{"metadata":{"kernelspec":{"language":"python","display_name":"Python 3","name":"python3"},"language_info":{"pygments_lexer":"ipython3","nbconvert_exporter":"python","version":"3.6.4","file_extension":".py","codemirror_mode":{"name":"ipython","version":3},"name":"python","mimetype":"text/x-python"}},"nbformat_minor":4,"nbformat":4,"cells":[{"cell_type":"code","source":"import numpy as np\nimport pandas as pd \nimport tensorflow as tf\n\nimport matplotlib.pyplot as plt\nimport random\nimport os\nprint(os.listdir(\"../input/dogs-vs-cats/\"))","metadata":{"_uuid":"d629ff2d2480ee46fbb7e2d37f6b5fab8052498a","_cell_guid":"79c7e3d0-c299-4dcb-8224-4455121ee9b0","scrolled":true,"execution":{"iopub.status.busy":"2022-10-24T17:53:44.791881Z","iopub.execute_input":"2022-10-24T17:53:44.792206Z","iopub.status.idle":"2022-10-24T17:53:44.798230Z","shell.execute_reply.started":"2022-10-24T17:53:44.792173Z","shell.execute_reply":"2022-10-24T17:53:44.797277Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"%%time\nimport zipfile\nwith zipfile.ZipFile(\"../input/dogs-vs-cats/train.zip\",'r') as z:\n    z.extractall(\".\")\nwith zipfile.ZipFile(\"../input/dogs-vs-cats/test1.zip\",'r') as z:\n    z.extractall(\".\")","metadata":{"scrolled":true,"execution":{"iopub.status.busy":"2022-10-24T17:53:44.804148Z","iopub.execute_input":"2022-10-24T17:53:44.804393Z","iopub.status.idle":"2022-10-24T17:53:57.258992Z","shell.execute_reply.started":"2022-10-24T17:53:44.804367Z","shell.execute_reply":"2022-10-24T17:53:57.258238Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"FAST_RUN = True\nIMAGE_WIDTH=128\nIMAGE_HEIGHT=128\nIMAGE_SIZE=(IMAGE_WIDTH, IMAGE_HEIGHT)\nIMAGE_CHANNELS=3","metadata":{"scrolled":true,"execution":{"iopub.status.busy":"2022-10-24T17:53:57.260688Z","iopub.execute_input":"2022-10-24T17:53:57.260940Z","iopub.status.idle":"2022-10-24T17:53:57.264960Z","shell.execute_reply.started":"2022-10-24T17:53:57.260914Z","shell.execute_reply":"2022-10-24T17:53:57.264149Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"## Предварительная обработка данных","metadata":{}},{"cell_type":"code","source":"from keras.preprocessing.image import ImageDataGenerator, load_img\nfrom keras.utils import to_categorical\nfrom sklearn.model_selection import train_test_split","metadata":{"scrolled":true,"execution":{"iopub.status.busy":"2022-10-24T17:53:57.266151Z","iopub.execute_input":"2022-10-24T17:53:57.266671Z","iopub.status.idle":"2022-10-24T17:53:58.580959Z","shell.execute_reply.started":"2022-10-24T17:53:57.266635Z","shell.execute_reply":"2022-10-24T17:53:58.580097Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"filenames = os.listdir(\"/kaggle/working/train\")\ncategories = []\nfor filename in filenames:\n    category = filename.split('.')[0]\n    categories.append(category)\n\nfiles_df = pd.DataFrame({\n    'filename': filenames,\n    'category': categories\n})","metadata":{"scrolled":true,"execution":{"iopub.status.busy":"2022-10-24T17:53:58.582335Z","iopub.execute_input":"2022-10-24T17:53:58.582843Z","iopub.status.idle":"2022-10-24T17:53:58.619049Z","shell.execute_reply.started":"2022-10-24T17:53:58.582804Z","shell.execute_reply":"2022-10-24T17:53:58.618391Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"train_df, validate_df = train_test_split(files_df, \n                                         test_size=0.20,                                         \n                                         random_state=0)\ntrain_df = train_df.reset_index(drop=True)\nvalidate_df = validate_df.reset_index(drop=True)","metadata":{"execution":{"iopub.status.busy":"2022-10-24T17:53:58.622345Z","iopub.execute_input":"2022-10-24T17:53:58.622641Z","iopub.status.idle":"2022-10-24T17:53:58.632782Z","shell.execute_reply.started":"2022-10-24T17:53:58.622615Z","shell.execute_reply":"2022-10-24T17:53:58.632095Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"total_train = train_df.shape[0]\ntotal_validate = validate_df.shape[0]\nbatch_size=15","metadata":{"execution":{"iopub.status.busy":"2022-10-24T17:53:58.635308Z","iopub.execute_input":"2022-10-24T17:53:58.635675Z","iopub.status.idle":"2022-10-24T17:53:58.639975Z","shell.execute_reply.started":"2022-10-24T17:53:58.635640Z","shell.execute_reply":"2022-10-24T17:53:58.639110Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"## Подготовка данных изображения","metadata":{}},{"cell_type":"code","source":"train_datagen = ImageDataGenerator(\n    rotation_range=15,\n    rescale=1./255,\n    shear_range=0.1,\n    zoom_range=0.2,\n    horizontal_flip=True,\n    width_shift_range=0.1,\n    height_shift_range=0.1\n)\n\ntrain_generator = train_datagen.flow_from_dataframe(\n    train_df, \n    \"/kaggle/working/train/\", \n    x_col='filename',\n    y_col='category',\n    target_size=IMAGE_SIZE,\n    class_mode='categorical',\n    batch_size=batch_size\n)","metadata":{"execution":{"iopub.status.busy":"2022-10-24T17:53:58.641210Z","iopub.execute_input":"2022-10-24T17:53:58.641759Z","iopub.status.idle":"2022-10-24T17:53:58.850399Z","shell.execute_reply.started":"2022-10-24T17:53:58.641723Z","shell.execute_reply":"2022-10-24T17:53:58.849366Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"validation_datagen = ImageDataGenerator(rescale=1./255)\n\nvalidation_generator = validation_datagen.flow_from_dataframe(validate_df, \n    \"/kaggle/working/train/\", \n    x_col='filename',\n    y_col='category',\n    target_size=IMAGE_SIZE,\n    class_mode='categorical',\n    batch_size=batch_size\n)","metadata":{"execution":{"iopub.status.busy":"2022-10-24T17:53:58.851556Z","iopub.execute_input":"2022-10-24T17:53:58.851906Z","iopub.status.idle":"2022-10-24T17:53:58.908390Z","shell.execute_reply.started":"2022-10-24T17:53:58.851866Z","shell.execute_reply":"2022-10-24T17:53:58.907602Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"## Создание модели CNN","metadata":{}},{"cell_type":"code","source":"from keras.models import Sequential\nfrom keras.applications import VGG16\nfrom keras.layers import Conv2D, MaxPooling2D, Dropout, Flatten, Dense, Activation, BatchNormalization, GlobalAveragePooling2D","metadata":{"execution":{"iopub.status.busy":"2022-10-24T17:53:58.910036Z","iopub.execute_input":"2022-10-24T17:53:58.910593Z","iopub.status.idle":"2022-10-24T17:53:58.917407Z","shell.execute_reply.started":"2022-10-24T17:53:58.910556Z","shell.execute_reply":"2022-10-24T17:53:58.916574Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"def create_model():        \n    \n    model = Sequential()\n    model.add(Conv2D(32, (3, 3), \n                     activation='relu',\n                     input_shape=(IMAGE_WIDTH, IMAGE_HEIGHT, IMAGE_CHANNELS)))\n    \n    model.add(BatchNormalization())\n    model.add(MaxPooling2D(pool_size=(2, 2)))\n    model.add(Dropout(0.25))\n\n    model.add(Conv2D(60, (3, 3), activation='relu',strides=2))\n    model.add(BatchNormalization())\n    model.add(MaxPooling2D(pool_size=(2, 2)))\n    model.add(Dropout(0.25))\n\n    model.add(Flatten())\n    \n    model.add(Dense(100, activation='relu'))\n    model.add(BatchNormalization())\n    model.add(Dropout(0.5))\n    \n    model.add(Dense(2, activation='sigmoid'))\n    \n    model.compile(loss='categorical_crossentropy', \n                  optimizer='adam',\n                  metrics=['accuracy'])\n\n    #model.summary()\n    return