train.ipynb

{
  "nbformat": 4,
  "nbformat_minor": 0,
  "metadata": {
    "kernelspec": {
      "display_name": "Python 3",
      "language": "python",
      "name": "python3"
    },
    "language_info": {
      "codemirror_mode": {
        "name": "ipython",
        "version": 3
      },
      "file_extension": ".py",
      "mimetype": "text/x-python",
      "name": "python",
      "nbconvert_exporter": "python",
      "pygments_lexer": "ipython3",
      "version": "3.7.9"
    },
    "colab": {
      "name": "train.ipynb",
      "provenance": [],
      "collapsed_sections": []
    }
  },
  "cells": [
    {
      "cell_type": "code",
      "metadata": {
        "id": "InSjFyqaHBBG"
      },
      "source": [
        "import datetime\n",
        "import numpy as np\n",
        "import matplotlib.pyplot as plt\n",
        "from keras.layers import Input, Activation, Conv2D, Flatten, Dense, MaxPooling2D\n",
        "from keras.models import Model, load_model\n",
        "# 데이터를 인위적으로 변형하여 증가(학습데이터 부족시 사용)\n",
        "from keras.preprocessing.image import ImageDataGenerator\n",
        "from keras.callbacks import ModelCheckpoint, ReduceLROnPlateau\n",
        "plt.style.use('dark_background')"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "XK-UtT2rHBBQ"
      },
      "source": [
        "# Load Dataset"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "UPvrocfzHBBR",
        "outputId": "e60ee887-a13f-41bd-8c9f-ccbc96761a33"
      },
      "source": [
        "# 데이터 로드\n",
        "x_train = np.load('dataset/x_train.npy').astype(np.float32) \n",
        "y_train = np.load('dataset/y_train.npy').astype(np.float32)\n",
        "x_val = np.load('dataset/x_val.npy').astype(np.float32)\n",
        "y_val = np.load('dataset/y_val.npy').astype(np.float32)\n",
        "\n",
        "print(x_train.shape, y_train.shape)\n",
        "print(x_val.shape, y_val.shape)\n",
        "# 0: 눈 감음 / 1: 눈 뜸"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "(2586, 26, 34, 1) (2586, 1)\n",
            "(288, 26, 34, 1) (288, 1)\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "sdb0Q2luHBBb"
      },
      "source": [
        "# Preview"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "tR2JondHHBBc",
        "outputId": "a6a9bd90-7d21-40bd-c307-e3881d5049b8"
      },
      "source": [
        "plt.subplot(2, 1, 1)\n",
        "plt.title(str(y_train[0]))\n",
        "plt.imshow(x_train[0].reshape((26, 34)), cmap='gray')\n",
        "plt.subplot(2, 1, 2)\n",
        "plt.title(str(y_val[4]))\n",
        "plt.imshow(x_val[4].reshape((26, 34)), cmap='gray')"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "<matplotlib.image.AxesImage at 0x197b652de88>"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 6
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 432x288 with 2 Axes>"
            ]
          },
          "metadata": {
            "tags": []
          }
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "fZj74gZbHBBj"
      },
      "source": [
        "# Data Augmentation"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "4HAJbG6SHBBk"
      },
      "source": [
        "# 데이터가 부족할 때 사용 -> 이미지형태가 다양해지므로 모델이 좀 더 잘 학습할 수 있음\n",
        "train_datagen = ImageDataGenerator( # 이미지 처리를 도와주는 클래스\n",
        "    rescale=1./255, # 원래이미지(0~255) -> 255로 나누어 줌으로써 0~1 값으로 변경\n",
        "    rotation_range=10, # 10도 돌리기\n",
        "    width_shift_range=0.2, # 가로크기 변경\n",
        "    height_shift_range=0.2, # 세로크기 변경\n",
        "    shear_range=0.2 # 이미지를 약간 비틈\n",
        ")\n",
        "\n",
        "# 검증 데이터셋은 데이터를 비틀 필요가 없으므로 rescale만 진행\n",
        "val_datagen = ImageDataGenerator(rescale=1./255)\n",
        "\n",
        "train_generator = train_datagen.flow(\n",
        "    x=x_train, y=y_train,\n",
        "    batch_size=32, # 32개씩 데이터 꺼내옴\n",
        "    shuffle=True # 순서를 섞음\n",
        ")\n",
        "\n",
        "val_generator = val_datagen.flow(\n",
        "    x=x_val, y=y_val,\n",
        "    batch_size=32,\n",
        "    shuffle=False\n",
