tpr_unique, kind='linear')\n",
        "  tprs_biasedLSBG[i,:] = interp(fprs_biasedLSBG)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Epoch 1/20\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 58.2149 - accuracy: 0.8024 - val_loss: 7.0462 - val_accuracy: 0.6448\n",
            "Epoch 2/20\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 2.6662 - accuracy: 0.8644 - val_loss: 1.1531 - val_accuracy: 0.8298\n",
            "Epoch 3/20\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.7424 - accuracy: 0.8814 - val_loss: 0.6134 - val_accuracy: 0.8562\n",
            "Epoch 4/20\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.4879 - accuracy: 0.8910 - val_loss: 0.4845 - val_accuracy: 0.8698\n",
            "Epoch 5/20\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.4125 - accuracy: 0.8952 - val_loss: 0.3812 - val_accuracy: 0.8952\n",
            "Epoch 6/20\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3770 - accuracy: 0.8971 - val_loss: 0.3381 - val_accuracy: 0.9028\n",
            "Epoch 7/20\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3551 - accuracy: 0.8994 - val_loss: 0.3281 - val_accuracy: 0.9018\n",
            "Epoch 8/20\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3445 - accuracy: 0.8988 - val_loss: 0.3278 - val_accuracy: 0.9052\n",
            "Epoch 9/20\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3343 - accuracy: 0.8985 - val_loss: 0.3131 - val_accuracy: 0.9068\n",
            "Epoch 10/20\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3281 - accuracy: 0.8987 - val_loss: 0.3135 - val_accuracy: 0.9086\n",
            "Epoch 11/20\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3230 - accuracy: 0.9015 - val_loss: 0.2989 - val_accuracy: 0.9068\n",
            "Epoch 12/20\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3172 - accuracy: 0.9012 - val_loss: 0.2886 - val_accuracy: 0.9106\n",
            "Epoch 13/20\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3153 - accuracy: 0.9008 - val_loss: 0.2978 - val_accuracy: 0.9084\n",
            "Epoch 14/20\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.3130 - accuracy: 0.9006 - val_loss: 0.2861 - val_accuracy: 0.9146\n",
            "Epoch 15/20\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.3095 - accuracy: 0.9020 - val_loss: 0.2798 - val_accuracy: 0.9104\n",
            "Epoch 16/20\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.3093 - accuracy: 0.9020 - val_loss: 0.2875 - val_accuracy: 0.9064\n",
            "Epoch 17/20\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.3069 - accuracy: 0.9014 - val_loss: 0.2804 - val_accuracy: 0.9142\n",
            "Epoch 18/20\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3066 - accuracy: 0.9017 - val_loss: 0.2839 - val_accuracy: 0.9102\n",
            "Epoch 19/20\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3046 - accuracy: 0.9007 - val_loss: 0.2798 - val_accuracy: 0.9112\n",
            "Epoch 20/20\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3027 - accuracy: 0.9026 - val_loss: 0.2872 - val_accuracy: 0.9010\n",
            "Epoch 1/20\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 56.4868 - accuracy: 0.7657 - val_loss: 6.1259 - val_accuracy: 0.8258\n",
            "Epoch 2/20\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 2.4055 - accuracy: 0.8320 - val_loss: 1.0353 - val_accuracy: 0.8644\n",
            "Epoch 3/20\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.7515 - accuracy: 0.8500 - val_loss: 0.5867 - val_accuracy: 0.8922\n",
            "Epoch 4/20\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.5367 - accuracy: 0.8547 - val_loss: 0.4456 - val_accuracy: 0.8900\n",
            "Epoch 5/20\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.4655 - accuracy: 0.8590 - val_loss: 0.3785 - val_accuracy: 0.8988\n",
            "Epoch 6/20\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.4319 - accuracy: 0.8590 - val_loss: 0.3514 - val_accuracy: 0.8992\n",
            "Epoch 7/20\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.4133 - accuracy: 0.8629 - val_loss: 0.3286 - val_accuracy: 0.9034\n",
            "Epoch 8/20\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.4011 - accuracy: 0.8628 - val_loss: 0.3313 - val_accuracy: 0.9102\n",
            "Epoch 9/20\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3928 - accuracy: 0.8627 - val_loss: 0.3317 - val_accuracy: 0.9040\n",
