Object_Detection_Section1.ipynb

{
  "nbformat": 4,
  "nbformat_minor": 0,
  "metadata": {
    "colab": {
      "name": "Object-Detection-Section1.ipynb",
      "provenance": [],
      "authorship_tag": "ABX9TyNof1GB/Vfn+kiblTgt860G",
      "include_colab_link": true
    },
    "kernelspec": {
      "name": "python3",
      "display_name": "Python 3"
    },
    "language_info": {
      "name": "python"
    }
  },
  "cells": [
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "view-in-github",
        "colab_type": "text"
      },
      "source": [
        "<a href=\"https://colab.research.google.com/github/SXV357/Object_Detection/blob/main/Object_Detection_Section1.ipynb\" target=\"_parent\"><img src=\"https://colab.research.google.com/assets/colab-badge.svg\" alt=\"Open In Colab\"/></a>"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": 1,
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "hQsh8RsGAPTu",
        "outputId": "87357b5a-093e-4bba-b096-8ee234a464c4"
      },
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "TensorFlow version: 2.8.2\n"
          ]
        }
      ],
      "source": [
        "import os\n",
        "import pandas as pd\n",
        "import seaborn as sns\n",
        "import numpy as np\n",
        "import matplotlib.pyplot as plt\n",
        "import tensorflow as tf\n",
        "from tensorflow.keras.datasets import cifar10\n",
        "\n",
        "print(\"TensorFlow version: {}\".format(tf.__version__))\n",
        "\n",
        "# Load data\n",
        "def load_cifar10():\n",
        "  (x_train_cifar, y_train_cifar), (x_test_cifar, y_test_cifar) = cifar10.load_data()\n",
        "  y_train_cifar = y_train_cifar.squeeze()\n",
        "  y_test_cifar = y_test_cifar.squeeze()\n",
        "  return (x_train_cifar, y_train_cifar), (x_test_cifar, y_test_cifar)\n",
        "\n",
        "# CIFAR100 classes\n",
        "idx_to_class = ['background', 'car', 'truck']\n",
        "\n",
        "# Construct vehicle dataset from CIFAR10\n",
        "def construct_vehicle_dataset(data, labels, images_per_class, label_car=1, label_truck=9):\n",
        "  mask_car = labels == label_car\n",
        "  mask_truck = labels == label_truck\n",
        "\n",
        "  mask_vehicles = mask_car | mask_truck\n",
        "  mask_background = np.invert(mask_vehicles)\n",
        "  \n",
        "  data_car = data[mask_car]\n",
        "  data_truck = data[mask_truck]\n",
        "  data_background = data[mask_background][:images_per_class]\n",
        "\n",
        "  new_data = np.vstack((data_background, data_car, data_truck))\n",
        "  new_labels = np.repeat(np.array([0, 1, 2]), images_per_class, axis=0)\n",
        "  \n",
        "  return new_data, new_labels\n",
        "\n",
        "def load_vehicle_dataset():\n",
        "  (x_train_cifar, y_train_cifar), (x_test_cifar, y_test_cifar) = load_cifar10()\n",
        "  x_train, y_train = construct_vehicle_dataset(x_train_cifar, y_train_cifar, 5000)\n",
        "  x_test, y_test = construct_vehicle_dataset(x_test_cifar, y_test_cifar, 1000)\n",
        "  return (x_train, y_train), (x_test, y_test)\n",
        "\n",
        "# Helper functions\n",
        "\n",
        "# plotting\n",
        "def plot_one_image(data, labels = [], index = None, image_shape = None):\n",
        "  '''\n",
        "  if data is a single image, display that image\n",
        "\n",
        "  if data is a 4d stack of images, display that image\n",
        "  '''\n",
        "  ### cv2.imshow('image', data)    \n",
        "  num_dims   = len(data.shape)\n",
        "  num_labels = len(labels)\n",
        "  if image_shape is not None:\n",
        "    target_shape = image_shape\n",