        ")"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "6P3Ytw10HBBq"
      },
      "source": [
        "# Build Model"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "CQfPYg6THBBr",
        "outputId": "b849ced7-bb9c-425f-9b4a-962ebfcb6e80"
      },
      "source": [
        "inputs = Input(shape=(26, 34, 1)) # 세로26 가로34 그레이스케일인 이미지\n",
        "\n",
        "# 커널수만 변경 32->64->128\n",
        "net = Conv2D(32, kernel_size=3, strides=1, padding='same', activation='relu')(inputs)\n",
        "net = MaxPooling2D(pool_size=2)(net) # 차원축소\n",
        "\n",
        "net = Conv2D(64, kernel_size=3, strides=1, padding='same', activation='relu')(net)\n",
        "net = MaxPooling2D(pool_size=2)(net)\n",
        "\n",
        "net = Conv2D(128, kernel_size=3, strides=1, padding='same', activation='relu')(net)\n",
        "net = MaxPooling2D(pool_size=2)(net)\n",
        "\n",
        "net = Flatten()(net) # convolution & Pooling layer(특징추출)를 1차원으로 펴줌\n",
        "\n",
        "net = Dense(512)(net) # 512개로 수렴. fully-connected layer(분류)와 연결\n",
        "net = Activation('relu')(net)\n",
        "net = Dense(1)(net) # 마지막 레이어는 1개(0~1사이 값 1개로로 나오게 함)\n",
        "outputs = Activation('sigmoid')(net) # 거의 0, 거의 1에 가까운 0~1사이값으로 나오게 됨\n",
        "\n",
        "model = Model(inputs=inputs, outputs=outputs) # 모델에 input, output 정의\n",
        "\n",
        "# binary_crossentropy : 0이냐? 1이냐?\n",
        "model.compile(optimizer='adam', loss='binary_crossentropy', metrics=['acc'])\n",
        "\n",
        "model.summary()\n",
        "\n",
        "# 880,129 파라미터를 학습시킬 예정"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Model: \"model_1\"\n",
            "_________________________________________________________________\n",
            "Layer (type)                 Output Shape              Param #   \n",
            "=================================================================\n",
            "input_1 (InputLayer)         (None, 26, 34, 1)         0         \n",
            "_________________________________________________________________\n",
            "conv2d_1 (Conv2D)            (None, 26, 34, 32)        320       \n",
            "_________________________________________________________________\n",
            "max_pooling2d_1 (MaxPooling2 (None, 13, 17, 32)        0         \n",
            "_________________________________________________________________\n",
            "conv2d_2 (Conv2D)            (None, 13, 17, 64)        18496     \n",
            "_________________________________________________________________\n",
            "max_pooling2d_2 (MaxPooling2 (None, 6, 8, 64)          0         \n",
            "_________________________________________________________________\n",
            "conv2d_3 (Conv2D)            (None, 6, 8, 128)         73856     \n",
            "_________________________________________________________________\n",
            "max_pooling2d_3 (MaxPooling2 (None, 3, 4, 128)         0         \n",
            "_________________________________________________________________\n",
            "flatten_1 (Flatten)          (None, 1536)              0         \n",
            "_________________________________________________________________\n",
            "dense_1 (Dense)              (None, 512)               786944    \n",
            "_________________________________________________________________\n",
            "activation_1 (Activation)    (None, 512)               0         \n",
            "_________________________________________________________________\n",
            "dense_2 (Dense)              (None, 1)                 513       \n",
            "_________________________________________________________________\n",
            "activation_2 (Activation)    (None, 1)                 0         \n",
            "=================================================================\n",
            "Total params: 880,129\n",
            "Trainable params: 880,129\n",
            "Non-trainable params: 0\n",