            "Epoch 10/20\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3881 - accuracy: 0.8643 - val_loss: 0.3020 - val_accuracy: 0.9062\n",
            "Epoch 11/20\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3832 - accuracy: 0.8650 - val_loss: 0.3137 - val_accuracy: 0.9088\n",
            "Epoch 12/20\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3798 - accuracy: 0.8641 - val_loss: 0.3048 - val_accuracy: 0.9080\n",
            "Epoch 13/20\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3757 - accuracy: 0.8662 - val_loss: 0.3003 - val_accuracy: 0.9098\n",
            "Epoch 14/20\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3757 - accuracy: 0.8652 - val_loss: 0.3157 - val_accuracy: 0.9038\n",
            "Epoch 15/20\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3740 - accuracy: 0.8661 - val_loss: 0.2908 - val_accuracy: 0.9074\n",
            "Epoch 16/20\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3720 - accuracy: 0.8669 - val_loss: 0.2940 - val_accuracy: 0.9012\n",
            "Epoch 17/20\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3711 - accuracy: 0.8660 - val_loss: 0.3239 - val_accuracy: 0.9054\n",
            "Epoch 18/20\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3692 - accuracy: 0.8659 - val_loss: 0.2873 - val_accuracy: 0.9078\n",
            "Epoch 19/20\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3698 - accuracy: 0.8656 - val_loss: 0.2928 - val_accuracy: 0.9142\n",
            "Epoch 20/20\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3686 - accuracy: 0.8660 - val_loss: 0.2997 - val_accuracy: 0.9056\n",
            "Epoch 1/20\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 57.8646 - accuracy: 0.7273 - val_loss: 6.8612 - val_accuracy: 0.5766\n",
            "Epoch 2/20\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 2.6165 - accuracy: 0.7900 - val_loss: 1.1205 - val_accuracy: 0.8528\n",
            "Epoch 3/20\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.8113 - accuracy: 0.8075 - val_loss: 0.6093 - val_accuracy: 0.8972\n",
            "Epoch 4/20\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.5805 - accuracy: 0.8130 - val_loss: 0.4434 - val_accuracy: 0.8948\n",
            "Epoch 5/20\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.5066 - accuracy: 0.8167 - val_loss: 0.4155 - val_accuracy: 0.9036\n",
            "Epoch 6/20\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.4711 - accuracy: 0.8196 - val_loss: 0.4049 - val_accuracy: 0.8848\n",
            "Epoch 7/20\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.4529 - accuracy: 0.8206 - val_loss: 0.3558 - val_accuracy: 0.9014\n",
            "Epoch 8/20\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.4438 - accuracy: 0.8218 - val_loss: 0.3495 - val_accuracy: 0.9012\n",
            "Epoch 9/20\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.4355 - accuracy: 0.8215 - val_loss: 0.3517 - val_accuracy: 0.8982\n",
            "Epoch 10/20\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.4319 - accuracy: 0.8222 - val_loss: 0.3596 - val_accuracy: 0.9060\n",
            "Epoch 11/20\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.4276 - accuracy: 0.8231 - val_loss: 0.3329 - val_accuracy: 0.9072\n",
            "Epoch 12/20\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.4253 - accuracy: 0.8235 - val_loss: 0.3424 - val_accuracy: 0.9036\n",
            "Epoch 13/20\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.4226 - accuracy: 0.8236 - val_loss: 0.3294 - val_accuracy: 0.9092\n",
            "Epoch 14/20\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.4205 - accuracy: 0.8230 - val_loss: 0.3082 - val_accuracy: 0.9074\n",
            "Epoch 15/20\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.4198 - accuracy: 0.8247 - val_loss: 0.3469 - val_accuracy: 0.8994\n",
            "Epoch 16/20\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.4203 - accuracy: 0.8209 - val_loss: 0.3281 - val_accuracy: 0.9102\n",
            "Epoch 17/20\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.4169 - accuracy: 0.8245 - val_loss: 0.3217 - val_accuracy: 0.9074\n",
            "Epoch 18/20\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.4185 - accuracy: 0.8240 - val_loss: 0.3306 - val_accuracy: 0.9034\n",
            "Epoch 19/20\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.4177 - accuracy: 0.8221 - val_loss: 0.3313 - val_accuracy: 0.9050\n",