        "  else:\n",
        "    target_shape = (32, 32, 3)\n",
        "  # reshape data if necessary\n",
        "  if num_dims == 1:\n",
        "    data = data.reshape(target_shape)\n",
        "  if num_dims == 2:\n",
        "    data = data.reshape(np.vstack[-1, image_shape])\n",
        "  num_dims   = len(data.shape)\n",
        "\n",
        "  # check if single or multiple images\n",
        "  if num_dims == 3:\n",
        "    if num_labels > 1:\n",
        "      print('Multiple labels does not make sense for single image.')\n",
        "      return\n",
        "\n",
        "    label = labels      \n",
        "    if num_labels == 0:\n",
        "      label = ''\n",
        "    image = data\n",
        "\n",
        "  if num_dims == 4:\n",
        "    image = data[index, :]\n",
        "    label = labels[index]\n",
        "\n",
        "  # plot image of interest\n",
        "  print('Label: %s'%label)\n",
        "  plt.imshow(image)\n",
        "  plt.show()\n",
        "\n",
        "def model_to_string(model):\n",
        "  import re\n",
        "  stringlist = []\n",
        "  model.summary(print_fn=lambda x: stringlist.append(x))\n",
        "  sms = \"\\n\".join(stringlist)\n",
        "  sms = re.sub('_\\d\\d\\d','', sms)\n",
        "  sms = re.sub('_\\d\\d','', sms)\n",
        "  sms = re.sub('_\\d','', sms)  \n",
        "  return sms\n",
        "\n",
        "def normalize(data):\n",
        "  # CIFAR100 mean (0.4914, 0.4822, 0.4465) std (0.2023, 0.1994, 0.2010)\n",
        "  return (data/255-np.array((0.4914, 0.4822, 0.4465))) / np.array((0.2023, 0.1994, 0.2010))\n",
        "\n",
        "def label_to_onehot(labels):\n",
        "  final_labels = np.zeros((len(labels), 3))\n",
        "  for i in range(len(labels)):\n",
        "    label = labels[i]\n",
        "    if label == 0:\n",
        "      final_labels[i,:] = np.array([1, 0, 0])\n",
        "    if label == 1:\n",
        "      final_labels[i,:] = np.array([0, 1, 0])\n",
        "    if label == 2:\n",
        "      final_labels[i,:] = np.array([0, 0, 1])\n",
        "  return final_labels\n",
        "\n",
        "def plot_acc(history, ax = None, xlabel = 'Epoch #'):\n",
        "  # i'm sorry for this function's code. i am so sorry. \n",
        "  history = history.history\n",
        "  history.update({'epoch':list(range(len(history['val_accuracy'])))})\n",
        "  history = pd.DataFrame.from_dict(history)\n",
        "\n",
        "  best_epoch = history.sort_values(by = 'val_accuracy', ascending = False).iloc[0]['epoch']\n",
        "\n",
        "  if not ax:\n",
        "    f, ax = plt.subplots(1,1)\n",
        "  sns.lineplot(x = 'epoch', y = 'val_accuracy', data = history, label = 'Validation', ax = ax)\n",
        "  sns.lineplot(x = 'epoch', y = 'accuracy', data = history, label = 'Training', ax = ax)\n",
        "  ax.axhline(0.333, linestyle = '--',color='red', label = 'Chance')\n",
        "  ax.axvline(x = best_epoch, linestyle = '--', color = 'green', label = 'Best Epoch')  \n",
        "  ax.legend(loc = 1)    \n",
        "  ax.set_ylim([0.01, 1])\n",
        "\n",
        "  ax.set_xlabel(xlabel)\n",
        "  ax.set_ylabel('Accuracy (Fraction)')\n",
        "  \n",
        "  plt.show()"
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "#Loading in training and testing datasets\n",
        "(x_train, y_train), (x_test, y_test) = load_vehicle_dataset()"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "fKyhCsEKAhWC",
        "outputId": "af66bfbc-40ef-478f-85cb-df06b0804bd5"
      },
      "execution_count": 2,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Downloading data from https://www.cs.toronto.edu/~kriz/cifar-10-python.tar.gz\n",