            "_________________________________________________________________\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "CdGHwwMbHBBz"
      },
      "source": [
        "# Train"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "pUe1xRF0HBB0",
        "outputId": "a4d9a101-e7e3-45e0-9414-3667c914d525"
      },
      "source": [
        "start_time = datetime.datetime.now().strftime('%Y_%m_%d_%H_%M_%S')\n",
        "\n",
        "# generator로 데이터를 로드했을 경우 fit_generator로 학습시켜주는 것이 좋음\n",
        "# 반복학습 : 50번\n",
        "model.fit_generator( \n",
        "    train_generator, epochs=50, validation_data=val_generator,\n",
        "    callbacks=[\n",
        "        # 모델이 좋으면 저장해라\n",
        "        ModelCheckpoint('models/blink/%s.h5' % (start_time), monitor='val_acc', save_best_only=True, mode='max', verbose=1),\n",
        "        # 학습이 잘 안된다면 러닝레이트를 줄여라\n",
        "        # plateau : 정체기라는 의미\n",
        "        # 모델의 정확도가 향상되지 않는 경우 learning rate(lr) 줄여주는 클래스\n",
        "        ReduceLROnPlateau(monitor='val_acc', factor=0.2, patience=10, verbose=1, mode='auto', min_lr=1e-05)\n",
        "    ]\n",
        ")"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Epoch 1/50\n",
            "81/81 [==============================] - 2s 30ms/step - loss: 0.1784 - acc: 0.9377 - val_loss: 0.0563 - val_acc: 0.9722\n",
            "\n",
            "Epoch 00001: val_acc improved from -inf to 0.97222, saving model to models/2020_10_22_14_47_26.h5\n",
            "Epoch 2/50\n",
            "81/81 [==============================] - 2s 29ms/step - loss: 0.1352 - acc: 0.9520 - val_loss: 0.0445 - val_acc: 0.9792\n",
            "\n",
            "Epoch 00002: val_acc improved from 0.97222 to 0.97917, saving model to models/2020_10_22_14_47_26.h5\n",
            "Epoch 3/50\n",
            "81/81 [==============================] - 2s 30ms/step - loss: 0.1075 - acc: 0.9637 - val_loss: 0.0240 - val_acc: 0.9826\n",
            "\n",
            "Epoch 00003: val_acc improved from 0.97917 to 0.98264, saving model to models/2020_10_22_14_47_26.h5\n",
            "Epoch 4/50\n",
            "81/81 [==============================] - 3s 31ms/step - loss: 0.0903 - acc: 0.9671 - val_loss: 0.0176 - val_acc: 0.9792\n",
            "\n",
            "Epoch 00004: val_acc did not improve from 0.98264\n",
            "Epoch 5/50\n",
            "81/81 [==============================] - 2s 31ms/step - loss: 0.0884 - acc: 0.9710 - val_loss: 0.0130 - val_acc: 0.9826\n",
            "\n",
            "Epoch 00005: val_acc did not improve from 0.98264\n",
            "Epoch 6/50\n",
            "81/81 [==============================] - 2s 29ms/step - loss: 0.0689 - acc: 0.9753 - val_loss: 0.0193 - val_acc: 0.9861\n",
            "\n",
            "Epoch 00006: val_acc improved from 0.98264 to 0.98611, saving model to models/2020_10_22_14_47_26.h5\n",
            "Epoch 7/50\n",
            "81/81 [==============================] - 2s 29ms/step - loss: 0.0520 - acc: 0.9826 - val_loss: 0.0081 - val_acc: 0.9861\n",
            "\n",
            "Epoch 00007: val_acc did not improve from 0.98611\n",
            "Epoch 8/50\n",
            "81/81 [==============================] - 2s 29ms/step - loss: 0.0587 - acc: 0.9807 - val_loss: 0.0288 - val_acc: 0.9792\n",
            "\n",
            "Epoch 00008: val_acc did not improve from 0.98611\n",
            "Epoch 9/50\n",
            "81/81 [==============================] - 2s 30ms/step - loss: 0.0581 - acc: 0.9814 - val_loss: 0.0045 - val_acc: 0.9965\n",
            "\n",
            "Epoch 00009: val_acc improved from 0.98611 to 0.99653, saving model to models/2020_10_22_14_47_26.h5\n",
            "Epoch 10/50\n",
            "81/81 [==============================] - 3s 32ms/step - loss: 0.0542 - acc: 0.9799 - val_loss: 0.0033 - val_acc: 0.9931\n",
            "\n",
            "Epoch 00010: val_acc did not improve from 0.99653\n",
            "Epoch 11/50\n",
            "81/81 [==============================] - 3s 34ms/step - loss: 0.0469 - acc: 0.9830 - val_loss: 0.0066 - val_acc: 0.9861\n",