            "Epoch 20/20\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.4149 - accuracy: 0.8248 - val_loss: 0.3957 - val_accuracy: 0.8668\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "EfJKTW7ZjuNi",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 102
        },
        "outputId": "e77696db-a6a5-4a72-b73f-23ca64ef1677"
      },
      "source": [
        "print(Accuracies_biasedLSBG)\n",
        "print(Accuracies_balanced_biasedLSBG)\n",
        "print(Precisions_biasedLSBG)\n",
        "print(Recalls_biasedLSBG)\n",
        "print(AUCs_biasedLSBG)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "[0.9102 0.9152 0.9154]\n",
            "[0.90903944 0.91433196 0.91453571]\n",
            "[0.86750789 0.88248017 0.88279841]\n",
            "[0.97211312 0.96150825 0.96150825]\n",
            "[0.9705666  0.97082476 0.96821348]\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "IR4Ifx-ijuRc"
      },
      "source": [
        "**Part II**\n",
        "\n",
        "In this part we assume a systematic mislabeling of artefacts as LSBGs 0 $\\to$ 1. This is something that can happen if we are too conservative, favoring purity over completeness in our visual inspection of the training set.\n",
        "\n",
        "\n",
        "We randomly select 300, 1500 and 3000 artefacts and we change their labels"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "0PrWP-nKV_1q"
      },
      "source": [
        "# Create empty arrays for the evaluation metrics of each loop\n",
        "Accuracies_biasedArtefacts = np.zeros(3)\n",
        "Accuracies_balanced_biasedArtefacts = np.zeros(3)\n",
        "Precisions_biasedArtefacts = np.zeros(3)\n",
        "Recalls_biasedArtefacts = np.zeros(3)\n",
        "tprs_biasedArtefacts = np.zeros((3,200))\n",
        "fprs_biasedArtefacts = np.linspace(0,1,200)\n",
        "AUCs_biasedArtefacts = np.zeros(3)\n",
        "# =================================\n",
        "# =================================\n",
        "num_examples = [300,1500,3000] #Number of examples to change the labels"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "t-kvdiZYW_Hf",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 1000
        },
        "outputId": "dd34e399-7a8f-404f-f53a-fe5ce2dab413"
      },
      "source": [
        "for i in range(3):\n",
        "  # Clear the backend session\n",
        "  tf.keras.backend.clear_session()\n",
        "  # ============================================================================\n",
        "  # ============================================================================\n",
        "  # Number of examples to change the labels\n",
        "  n_examples = num_examples[i]\n",
        "\n",
        "  # Make now a copy of y_train\n",
        "  y_train_loc = np.copy(y_tr)\n",
        "\n",
        "  # Now get an array with the indices in the y_train array that \n",
        "  # point to the elements with y_train = 1 (LSBGs) \n",
        "  indices_LSBG = np.asarray(np.where(y_train_loc==0.0)[0])\n",
        "\n",
        "  # Now get n_examples random indices from 0 to len(indices_LSBG)\n",
        "  rands_loc = random.sample(range(0, len(indices_LSBG)),n_examples)\n",
        "\n",
        "  # Now get indices of the objects to change\n",
        "  indices_change = indices_LSBG[rands_loc]\n",
        "\n",
        "  # Now, change the indices \n",
        "  y_train_loc[indices_change] = 1.\n",
        "  # ============================================================================\n",
        "  # ============================================================================\n",
        "  # ============================================================================\n",
        "  # Now build the model\n",
        "  model = keras.models.Sequential()\n",
        "  model.add(InputLayer((64,64,3)))\n",
        "  model.add(Conv2D(filters=16, kernel_size=(3,3), padding='same', activation='relu',kernel_regularizer=regularizers.l2(0.13)))\n",
        "  model.add(BatchNormalization())\n",
        "  model.add(MaxPool2D(pool_size=(2, 2), strides=None, padding='valid'))\n",
        "  model.add(Dropout(rate=0.4))\n",
        "  model.add(Conv2D(filters=2*16, kernel_size=(3,3), padding='same', activation='relu',kernel_regularizer=regularizers.l2(0.13)))\n",
        "  model.add(BatchNormalization())\n",
        "  model.add(MaxPool2D(pool_size=(2, 2), strides=None, padding='valid'))\n",
        "  model.add(Dropout(rate=0.4))\n",