            "170500096/170498071 [==============================] - 4s 0us/step\n",
            "170508288/170498071 [==============================] - 4s 0us/step\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "#checking type and shape of labels training + testing dataset\n",
        "print(f'Our labels are stored as {type(y_train)} in Python')\n",
        "print(f'The label vector of the training set has dimensions of {y_train.shape}') # 15000 training labels\n",
        "print(f'The label vector of the test set has dimensions of {y_test.shape}') # 3000 testing labels"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "XOGo8hEnArgZ",
        "outputId": "1573de79-df3e-4f22-e697-c8ef84900d0f"
      },
      "execution_count": 3,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Our labels are stored as <class 'numpy.ndarray'> in Python\n",
            "The label vector of the training set has dimensions of (15000,)\n",
            "The label vector of the test set has dimensions of (3000,)\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "for index, element in enumerate(idx_to_class):\n",
        "  print(f'Class {index} contains {element} images')"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "2OPpOuoCA0mx",
        "outputId": "3429f2fd-dab8-4459-f3a0-8491081ab1ba"
      },
      "execution_count": 4,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Class 0 contains background images\n",
            "Class 1 contains car images\n",
            "Class 2 contains truck images\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "# checking number of images in each class\n",
        "import collections\n",
        "\n",
        "print(collections.Counter(y_train)) # tells how many images per class(5x more for training dataset than testing dataset)\n",
        "print(collections.Counter(y_test))\n"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "yZXJyis5A5c4",
        "outputId": "d1a5c45d-8eff-4257-e679-304777c8e6f0"
      },
      "execution_count": 5,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Counter({0: 5000, 1: 5000, 2: 5000})\n",
            "Counter({0: 1000, 1: 1000, 2: 1000})\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "print(x_train.shape) # 15,000 images with height = 32, width = 32 and 3 color spectrums\n",
        "print(x_test.shape) # 3,000 images with height = 32, width = 32 and 3 color spectrums"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "4f1nogS0A_Ou",
        "outputId": "61a24e88-e473-4e06-bfb2-1e41626cc4fb"
      },
      "execution_count": 6,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "(15000, 32, 32, 3)\n",
            "(3000, 32, 32, 3)\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "# plot a SINGLE image\n",
        "plot_one_image(x_train, y_train, 1200) # 1st 5000 are background, next 5000 are cars, and last 5000 are trucks"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 283
        },
        "id": "E6uM4LvwBB0t",
        "outputId": "db2a59fc-8adb-4b18-adae-f5fe104721f3"
      },
      "execution_count": 7,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Label: 0\n"
          ]
        },
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ],
            "image/png": "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\n"
          },
          "metadata": {
            "needs_background": "light"
          }
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "# building a simple neural network\n",
        "import tensorflow as tf\n",
        "from tensorflow.keras.models import Sequential\n",
        "from tensorflow.keras.layers import Activation, Dropout, Flatten, Dense, MaxPooling2D, Convolution2D\n",