            "\n",
            "Epoch 00011: val_acc did not improve from 0.99653\n",
            "Epoch 12/50\n",
            "81/81 [==============================] - 3s 34ms/step - loss: 0.0372 - acc: 0.9849 - val_loss: 0.0030 - val_acc: 0.9896\n",
            "\n",
            "Epoch 00012: val_acc did not improve from 0.99653\n",
            "Epoch 13/50\n",
            "81/81 [==============================] - 3s 33ms/step - loss: 0.0444 - acc: 0.9841 - val_loss: 0.0063 - val_acc: 0.9861\n",
            "\n",
            "Epoch 00013: val_acc did not improve from 0.99653\n",
            "Epoch 14/50\n",
            "81/81 [==============================] - 3s 33ms/step - loss: 0.0302 - acc: 0.9899 - val_loss: 0.0038 - val_acc: 0.9896\n",
            "\n",
            "Epoch 00014: val_acc did not improve from 0.99653\n",
            "Epoch 15/50\n",
            "81/81 [==============================] - 3s 33ms/step - loss: 0.0293 - acc: 0.9876 - val_loss: 0.0016 - val_acc: 0.9931\n",
            "\n",
            "Epoch 00015: val_acc did not improve from 0.99653\n",
            "Epoch 16/50\n",
            "81/81 [==============================] - 3s 33ms/step - loss: 0.0387 - acc: 0.9888 - val_loss: 0.0079 - val_acc: 0.9896\n",
            "\n",
            "Epoch 00016: val_acc did not improve from 0.99653\n",
            "Epoch 17/50\n",
            "81/81 [==============================] - 3s 34ms/step - loss: 0.0409 - acc: 0.9861 - val_loss: 0.0196 - val_acc: 0.9826\n",
            "\n",
            "Epoch 00017: val_acc did not improve from 0.99653\n",
            "Epoch 18/50\n",
            "81/81 [==============================] - 3s 33ms/step - loss: 0.0318 - acc: 0.9884 - val_loss: 0.0164 - val_acc: 0.9896\n",
            "\n",
            "Epoch 00018: val_acc did not improve from 0.99653\n",
            "Epoch 19/50\n",
            "81/81 [==============================] - 3s 33ms/step - loss: 0.0254 - acc: 0.9930 - val_loss: 0.0042 - val_acc: 0.9896\n",
            "\n",
            "Epoch 00019: val_acc did not improve from 0.99653\n",
            "\n",
            "Epoch 00019: ReduceLROnPlateau reducing learning rate to 0.00020000000949949026.\n",
            "Epoch 20/50\n",
            "81/81 [==============================] - 3s 35ms/step - loss: 0.0182 - acc: 0.9938 - val_loss: 0.0060 - val_acc: 0.9896\n",
            "\n",
            "Epoch 00020: val_acc did not improve from 0.99653\n",
            "Epoch 21/50\n",
            "81/81 [==============================] - 3s 36ms/step - loss: 0.0099 - acc: 0.9981 - val_loss: 0.0028 - val_acc: 0.9896\n",
            "\n",
            "Epoch 00021: val_acc did not improve from 0.99653\n",
            "Epoch 22/50\n",
            "81/81 [==============================] - 3s 36ms/step - loss: 0.0166 - acc: 0.9950 - val_loss: 0.0023 - val_acc: 0.9896\n",
            "\n",
            "Epoch 00022: val_acc did not improve from 0.99653\n",
            "Epoch 23/50\n",
            "81/81 [==============================] - 3s 36ms/step - loss: 0.0132 - acc: 0.9954 - val_loss: 0.0027 - val_acc: 0.9931\n",
            "\n",
            "Epoch 00023: val_acc did not improve from 0.99653\n",
            "Epoch 24/50\n",
            "81/81 [==============================] - 3s 34ms/step - loss: 0.0103 - acc: 0.9969 - val_loss: 0.0014 - val_acc: 0.9931\n",
            "\n",
            "Epoch 00024: val_acc did not improve from 0.99653\n",
            "Epoch 25/50\n",
            "81/81 [==============================] - 3s 34ms/step - loss: 0.0111 - acc: 0.9961 - val_loss: 0.0014 - val_acc: 0.9931\n",
            "\n",
            "Epoch 00025: val_acc did not improve from 0.99653\n",
            "Epoch 26/50\n",
            "81/81 [==============================] - 3s 34ms/step - loss: 0.0095 - acc: 0.9969 - val_loss: 8.3772e-04 - val_acc: 0.9931\n",
            "\n",
            "Epoch 00026: val_acc did not improve from 0.99653\n",
            "Epoch 27/50\n",