        "  model.add(Conv2D(filters=2*32, kernel_size=(3,3), padding='same', activation='relu',kernel_regularizer=regularizers.l2(0.13)))\n",
        "  model.add(BatchNormalization())\n",
        "  model.add(MaxPool2D(pool_size=(2, 2), strides=None, padding='valid'))\n",
        "  model.add(Dropout(rate=0.4))\n",
        "  model.add(Flatten())\n",
        "  model.add(Dense(units=1024, activation='relu',kernel_regularizer=regularizers.l2(0.12)))\n",
        "  model.add(Dense(units=1, activation='sigmoid'))\n",
        "  model.compile(optimizer=optimizers.Adadelta(0.1),\n",
        "                loss= 'binary_crossentropy',\n",
        "                metrics=['accuracy'])\n",
        "  # ============================================================================\n",
        "  # Now fit the model\n",
        "  model.fit(x=X_tr, y=y_train_loc,\n",
        "          epochs=20, batch_size=64,\n",
        "          shuffle=True,\n",
        "          validation_data=(X_val,y_val))\n",
        "  # ============================================================================\n",
        "  # ============================================================================\n",
        "  # ============================================================================\n",
        "  # Now predict on the test set\n",
        "  y_pred_loc = model.predict_classes(X_test)\n",
        "  # Predict probability scores\n",
        "  scores_loc = model.predict(X_test)\n",
        "  # ============================================================================\n",
        "  # ============================================================================\n",
        "  # ============================================================================\n",
        "  # Evaluation metrics \n",
        "\n",
        "  Accuracies_biasedArtefacts[i] = accuracy_score(y_test, y_pred_loc)\n",
        "  Accuracies_balanced_biasedArtefacts[i] = balanced_accuracy_score(y_test, y_pred_loc)\n",
        "  Precisions_biasedArtefacts[i] = precision_score(y_test, y_pred_loc)\n",
        "  Recalls_biasedArtefacts[i] = recall_score(y_test, y_pred_loc)\n",
        "  AUCs_biasedArtefacts[i] = roc_auc_score(y_test, scores_loc[:,0])\n",
        "  # ============================================================================\n",
        "  # Get now false positives and true positives\n",
        "  fpr, tpr, thresholds = roc_curve(y_test, scores_loc[:,0], pos_label=1.0)\n",
        "  fpr_unique, ind_unique = np.unique(fpr,return_index=True)\n",
        "  tpr_unique = tpr[ind_unique]\n",
        "\n",
        "  # Interpolate the false positives - true positives\n",
        "  interp = interpolate.interp1d(fpr_unique, tpr_unique, kind='linear')\n",
        "  tprs_biasedArtefacts[i,:] = interp(fprs_biasedLSBG)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Epoch 1/20\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 56.5264 - accuracy: 0.7977 - val_loss: 6.6501 - val_accuracy: 0.6608\n",
            "Epoch 2/20\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 2.5578 - accuracy: 0.8656 - val_loss: 1.0837 - val_accuracy: 0.8638\n",
            "Epoch 3/20\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.7308 - accuracy: 0.8827 - val_loss: 0.5806 - val_accuracy: 0.8654\n",
            "Epoch 4/20\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.4846 - accuracy: 0.8909 - val_loss: 0.4368 - val_accuracy: 0.8916\n",
            "Epoch 5/20\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.4098 - accuracy: 0.8947 - val_loss: 0.3713 - val_accuracy: 0.9024\n",
            "Epoch 6/20\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3753 - accuracy: 0.8973 - val_loss: 0.3411 - val_accuracy: 0.9012\n",
            "Epoch 7/20\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3562 - accuracy: 0.8994 - val_loss: 0.3386 - val_accuracy: 0.8978\n",
            "Epoch 8/20\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3444 - accuracy: 0.9001 - val_loss: 0.3092 - val_accuracy: 0.9094\n",
            "Epoch 9/20\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3364 - accuracy: 0.8997 - val_loss: 0.3131 - val_accuracy: 0.9074\n",
            "Epoch 10/20\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3311 - accuracy: 0.9008 - val_loss: 0.2969 - val_accuracy: 0.9078\n",
            "Epoch 11/20\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3266 - accuracy: 0.9009 - val_loss: 0.2926 - val_accuracy: 0.9058\n",