        "from tensorflow.keras import optimizers\n",
        "\n",
        "model = Sequential()\n",
        "model.add(Dense(4, input_shape=(3,),activation = 'relu'))\n",
        "model.add(Dense(1, activation = 'linear'))\n",
        "model.compile(loss='mean_squared_error',\n",
        "                optimizer='adam',\n",
        "                metrics=['mean_squared_error'])"
      ],
      "metadata": {
        "id": "ltCaQoqwBHEj"
      },
      "execution_count": 9,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "# building a 2-layer model\n",
        "model_1 = Sequential()\n",
        "model_1.add(Dense(4, input_shape=(3,),activation = 'relu'))\n",
        "model_1.add(Dense(2, activation = 'softmax'))\n",
        "model_1.compile(loss = 'categorical_crossentropy', optimizer = 'adam', metrics = ['accuracy'])\n",
        "\n",
        "# checking to see if model works\n",
        "model_1_answer = Sequential()\n",
        "model_1_answer.add(Dense(4, input_shape = (3,), activation = 'relu'))\n",
        "model_1_answer.add(Dense(2, activation = 'softmax'))\n",
        "model_1_answer.compile(loss='categorical_crossentropy',\n",
        "optimizer = 'adam', \n",
        "metrics = ['accuracy'])\n",
        "\n",
        "if model_to_string(model_1) == model_to_string(model_1_answer):\n",
        "  print('Good job! Your model worked')\n",
        "else: \n",
        "  print('Please check your code again!')"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "ZOcEg83ABV4n",
        "outputId": "6efb6b67-4a0a-4ede-e726-0b1520cb7f14"
      },
      "execution_count": 10,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Good job! Your model worked\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "#building a custom 2-layer neural network using a multi-layer perceptron model\n",
        "\n",
        "perceptron = Sequential()\n",
        "\n",
        "perceptron.add(Convolution2D(64, 3, 3, input_shape = (32, 32, 3), activation = 'linear'))\n",
        "perceptron.add(Flatten(input_shape = (32, 32, 3)))\n",
        "\n",
        "perceptron.add(Dense(312, activation = 'relu'))\n",
        "perceptron.add(Dense(256, activation = 'relu'))\n",
        "perceptron.add(Dense(256, activation = 'linear'))\n",
        "perceptron.add(Dropout(0.4))\n",
        "perceptron.add(Dropout(0.25))\n",
        "perceptron.add(Dense(3, activation = 'sigmoid'))\n",
        "\n",
        "perceptron.compile(loss = 'categorical_crossentropy', \n",
        "                   optimizer = optimizers.SGD(learning_rate=1e-3, momentum=0.9), \n",
        "                   metrics = ['accuracy'])\n",
        "\n",
        "# checking to see if model works\n",
        "\n",
        "perceptron_answer = Sequential()\n",
        "perceptron = perceptron_answer\n",
        "perceptron_answer.add(Flatten(input_shape = (32, 32, 3)))\n",
        "perceptron_answer.add(Dense(units = 128, activation = 'relu'))\n",
        "perceptron_answer.add(Dense(units = 3, activation = 'softmax'))\n",
        "    \n",
        "perceptron_answer.compile(loss='categorical_crossentropy',\n",
        "              optimizer=optimizers.SGD(learning_rate=1e-3, momentum=0.9),\n",
        "              metrics=['accuracy'])\n",
        "\n",
        "if model_to_string(perceptron) == model_to_string(perceptron_answer):\n",
        "  print('Good job, you specified it correctly!')\n",
        "else: \n",
        "  print('Please check your code again!')\n"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "foeRDGOQBkZ3",
        "outputId": "23c6bf76-7f50-48bd-ebf6-d06e77340720"
      },