            "81/81 [==============================] - 3s 34ms/step - loss: 0.0147 - acc: 0.9942 - val_loss: 0.0040 - val_acc: 0.9931\n",
            "\n",
            "Epoch 00027: val_acc did not improve from 0.99653\n",
            "Epoch 28/50\n",
            "81/81 [==============================] - 3s 36ms/step - loss: 0.0105 - acc: 0.9965 - val_loss: 0.0012 - val_acc: 0.9931\n",
            "\n",
            "Epoch 00028: val_acc did not improve from 0.99653\n",
            "Epoch 29/50\n",
            "81/81 [==============================] - 3s 41ms/step - loss: 0.0073 - acc: 0.9985 - val_loss: 9.4678e-04 - val_acc: 0.9931\n",
            "\n",
            "Epoch 00029: val_acc did not improve from 0.99653\n",
            "\n",
            "Epoch 00029: ReduceLROnPlateau reducing learning rate to 4.0000001899898055e-05.\n",
            "Epoch 30/50\n",
            "81/81 [==============================] - 3s 39ms/step - loss: 0.0097 - acc: 0.9961 - val_loss: 0.0016 - val_acc: 0.9931\n",
            "\n",
            "Epoch 00030: val_acc did not improve from 0.99653\n",
            "Epoch 31/50\n",
            "81/81 [==============================] - 3s 41ms/step - loss: 0.0083 - acc: 0.9973 - val_loss: 0.0014 - val_acc: 0.9931\n",
            "\n",
            "Epoch 00031: val_acc did not improve from 0.99653\n",
            "Epoch 32/50\n",
            "81/81 [==============================] - 3s 39ms/step - loss: 0.0107 - acc: 0.9957 - val_loss: 0.0011 - val_acc: 0.9931\n",
            "\n",
            "Epoch 00032: val_acc did not improve from 0.99653\n",
            "Epoch 33/50\n",
            "81/81 [==============================] - 3s 38ms/step - loss: 0.0084 - acc: 0.9985 - val_loss: 0.0012 - val_acc: 0.9931\n",
            "\n",
            "Epoch 00033: val_acc did not improve from 0.99653\n",
            "Epoch 34/50\n",
            "81/81 [==============================] - 3s 38ms/step - loss: 0.0120 - acc: 0.9965 - val_loss: 0.0019 - val_acc: 0.9931\n",
            "\n",
            "Epoch 00034: val_acc did not improve from 0.99653\n",
            "Epoch 35/50\n",
            "81/81 [==============================] - 3s 41ms/step - loss: 0.0080 - acc: 0.9977 - val_loss: 0.0019 - val_acc: 0.9931\n",
            "\n",
            "Epoch 00035: val_acc did not improve from 0.99653\n",
            "Epoch 36/50\n",
            "81/81 [==============================] - 3s 40ms/step - loss: 0.0080 - acc: 0.9973 - val_loss: 0.0018 - val_acc: 0.9965\n",
            "\n",
            "Epoch 00036: val_acc did not improve from 0.99653\n",
            "Epoch 37/50\n",
            "81/81 [==============================] - 3s 36ms/step - loss: 0.0073 - acc: 0.9977 - val_loss: 0.0015 - val_acc: 0.9931\n",
            "\n",
            "Epoch 00037: val_acc did not improve from 0.99653\n",
            "Epoch 38/50\n",
            "81/81 [==============================] - 3s 41ms/step - loss: 0.0057 - acc: 0.9981 - val_loss: 0.0018 - val_acc: 0.9931\n",
            "\n",
            "Epoch 00038: val_acc did not improve from 0.99653\n",
            "Epoch 39/50\n",
            "81/81 [==============================] - 4s 44ms/step - loss: 0.0074 - acc: 0.9981 - val_loss: 0.0012 - val_acc: 0.9931\n",
            "\n",
            "Epoch 00039: val_acc did not improve from 0.99653\n",
            "\n",
            "Epoch 00039: ReduceLROnPlateau reducing learning rate to 1e-05.\n",
            "Epoch 40/50\n",
            "81/81 [==============================] - 4s 43ms/step - loss: 0.0074 - acc: 0.9973 - val_loss: 0.0015 - val_acc: 0.9931\n",
            "\n",
            "Epoch 00040: val_acc did not improve from 0.99653\n",
            "Epoch 41/50\n",
            "81/81 [==============================] - 4s 46ms/step - loss: 0.0063 - acc: 0.9977 - val_loss: 0.0017 - val_acc: 0.9931\n",
            "\n",
            "Epoch 00041: val_acc did not improve from 0.99653\n",
            "Epoch 42/50\n",
            "81/81 [==============================] - 4s 46ms/step - loss: 0.0059 - acc: 0.9985 - val_loss: 0.0015 - val_acc: 0.9931\n",