            "Epoch 12/20\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3247 - accuracy: 0.9014 - val_loss: 0.2936 - val_accuracy: 0.9060\n",
            "Epoch 13/20\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3204 - accuracy: 0.9013 - val_loss: 0.3053 - val_accuracy: 0.9056\n",
            "Epoch 14/20\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.3201 - accuracy: 0.8992 - val_loss: 0.3291 - val_accuracy: 0.8934\n",
            "Epoch 15/20\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3171 - accuracy: 0.9016 - val_loss: 0.3064 - val_accuracy: 0.9056\n",
            "Epoch 16/20\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3169 - accuracy: 0.9012 - val_loss: 0.2803 - val_accuracy: 0.9102\n",
            "Epoch 17/20\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3154 - accuracy: 0.9014 - val_loss: 0.2985 - val_accuracy: 0.9038\n",
            "Epoch 18/20\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3139 - accuracy: 0.9011 - val_loss: 0.3004 - val_accuracy: 0.8996\n",
            "Epoch 19/20\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3111 - accuracy: 0.9024 - val_loss: 0.3391 - val_accuracy: 0.8716\n",
            "Epoch 20/20\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3119 - accuracy: 0.9022 - val_loss: 0.2790 - val_accuracy: 0.9100\n",
            "Epoch 1/20\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 63.2736 - accuracy: 0.7520 - val_loss: 8.8413 - val_accuracy: 0.7628\n",
            "Epoch 2/20\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 3.3938 - accuracy: 0.8283 - val_loss: 1.3796 - val_accuracy: 0.7864\n",
            "Epoch 3/20\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.9286 - accuracy: 0.8494 - val_loss: 0.7676 - val_accuracy: 0.8148\n",
            "Epoch 4/20\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.6170 - accuracy: 0.8599 - val_loss: 0.4755 - val_accuracy: 0.8944\n",
            "Epoch 5/20\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.5181 - accuracy: 0.8622 - val_loss: 0.4125 - val_accuracy: 0.8964\n",
            "Epoch 6/20\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.4726 - accuracy: 0.8626 - val_loss: 0.4228 - val_accuracy: 0.8908\n",
            "Epoch 7/20\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.4484 - accuracy: 0.8627 - val_loss: 0.3569 - val_accuracy: 0.8998\n",
            "Epoch 8/20\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.4334 - accuracy: 0.8641 - val_loss: 0.3745 - val_accuracy: 0.8918\n",
            "Epoch 9/20\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.4237 - accuracy: 0.8642 - val_loss: 0.3639 - val_accuracy: 0.8960\n",
            "Epoch 10/20\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.4184 - accuracy: 0.8673 - val_loss: 0.3686 - val_accuracy: 0.8886\n",
            "Epoch 11/20\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.4134 - accuracy: 0.8661 - val_loss: 0.3802 - val_accuracy: 0.8922\n",
            "Epoch 12/20\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.4079 - accuracy: 0.8656 - val_loss: 0.3406 - val_accuracy: 0.8992\n",
            "Epoch 13/20\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.4033 - accuracy: 0.8671 - val_loss: 0.3307 - val_accuracy: 0.8998\n",
            "Epoch 14/20\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.4018 - accuracy: 0.8674 - val_loss: 0.3201 - val_accuracy: 0.9096\n",
            "Epoch 15/20\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3988 - accuracy: 0.8671 - val_loss: 0.3159 - val_accuracy: 0.9090\n",
            "Epoch 16/20\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3981 - accuracy: 0.8675 - val_loss: 0.3416 - val_accuracy: 0.8938\n",
            "Epoch 17/20\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.3949 - accuracy: 0.8692 - val_loss: 0.3766 - val_accuracy: 0.8868\n",
            "Epoch 18/20\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3955 - accuracy: 0.8679 - val_loss: 0.3730 - val_accuracy: 0.8812\n",
            "Epoch 19/20\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3934 - accuracy: 0.8673 - val_loss: 0.3099 - val_accuracy: 0.9010\n",
            "Epoch 20/20\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.3925 - accuracy: 0.8668 - val_loss: 0.3376 - val_accuracy: 0.8990\n",
            "Epoch 1/20\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 58.2362 - accuracy: 0.7277 - val_loss: 7.1588 - val_accuracy: 0.5224\n",