      "execution_count": 13,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Good job, you specified it correctly!\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "#training perceptron on images from training data\n",
        "\n",
        "from tensorflow.keras.callbacks import ModelCheckpoint\n",
        "\n",
        "# define our monitor\n",
        "monitor = ModelCheckpoint('./model.h5', monitor='val_loss', verbose=0, save_best_only=True, save_weights_only=False, mode='auto', save_freq='epoch')\n",
        "\n",
        "# Normalize the data.\n",
        "x_train_norm = normalize(x_train)\n",
        "x_test_norm = normalize(x_test)\n",
        "\n",
        "# Convert labels into one-hot numpy arrays.\n",
        "y_train_onehot = label_to_onehot(y_train)\n",
        "y_test_onehot = label_to_onehot(y_test)\n",
        "\n",
        "\n",
        "perceptron.fit(x_train_norm, y_train_onehot)\n"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "lNxZ7mqxB7r4",
        "outputId": "99a73b70-ef0e-4647-c6ec-835b98c87162"
      },
      "execution_count": 14,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "469/469 [==============================] - 2s 3ms/step - loss: 0.8818 - accuracy: 0.6279\n"
          ]
        },
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "<keras.callbacks.History at 0x7fc5644fa8d0>"
            ]
          },
          "metadata": {},
          "execution_count": 14
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "# plotting training vs testing accuracy of model\n",
        "\n",
        "history = perceptron.fit(x_train_norm, y_train_onehot, epochs = 50, validation_data = (x_test_norm, y_test_onehot), shuffle = True, callbacks = [monitor])\n",
        "plot_acc(history)"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 1000
        },
        "id": "qPkmQffgCGVX",
        "outputId": "2b9851bd-d5a0-4815-f367-7462fdc1312c"
      },
      "execution_count": 15,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Epoch 1/50\n",
            "469/469 [==============================] - 2s 4ms/step - loss: 0.6732 - accuracy: 0.7185 - val_loss: 0.7146 - val_accuracy: 0.6970\n",
            "Epoch 2/50\n",
            "469/469 [==============================] - 2s 4ms/step - loss: 0.6002 - accuracy: 0.7581 - val_loss: 0.7110 - val_accuracy: 0.7070\n",
            "Epoch 3/50\n",
            "469/469 [==============================] - 2s 4ms/step - loss: 0.5529 - accuracy: 0.7793 - val_loss: 0.7197 - val_accuracy: 0.6990\n",
            "Epoch 4/50\n",
            "469/469 [==============================] - 3s 6ms/step - loss: 0.5116 - accuracy: 0.7987 - val_loss: 0.6899 - val_accuracy: 0.7180\n",
            "Epoch 5/50\n",
            "469/469 [==============================] - 2s 4ms/step - loss: 0.4749 - accuracy: 0.8145 - val_loss: 0.7257 - val_accuracy: 0.7133\n",
            "Epoch 6/50\n",
            "469/469 [==============================] - 2s 4ms/step - loss: 0.4383 - accuracy: 0.8287 - val_loss: 0.7352 - val_accuracy: 0.7100\n",
            "Epoch 7/50\n",
            "469/469 [==============================] - 2s 4ms/step - loss: 0.4078 - accuracy: 0.8469 - val_loss: 0.7253 - val_accuracy: 0.7047\n",
            "Epoch 8/50\n",
            "469/469 [==============================] - 2s 4ms/step - loss: 0.3784 - accuracy: 0.8581 - val_loss: 0.7671 - val_accuracy: 0.7080\n",
            "Epoch 9/50\n",
            "469/469 [==============================] - 2s 4ms/step - loss: 0.3509 - accuracy: 0.8697 - val_loss: 0.7699 - val_accuracy: 0.7067\n",
            "Epoch 10/50\n",
            "469/469 [==============================] - 2s 4ms/step - loss: 0.3218 - accuracy: 0.8821 - val_loss: 0.7711 - val_accuracy: 0.7227\n",