            "\n",
            "Epoch 00042: val_acc did not improve from 0.99653\n",
            "Epoch 43/50\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "81/81 [==============================] - 4s 43ms/step - loss: 0.0069 - acc: 0.9988 - val_loss: 0.0016 - val_acc: 0.9931\n",
            "\n",
            "Epoch 00043: val_acc did not improve from 0.99653\n",
            "Epoch 44/50\n",
            "81/81 [==============================] - 3s 35ms/step - loss: 0.0051 - acc: 0.9988 - val_loss: 0.0015 - val_acc: 0.9931\n",
            "\n",
            "Epoch 00044: val_acc did not improve from 0.99653\n",
            "Epoch 45/50\n",
            "81/81 [==============================] - 3s 34ms/step - loss: 0.0069 - acc: 0.9977 - val_loss: 0.0019 - val_acc: 0.9931\n",
            "\n",
            "Epoch 00045: val_acc did not improve from 0.99653\n",
            "Epoch 46/50\n",
            "81/81 [==============================] - 3s 35ms/step - loss: 0.0051 - acc: 0.9977 - val_loss: 0.0017 - val_acc: 0.9931\n",
            "\n",
            "Epoch 00046: val_acc did not improve from 0.99653\n",
            "Epoch 47/50\n",
            "81/81 [==============================] - 3s 35ms/step - loss: 0.0059 - acc: 0.9985 - val_loss: 0.0016 - val_acc: 0.9931\n",
            "\n",
            "Epoch 00047: val_acc did not improve from 0.99653\n",
            "Epoch 48/50\n",
            "81/81 [==============================] - 3s 37ms/step - loss: 0.0082 - acc: 0.9977 - val_loss: 0.0015 - val_acc: 0.9931\n",
            "\n",
            "Epoch 00048: val_acc did not improve from 0.99653\n",
            "Epoch 49/50\n",
            "81/81 [==============================] - 3s 36ms/step - loss: 0.0071 - acc: 0.9977 - val_loss: 0.0014 - val_acc: 0.9931\n",
            "\n",
            "Epoch 00049: val_acc did not improve from 0.99653\n",
            "Epoch 50/50\n",
            "81/81 [==============================] - 3s 36ms/step - loss: 0.0086 - acc: 0.9977 - val_loss: 0.0016 - val_acc: 0.9931\n",
            "\n",
            "Epoch 00050: val_acc did not improve from 0.99653\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "<keras.callbacks.callbacks.History at 0x197b694f708>"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 11
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "7IN2fTphHBB7"
      },
      "source": [
        "# Confusion Matrix\n",
        "- 모델이 얼마나 잘 학습했는지 확인"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "yuXNExxTHBB8",
        "outputId": "fdebe0a1-ea1e-4b93-eb85-ce715351870b"
      },
      "source": [
        "from sklearn.metrics import accuracy_score, confusion_matrix # 정확도점수, 오차행렬\n",
        "import seaborn as sns\n",
        "\n",
        "model = load_model('models/blink/%s.h5' % (start_time)) # 모델 로드\n",
        "\n",
        "y_pred = model.predict(x_val/255.) # 검증 이미지데이터로 예측, 이미지 0~1로 바꿔서\n",
        "y_pred_logical = (y_pred > 0.5).astype(np.int) # 0.5미만(눈 감음) / 나머지(눈뜸), True.. -> int형 변환 -> 0 or 1\n",
        "\n",
        "print ('test acc: %s' % accuracy_score(y_val, y_pred_logical))\n",
        "cm = confusion_matrix(y_val, y_pred_logical)\n",
        "sns.heatmap(cm, annot=True)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "test acc: 0.9965277777777778\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "<AxesSubplot:>"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 12
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 432x288 with 2 Axes>"
            ]
          },
          "metadata": {
            "tags": []
          }
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "zU7wV8rTHBCD"
      },
      "source": [
        "# Distribution of Prediction"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "K2OT8yGoHBCE",
        "outputId": "5c9a97f1-47b6-4b20-909a-f2cd4562dee3"
      },
      "source": [
        "ax = sns.histplot(y_pred, kde=False)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": []
          }
        }
      ]
    }
  ]
}