            "Epoch 2/20\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 2.6555 - accuracy: 0.7933 - val_loss: 1.0499 - val_accuracy: 0.8020\n",
            "Epoch 3/20\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.7729 - accuracy: 0.8138 - val_loss: 0.5864 - val_accuracy: 0.8486\n",
            "Epoch 4/20\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.5716 - accuracy: 0.8206 - val_loss: 0.4814 - val_accuracy: 0.8682\n",
            "Epoch 5/20\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.5124 - accuracy: 0.8252 - val_loss: 0.4533 - val_accuracy: 0.8956\n",
            "Epoch 6/20\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.4851 - accuracy: 0.8260 - val_loss: 0.4303 - val_accuracy: 0.8786\n",
            "Epoch 7/20\n",
            "469/469 [==============================] - 3s 7ms/step - loss: 0.4739 - accuracy: 0.8261 - val_loss: 0.4775 - val_accuracy: 0.8618\n",
            "Epoch 8/20\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.4646 - accuracy: 0.8278 - val_loss: 0.4317 - val_accuracy: 0.8700\n",
            "Epoch 9/20\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.4586 - accuracy: 0.8273 - val_loss: 0.4115 - val_accuracy: 0.8948\n",
            "Epoch 10/20\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.4537 - accuracy: 0.8291 - val_loss: 0.4124 - val_accuracy: 0.8794\n",
            "Epoch 11/20\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.4513 - accuracy: 0.8275 - val_loss: 0.3794 - val_accuracy: 0.8834\n",
            "Epoch 12/20\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.4486 - accuracy: 0.8291 - val_loss: 0.3759 - val_accuracy: 0.9046\n",
            "Epoch 13/20\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.4464 - accuracy: 0.8293 - val_loss: 0.3971 - val_accuracy: 0.8812\n",
            "Epoch 14/20\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.4445 - accuracy: 0.8306 - val_loss: 0.3744 - val_accuracy: 0.9020\n",
            "Epoch 15/20\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.4457 - accuracy: 0.8296 - val_loss: 0.3760 - val_accuracy: 0.9036\n",
            "Epoch 16/20\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.4418 - accuracy: 0.8315 - val_loss: 0.4214 - val_accuracy: 0.8742\n",
            "Epoch 17/20\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.4418 - accuracy: 0.8300 - val_loss: 0.3751 - val_accuracy: 0.9030\n",
            "Epoch 18/20\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.4393 - accuracy: 0.8316 - val_loss: 0.3802 - val_accuracy: 0.8900\n",
            "Epoch 19/20\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.4393 - accuracy: 0.8298 - val_loss: 0.3992 - val_accuracy: 0.8844\n",
            "Epoch 20/20\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.4390 - accuracy: 0.8318 - val_loss: 0.4116 - val_accuracy: 0.8874\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "QNtOLlY5Y_lH",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 102
        },
        "outputId": "cb6dbe45-ffbf-4773-e0f4-d9b2e15fbf4d"
      },
      "source": [
        "print(Accuracies_biasedArtefacts)\n",
        "print(Accuracies_balanced_biasedArtefacts)\n",
        "print(Precisions_biasedArtefacts)\n",
        "print(Recalls_biasedArtefacts)\n",
        "print(AUCs_biasedArtefacts)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "[0.9158 0.9094 0.8922]\n",
            "[0.91523034 0.90815819 0.89056215]\n",
            "[0.8945414 0.864     0.836632 ]\n",
            "[0.9461901  0.97564808 0.97957581]\n",
            "[0.97113767 0.9697424  0.96984619]\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "zUkhJMWBW_TD"
      },
      "source": [
        "### Make a ROC curve"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "2mu6WdhJW_cS",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 514
        },
        "outputId": "5fd8c800-b260-40b2-901f-81f672088902"
      },
      "source": [
        "custom_blues = [\"#66CCFF\", \"#33BBFF\", \"#00AAFF\", \"#0088CC\", \"#006699\", \"#004466\"]\n",
        "\n",
        "x = np.linspace(0,1,100)\n",
        "# ==================================================================\n",
        "# ==================================================================\n",
        "fig, ax1 = plt.subplots(figsize = (6.5,6.5))\n",
        "# =============================================================\n",
        "# =============================================================\n",
        "# Main figure\n",
        "\n",