model","metadata":{"execution":{"iopub.status.busy":"2022-10-24T17:53:58.918804Z","iopub.execute_input":"2022-10-24T17:53:58.919135Z","iopub.status.idle":"2022-10-24T17:53:58.931392Z","shell.execute_reply.started":"2022-10-24T17:53:58.919101Z","shell.execute_reply":"2022-10-24T17:53:58.930563Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"def createVGG16model():\n    pretrained_model = VGG16(weights='imagenet', include_top=False ,input_shape=(IMAGE_WIDTH, IMAGE_HEIGHT, IMAGE_CHANNELS))\n    pretrained_model.trainable = False \n    \n    model = Sequential([pretrained_model,\n                        GlobalAveragePooling2D(),\n                        Dense(2, activation='softmax')])\n\n    model.compile(optimizer='adam',\n                    loss = 'categorical_crossentropy',\n                    metrics=['accuracy'])\n\n    return model","metadata":{"execution":{"iopub.status.busy":"2022-10-24T17:53:58.932786Z","iopub.execute_input":"2022-10-24T17:53:58.933126Z","iopub.status.idle":"2022-10-24T17:53:58.941535Z","shell.execute_reply.started":"2022-10-24T17:53:58.933090Z","shell.execute_reply":"2022-10-24T17:53:58.940637Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"### Обратные вызовы","metadata":{}},{"cell_type":"code","source":"from keras.callbacks import EarlyStopping, ReduceLROnPlateau\nearlystop = EarlyStopping(patience=10)\n\nlearning_rate_reduction = ReduceLROnPlateau(monitor='val_acc', \n                                            patience=2, \n                                            verbose=1, \n                                            factor=0.5, \n                                            min_lr=0.00001)\n\ncallbacks = [earlystop, learning_rate_reduction]","metadata":{"execution":{"iopub.status.busy":"2022-10-24T17:53:58.943379Z","iopub.execute_input":"2022-10-24T17:53:58.944046Z","iopub.status.idle":"2022-10-24T17:53:58.957011Z","shell.execute_reply.started":"2022-10-24T17:53:58.944010Z","shell.execute_reply":"2022-10-24T17:53:58.956246Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"## Подгонка модели","metadata":{}},{"cell_type":"code","source":"%%time\nif FAST_RUN :\n    epochs = 3\nelse:\n    epochs = 30\n\n#with strategy.scope():\nwith tf.device(\"/gpu:0\"):   \n    #model = createVGG16model()\n    model = create_model()\n    history = model.fit(train_generator, \n                        epochs=epochs,\n                        validation_data=validation_generator,\n                        validation_steps=1,\n                        steps_per_epoch=total_train//batch_size,\n                        callbacks=callbacks)\n\nmodel.save_weights(\"model.h5\")","metadata":{"execution":{"iopub.status.busy":"2022-10-24T17:53:58.960105Z","iopub.execute_input":"2022-10-24T17:53:58.960600Z","iopub.status.idle":"2022-10-24T18:00:09.135233Z","shell.execute_reply.started":"2022-10-24T17:53:58.960562Z","shell.execute_reply":"2022-10-24T18:00:09.134273Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"## Выходной график","metadata":{}},{"cell_type":"code","source":"import seaborn as sns\nhis_dict = history.history\nfig = plt.figure(figsize=(12, 15))\nx_range = range(len(history.history['loss']))\nsns.set_style('darkgrid')\n\nfig.add_subplot(2,1,1)\nsns.lineplot(x=x_range , y=his_dict[\"val_loss\"],label='Validation Loss')\nsns.lineplot(x=x_range , y=his_dict[\"loss\"],label='Training Loss')\n\nfig.add_subplot(2,1,2)\nsns.lineplot(x=x_range , y=his_dict[\"val_accuracy\"],label='Validation Accuracy')\nsns.lineplot(x=x_range , y=his_dict[\"accuracy\"],label='Training Accuracy')","metadata":{"execution":{"iopub.status.busy":"2022-10-24T18:00:09.136884Z","iopub.execute_input":"2022-10-24T18:00:09.137324Z","iopub.status.idle":"2022-10-24T18:00:09.859190Z","shell.execute_reply.started":"2022-10-24T18:00:09.137282Z","shell.execute_reply":"2022-10-24T18:00:09.858213Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"## Предсказания","metadata":{}},{"cell_type":"code","source":"test_filenames = os.listdir(\"/kaggle/working/test1\")\ntest_df = pd.DataFrame({\n    'filename': test_filenames\n})\nidlist=[]\nfor file in test_df[\"filename\"]:\n    idlist.append(file.split(\".