            "Epoch 11/50\n",
            "469/469 [==============================] - 2s 4ms/step - loss: 0.2931 - accuracy: 0.8927 - val_loss: 0.8125 - val_accuracy: 0.7187\n",
            "Epoch 12/50\n",
            "469/469 [==============================] - 2s 4ms/step - loss: 0.2718 - accuracy: 0.9067 - val_loss: 0.8113 - val_accuracy: 0.7167\n",
            "Epoch 13/50\n",
            "469/469 [==============================] - 2s 4ms/step - loss: 0.2460 - accuracy: 0.9147 - val_loss: 0.8342 - val_accuracy: 0.7177\n",
            "Epoch 14/50\n",
            "469/469 [==============================] - 2s 4ms/step - loss: 0.2283 - accuracy: 0.9218 - val_loss: 0.8563 - val_accuracy: 0.7150\n",
            "Epoch 15/50\n",
            "469/469 [==============================] - 2s 5ms/step - loss: 0.1984 - accuracy: 0.9339 - val_loss: 0.8784 - val_accuracy: 0.7163\n",
            "Epoch 16/50\n",
            "469/469 [==============================] - 2s 4ms/step - loss: 0.1821 - accuracy: 0.9429 - val_loss: 0.9462 - val_accuracy: 0.7067\n",
            "Epoch 17/50\n",
            "469/469 [==============================] - 2s 4ms/step - loss: 0.1622 - accuracy: 0.9503 - val_loss: 0.9465 - val_accuracy: 0.7167\n",
            "Epoch 18/50\n",
            "469/469 [==============================] - 2s 4ms/step - loss: 0.1557 - accuracy: 0.9516 - val_loss: 0.9689 - val_accuracy: 0.7100\n",
            "Epoch 19/50\n",
            "469/469 [==============================] - 2s 4ms/step - loss: 0.1333 - accuracy: 0.9608 - val_loss: 1.0123 - val_accuracy: 0.7123\n",
            "Epoch 20/50\n",
            "469/469 [==============================] - 2s 4ms/step - loss: 0.1203 - accuracy: 0.9653 - val_loss: 1.0812 - val_accuracy: 0.7087\n",
            "Epoch 21/50\n",
            "469/469 [==============================] - 2s 4ms/step - loss: 0.1041 - accuracy: 0.9729 - val_loss: 1.0134 - val_accuracy: 0.7193\n",
            "Epoch 22/50\n",
            "469/469 [==============================] - 2s 4ms/step - loss: 0.1181 - accuracy: 0.9669 - val_loss: 1.0240 - val_accuracy: 0.7153\n",
            "Epoch 23/50\n",
            "469/469 [==============================] - 2s 4ms/step - loss: 0.0851 - accuracy: 0.9787 - val_loss: 1.0467 - val_accuracy: 0.7190\n",
            "Epoch 24/50\n",
            "469/469 [==============================] - 2s 4ms/step - loss: 0.0801 - accuracy: 0.9802 - val_loss: 1.1114 - val_accuracy: 0.7040\n",
            "Epoch 25/50\n",
            "469/469 [==============================] - 2s 4ms/step - loss: 0.0763 - accuracy: 0.9829 - val_loss: 1.1194 - val_accuracy: 0.7093\n",
            "Epoch 26/50\n",
            "469/469 [==============================] - 2s 4ms/step - loss: 0.0559 - accuracy: 0.9899 - val_loss: 1.1300 - val_accuracy: 0.7107\n",
            "Epoch 27/50\n",
            "469/469 [==============================] - 2s 4ms/step - loss: 0.0505 - accuracy: 0.9916 - val_loss: 1.2185 - val_accuracy: 0.6973\n",
            "Epoch 28/50\n",
            "469/469 [==============================] - 2s 4ms/step - loss: 0.0403 - accuracy: 0.9947 - val_loss: 1.1664 - val_accuracy: 0.7177\n",
            "Epoch 29/50\n",
            "469/469 [==============================] - 2s 4ms/step - loss: 0.0311 - accuracy: 0.9972 - val_loss: 1.1560 - val_accuracy: 0.7203\n",
            "Epoch 30/50\n",
            "469/469 [==============================] - 2s 4ms/step - loss: 0.0290 - accuracy: 0.9973 - val_loss: 1.1909 - val_accuracy: 0.7100\n",
            "Epoch 31/50\n",
            "469/469 [==============================] - 2s 4ms/step - loss: 0.0258 - accuracy: 0.9983 - val_loss: 1.1912 - val_accuracy: 0.7140\n",