        "ax1.plot(fprs,tpr_baseline,c='coral',linewidth=2.0, label='No noise')\n",
        "ax1.plot(fprs_noise,tprs_noise[0],c=custom_blues[0],ls='--',linewidth=0.8, label='Noise 1$\\%$')\n",
        "ax1.plot(fprs_noise,tprs_noise[1],c=custom_blues[1],linewidth=0.8, label='Noise 5$\\%$')\n",
        "ax1.plot(fprs_noise,tprs_noise[2],c=custom_blues[4],ls='-.',linewidth=0.8, label='Noise 10$\\%$')\n",
        "ax1.plot(fprs_noise,tprs_noise[3],c=custom_blues[3],linewidth=0.8, label='Noise 33$\\%$')\n",
        "\n",
        "\n",
        "#Plot diagonal\n",
        "ax1.plot(x,x, c='k',ls='--',linewidth=0.8)\n",
        "ax1.grid(ls='--', alpha=0.6)\n",
        "\n",
        "ax1.set_xlabel('False Positive Rate',fontsize=14);ax1.set_ylabel('True Positive Rate',fontsize=14)\n",
        "ax1.tick_params(axis='both', labelsize=13)\n",
        "\n",
        "\n",
        "ax1.legend(frameon=True, loc='lower right', fontsize=15)\n",
        "# =============================================================\n",
        "# =============================================================\n",
        "# Inset figure\n",
        "left, bottom, width, height = [0.55, 0.45, 0.4, 0.4]\n",
        "ax2 = fig.add_axes([left, bottom, width, height])\n",
        "\n",
        "ax2.plot(fprs,tpr_baseline,c='coral',linewidth=1.8)\n",
        "ax2.plot(fprs_noise,tprs_noise[0],c=custom_blues[0],ls='--',linewidth=1.2)\n",
        "ax2.plot(fprs_noise,tprs_noise[1],c=custom_blues[1],linewidth=1.2)\n",
        "ax2.plot(fprs_noise,tprs_noise[2],c=custom_blues[4],ls='-.',linewidth=1.2)\n",
        "ax2.plot(fprs_noise,tprs_noise[3],c=custom_blues[3],linewidth=1.2)\n",
        "\n",
        "\n",
        "\n",
        "ax2.set_xlim(0.00,0.2)\n",
        "ax2.set_ylim(0.8,1.00)\n",
        "ax2.set_xticks([0.0,0.1,0.2])\n",
        "ax2.set_yticks([0.8,0.9,1.0])\n",
        "ax2.tick_params(axis='both', labelsize=12)\n",
        "ax2.grid(ls='--', alpha=0.6)\n",
        "# =============================================================\n",
        "# =============================================================\n",
        "# Format the whole figure\n",
        "# Mark the region corresponding to the inset axes on ax1 and draw lines\n",
        "# in grey linking the two axes.\n",
        "mark_inset(ax1, ax2, loc1=1, loc2=3, fc=\"none\",ls='--', ec='0.5')\n",
        "ax1.set_title('ROC curves - Different label noise levels',fontsize=14)\n",
        "\n",
        "\n",
        "\n",
        "# ==================================================================\n",
        "# ==================================================================\n",
        "plt.tight_layout()\n",
        "\n",
        "plt.savefig(\"/content/drive/My Drive/Plots/ROC_noisy.pdf\")\n",
        "#files.download(\"/content/drive/My Drive/Plots/ROC_noisy.pdf\")\n",
        "plt.show()"
      ],
      "execution_count": 45,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "/usr/local/lib/python3.6/dist-packages/ipykernel_launcher.py:59: UserWarning: This figure includes Axes that are not compatible with tight_layout, so results might be incorrect.\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 468x468 with 2 Axes>"
            ]
          },
          "metadata": {
            "tags": [],
            "needs_background": "light"
          }
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "2ziWwHOYfNbj",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 613
        },
        "outputId": "8befbb09-e2b1-479a-b175-177c04f12e87"
      },
      "source": [
        "x = np.linspace(0.0,0.5,100)\n",
        "y_loc = np.ones(100)\n",
        "plt.figure(figsize = (6.5,6.5))\n",
        "\n",
        "\n",
        "plt.plot(fprs_noise,tprs_noise[0]/tpr_baseline,c=custom_blues[0],ls='--',linewidth=1.5,label='Noise 1$\\%$')\n",
        "plt.plot(fprs_noise,tprs_noise[1]/tpr_baseline,c=custom_blues[1],linewidth=1.5,label='Noise 5$\\%$')\n",
        "plt.plot(fprs_noise,tprs_noise[2]/tpr_baseline,c=custom_blues[4],ls='-.',linewidth=1.5,label='Noise 10$\\%$')\n",
        "plt.plot(fprs_noise,tprs_noise[3]/tpr_baseline,c=custom_blues[3],linewidth=1.5,label='Noise 33$\\%$')\n",
        "\n",
        "plt.plot(x,y_loc,c='coral',ls='--',linewidth=1.9)\n",
        "\n",
        "plt.xlim(0,0.5)\n",
        "plt.ylim(0.8,1.1)\n",
        "\n",
        "plt.tick_params(axis='both', labelsize=12)\n",
        "plt.grid(ls='--', alpha=0.6)\n",
        "\n",
        "plt.xlabel('False Positive Rate',fontsize=14)\n",