\")[0])\n\n#test_df['ids'] = alist\nnb_samples = test_df.shape[0]","metadata":{"execution":{"iopub.status.busy":"2022-10-24T18:00:09.860389Z","iopub.execute_input":"2022-10-24T18:00:09.860738Z","iopub.status.idle":"2022-10-24T18:00:09.884258Z","shell.execute_reply.started":"2022-10-24T18:00:09.860685Z","shell.execute_reply":"2022-10-24T18:00:09.883614Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"test_gen = ImageDataGenerator(rescale=1./255)\ntest_generator = test_gen.flow_from_dataframe(\n    test_df, \n    \"/kaggle/working/test1\", \n    x_col='filename',\n    y_col=None,\n    class_mode=None,\n    target_size=IMAGE_SIZE,\n    batch_size=batch_size,\n    shuffle=False\n)","metadata":{"execution":{"iopub.status.busy":"2022-10-24T18:00:09.885452Z","iopub.execute_input":"2022-10-24T18:00:09.885806Z","iopub.status.idle":"2022-10-24T18:00:09.990973Z","shell.execute_reply.started":"2022-10-24T18:00:09.885778Z","shell.execute_reply":"2022-10-24T18:00:09.990095Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"predict = model.predict_generator(test_generator, steps=np.ceil(nb_samples/batch_size))","metadata":{"execution":{"iopub.status.busy":"2022-10-24T18:00:09.992313Z","iopub.execute_input":"2022-10-24T18:00:09.992729Z","iopub.status.idle":"2022-10-24T18:00:39.447378Z","shell.execute_reply.started":"2022-10-24T18:00:09.992672Z","shell.execute_reply":"2022-10-24T18:00:39.446469Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"test_df['category'] = np.argmax(predict, axis=-1)","metadata":{"execution":{"iopub.status.busy":"2022-10-24T18:00:39.451312Z","iopub.execute_input":"2022-10-24T18:00:39.451588Z","iopub.status.idle":"2022-10-24T18:00:39.456977Z","shell.execute_reply.started":"2022-10-24T18:00:39.451560Z","shell.execute_reply":"2022-10-24T18:00:39.456165Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"label_map = dict((v,k) for k,v in train_generator.class_indices.items())\ntest_df['category'] = test_df['category'].replace(label_map)","metadata":{"execution":{"iopub.status.busy":"2022-10-24T18:00:39.458631Z","iopub.execute_input":"2022-10-24T18:00:39.459311Z","iopub.status.idle":"2022-10-24T18:00:39.470948Z","shell.execute_reply.started":"2022-10-24T18:00:39.459272Z","shell.execute_reply":"2022-10-24T18:00:39.468830Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"test_df['category'] = test_df['category'].replace({ 'dog': 1, 'cat': 0 })","metadata":{"execution":{"iopub.status.busy":"2022-10-24T18:00:39.473299Z","iopub.execute_input":"2022-10-24T18:00:39.474411Z","iopub.status.idle":"2022-10-24T18:00:39.493250Z","shell.execute_reply.started":"2022-10-24T18:00:39.474362Z","shell.execute_reply":"2022-10-24T18:00:39.492384Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"## представление","metadata":{}},{"cell_type":"markdown","source":"","metadata":{}},{"cell_type":"code","source":"submission_df = test_df.copy()\nsubmission_df['id'] = submission_df['filename'].str.split('.').str[0]\nsubmission_df['label'] = submission_df['category']\nsubmission_df.drop(['filename', 'category'], axis=1, inplace=True)\nsubmission_df.to_csv('submission.csv', index=False)","metadata":{"execution":{"iopub.status.busy":"2022-10-24T18:12:32.689170Z","iopub.execute_input":"2022-10-24T18:12:32.689628Z","iopub.status.idle":"2022-10-24T18:12:32.731922Z","shell.execute_reply.started":"2022-10-24T18:12:32.689578Z","shell.execute_reply":"2022-10-24T18:12:32.730960Z"},"trusted":true},"execution_count":null,"outputs":[]}]}