            "Epoch 32/50\n",
            "469/469 [==============================] - 2s 4ms/step - loss: 0.0223 - accuracy: 0.9987 - val_loss: 1.2504 - val_accuracy: 0.7133\n",
            "Epoch 33/50\n",
            "469/469 [==============================] - 2s 4ms/step - loss: 0.0202 - accuracy: 0.9988 - val_loss: 1.2191 - val_accuracy: 0.7220\n",
            "Epoch 34/50\n",
            "469/469 [==============================] - 2s 4ms/step - loss: 0.0183 - accuracy: 0.9994 - val_loss: 1.2438 - val_accuracy: 0.7143\n",
            "Epoch 35/50\n",
            "469/469 [==============================] - 2s 4ms/step - loss: 0.0160 - accuracy: 0.9995 - val_loss: 1.2449 - val_accuracy: 0.7200\n",
            "Epoch 36/50\n",
            "469/469 [==============================] - 2s 4ms/step - loss: 0.0143 - accuracy: 0.9997 - val_loss: 1.2453 - val_accuracy: 0.7167\n",
            "Epoch 37/50\n",
            "469/469 [==============================] - 2s 4ms/step - loss: 0.0127 - accuracy: 0.9998 - val_loss: 1.2580 - val_accuracy: 0.7143\n",
            "Epoch 38/50\n",
            "469/469 [==============================] - 2s 4ms/step - loss: 0.0122 - accuracy: 0.9998 - val_loss: 1.2677 - val_accuracy: 0.7153\n",
            "Epoch 39/50\n",
            "469/469 [==============================] - 2s 4ms/step - loss: 0.0110 - accuracy: 0.9999 - val_loss: 1.2845 - val_accuracy: 0.7167\n",
            "Epoch 40/50\n",
            "469/469 [==============================] - 3s 5ms/step - loss: 0.0101 - accuracy: 0.9999 - val_loss: 1.2857 - val_accuracy: 0.7143\n",
            "Epoch 41/50\n",
            "469/469 [==============================] - 2s 5ms/step - loss: 0.0099 - accuracy: 0.9999 - val_loss: 1.3086 - val_accuracy: 0.7143\n",
            "Epoch 42/50\n",
            "469/469 [==============================] - 2s 4ms/step - loss: 0.0092 - accuracy: 1.0000 - val_loss: 1.3023 - val_accuracy: 0.7187\n",
            "Epoch 43/50\n",
            "469/469 [==============================] - 2s 4ms/step - loss: 0.0087 - accuracy: 0.9999 - val_loss: 1.3206 - val_accuracy: 0.7163\n",
            "Epoch 44/50\n",
            "469/469 [==============================] - 2s 4ms/step - loss: 0.0083 - accuracy: 0.9999 - val_loss: 1.3302 - val_accuracy: 0.7180\n",
            "Epoch 45/50\n",
            "469/469 [==============================] - 2s 4ms/step - loss: 0.0082 - accuracy: 1.0000 - val_loss: 1.3301 - val_accuracy: 0.7183\n",
            "Epoch 46/50\n",
            "469/469 [==============================] - 2s 4ms/step - loss: 0.0075 - accuracy: 1.0000 - val_loss: 1.3302 - val_accuracy: 0.7183\n",
            "Epoch 47/50\n",
            "469/469 [==============================] - 2s 4ms/step - loss: 0.0071 - accuracy: 1.0000 - val_loss: 1.3420 - val_accuracy: 0.7170\n",
            "Epoch 48/50\n",
            "469/469 [==============================] - 2s 4ms/step - loss: 0.0068 - accuracy: 1.0000 - val_loss: 1.3477 - val_accuracy: 0.7150\n",
            "Epoch 49/50\n",
            "469/469 [==============================] - 2s 5ms/step - loss: 0.0066 - accuracy: 1.0000 - val_loss: 1.3563 - val_accuracy: 0.7170\n",
            "Epoch 50/50\n",
            "469/469 [==============================] - 2s 4ms/step - loss: 0.0064 - accuracy: 0.9999 - val_loss: 1.3568 - val_accuracy: 0.7190\n"
          ]
        },
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ],
            "image/png": "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\n"
          },
          "metadata": {
            "needs_background": "light"
          }
        }
      ]
    }
  ]
}