        "plt.ylabel('Ratio TPR/TPR(no noise)', fontsize=14)\n",
        "plt.legend(frameon=True, loc='lower right', fontsize=16)\n",
        "\n",
        "\n",
        "plt.title('Ratios of ROC curves', fontsize=15)\n",
        "plt.tight_layout()\n",
        "\n",
        "plt.savefig(\"/content/drive/My Drive/Plots/ROC_ratios.pdf\")\n",
        "#files.download(\"/content/drive/My Drive/Plots/ROC_ratios.pdf\")\n",
        "plt.show()"
      ],
      "execution_count": 74,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "/usr/local/lib/python3.6/dist-packages/ipykernel_launcher.py:6: RuntimeWarning: invalid value encountered in true_divide\n",
            "  \n",
            "/usr/local/lib/python3.6/dist-packages/ipykernel_launcher.py:7: RuntimeWarning: invalid value encountered in true_divide\n",
            "  import sys\n",
            "/usr/local/lib/python3.6/dist-packages/ipykernel_launcher.py:8: RuntimeWarning: invalid value encountered in true_divide\n",
            "  \n",
            "/usr/local/lib/python3.6/dist-packages/ipykernel_launcher.py:9: RuntimeWarning: invalid value encountered in true_divide\n",
            "  if __name__ == '__main__':\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "display_data",
          "data": {
            "application/javascript": [
              "\n",
              "    async function download(id, filename, size) {\n",
              "      if (!google.colab.kernel.accessAllowed) {\n",
              "        return;\n",
              "      }\n",
              "      const div = document.createElement('div');\n",
              "      const label = document.createElement('label');\n",
              "      label.textContent = `Downloading \"${filename}\": `;\n",
              "      div.appendChild(label);\n",
              "      const progress = document.createElement('progress');\n",
              "      progress.max = size;\n",
              "      div.appendChild(progress);\n",
              "      document.body.appendChild(div);\n",
              "\n",
              "      const buffers = [];\n",
              "      let downloaded = 0;\n",
              "\n",
              "      const channel = await google.colab.kernel.comms.open(id);\n",
              "      // Send a message to notify the kernel that we're ready.\n",
              "      channel.send({})\n",
              "\n",
              "      for await (const message of channel.messages) {\n",
              "        // Send a message to notify the kernel that we're ready.\n",
              "        channel.send({})\n",
              "        if (message.buffers) {\n",
              "          for (const buffer of message.buffers) {\n",
              "            buffers.push(buffer);\n",
              "            downloaded += buffer.byteLength;\n",
              "            progress.value = downloaded;\n",
              "          }\n",
              "        }\n",
              "      }\n",
              "      const blob = new Blob(buffers, {type: 'application/binary'});\n",
              "      const a = document.createElement('a');\n",
              "      a.href = window.URL.createObjectURL(blob);\n",
              "      a.download = filename;\n",
              "      div.appendChild(a);\n",
              "      a.click();\n",
              "      div.remove();\n",
              "    }\n",
              "  "
            ],
            "text/plain": [
              "<IPython.core.display.Javascript object>"
            ]
          },
          "metadata": {
            "tags": []
          }
        },
        {
          "output_type": "display_data",
          "data": {
            "application/javascript": [
              "download(\"download_222bcbca-6cdb-45ea-bd64-62fee152ba08\", \"ROC_ratios.pdf\", 19571)"
            ],
            "text/plain": [
              "<IPython.core.display.Javascript object>"
            ]
          },
          "metadata": {
            "tags": []
          }
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 468x468 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": [],
            "needs_background": "light"
          }
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "gOaF8YwWQeUq"
      },
      "source": [
        ""
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "D35_Ymy7QehH"
      },
      "source": [
        ""
      ],
      "execution_count": null,
      "outputs": []
    }
  ]
}