{"metadata":{"kernelspec":{"language":"python","display_name":"Python 3","name":"python3"},"language_info":{"name":"python","version":"3.10.13","mimetype":"text/x-python","codemirror_mode":{"name":"ipython","version":3},"pygments_lexer":"ipython3","nbconvert_exporter":"python","file_extension":".py"},"kaggle":{"accelerator":"tpu1vmV38","dataSources":[{"sourceId":21154,"databundleVersionId":1243559,"sourceType":"competition"}],"dockerImageVersionId":30626,"isInternetEnabled":true,"language":"python","sourceType":"notebook","isGpuEnabled":false}},"nbformat_minor":4,"nbformat":4,"cells":[{"cell_type":"code","source":"!pip install -q efficientnet\n","metadata":{"_uuid":"8f2839f25d086af736a60e9eeb907d3b93b6e0e5","_cell_guid":"b1076dfc-b9ad-4769-8c92-a6c4dae69d19","execution":{"iopub.status.busy":"2023-12-22T09:02:10.723071Z","iopub.execute_input":"2023-12-22T09:02:10.723884Z","iopub.status.idle":"2023-12-22T09:02:17.498043Z","shell.execute_reply.started":"2023-12-22T09:02:10.723846Z","shell.execute_reply":"2023-12-22T09:02:17.497177Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"# **Загрузка, настройка**","metadata":{}},{"cell_type":"code","source":"#предварительно обученные модели\nfrom tensorflow.keras.applications.vgg16 import VGG16\nfrom tensorflow.keras.applications.vgg19 import VGG19\nfrom tensorflow.keras.applications.inception_v3 import InceptionV3\nfrom tensorflow.keras.applications.inception_resnet_v2 import InceptionResNetV2\nfrom tensorflow.keras.applications.densenet import DenseNet121, DenseNet169, DenseNet201 \nfrom tensorflow.keras.applications.xception import Xception\nfrom tensorflow.keras.applications.resnet50 import ResNet50\nfrom tensorflow.keras.applications.resnet_v2 import ResNet50V2, ResNet101V2, ResNet152V2\nfrom tensorflow.keras.applications.nasnet import NASNetLarge\nfrom efficientnet.tfkeras import EfficientNetB7, EfficientNetL2, EfficientNetB0, EfficientNetB1","metadata":{"execution":{"iopub.status.busy":"2023-12-22T09:02:34.955680Z","iopub.execute_input":"2023-12-22T09:02:34.956066Z","iopub.status.idle":"2023-12-22T09:02:49.409905Z","shell.execute_reply.started":"2023-12-22T09:02:34.956030Z","shell.execute_reply":"2023-12-22T09:02:49.409159Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"import tensorflow as tf\nfrom tensorflow.keras.models import Model\nfrom tensorflow.keras import layers\nfrom tensorflow.keras.layers import Dense, Flatten\nfrom tensorflow.keras.layers import Dropout, BatchNormalization, GaussianDropout\nfrom tensorflow.keras.layers import GlobalAveragePooling2D\nfrom tensorflow.keras.callbacks import ModelCheckpoint\n#  библиотека для работы с наборами данных на Kaggle\nfrom kaggle_datasets import KaggleDatasets\nimport re\nimport numpy as np\nimport random\nimport matplotlib.pyplot as plt\n%matplotlib inline \nprint(\"Tensorflow version \" + tf.__version__)","metadata":{"execution":{"iopub.status.busy":"2023-12-22T09:02:57.260732Z","iopub.execute_input":"2023-12-22T09:02:57.261294Z","iopub.status.idle":"2023-12-22T09:02:58.146884Z","shell.execute_reply.started":"2023-12-22T09:02:57.261261Z","shell.execute_reply":"2023-12-22T09:02:58.146113Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"AUTO = tf.data.experimental.AUTOTUNE\n# Обнаружение оборудования, возврат соответствующей стратегии распространения: TPU, GPU, CPU\ntry:\n    tpu = tf.distribute.cluster_resolver.TPUClusterResolver()  # Обнаружение TPU. Параметры среды не требуются, если задана переменная среды TPU_NAME. На Kaggle это всегда так.\n    print('Running on TPU ', tpu.master())\nexcept ValueError:\n    tpu = None\n\nif tpu:\n    tf.config.experimental_connect_to_cluster(tpu)\n    tf.tpu.experimental.initialize_tpu_system(tpu)\n    strategy = tf.distribute.experimental.TPUStrategy(tpu)\nelse:\n    strategy = tf.distribute.get_strategy() # стратегия распространения по умолчанию в Tensorflow. Работает на CPU и одном GPU.\nprint(\"REPLICAS: \", strategy.num_replicas_in_sync)","metadata":{"execution":{"iopub.status.busy":"2023-12-22T09:03:15.217663Z","iopub.execute_input":"2023-12-22T09:03:15.218043Z","iopub.status.idle":"2023-12-22T09:03:22.687703Z","shell.execute_reply.started":"2023-12-22T09:03:15.218013Z","shell.execute_reply":"2023-12-22T09:03:22.686846Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"from kaggle_datasets import KaggleDatasets\n\nGCS_DS_PATH = KaggleDatasets().get_gcs_path('tpu-getting-started') #получаем путь к наборам данных\nprint(GCS_DS_PATH)","metadata":{"execution":{"iopub.status.busy":"2023-12-22T10:30:55.294824Z","iopub.execute_input":"2023-12-22T10:30:55.295844Z","iopub.status.idle":"2023-12-22T10:30:55.301235Z","shell.execute_reply.started":"2023-12-22T10:30:55.295773Z","shell.execute_reply":"2023-12-22T10:30:55.300186Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"IMAGE_SIZE = [512, 512] # при таком размере графическому процессору не хватит памяти. Используйте TPU\nEPOCHS = 25\nBATCH_SIZE = 16 * strategy.num_replicas_in_sync\n\nGCS_PATH_SELECT = { # available image sizes\n    192: GCS_DS_PATH + '/tfrecords-jpeg-192x192',\n    224: GCS_DS_PATH + '/tfrecords-jpeg-224x224',\n    331: GCS_DS_PATH + '/tfrecords-jpeg-331x331',\n    512: GCS_DS_PATH + '/tfrecords-jpeg-512x512'\n}\nGCS_PATH = GCS_PATH_SELECT[IMAGE_SIZE[0]]\n\nTRAINING_FILENAMES = tf.io.gfile.glob(GCS_PATH + '/train/*.tfrec')\nVALIDATION_FILENAMES = tf.io.gfile.glob(GCS_PATH + '/val/*.tfrec')\nTEST_FILENAMES = tf.io.gfile.glob(GCS_PATH + '/test/*.tfrec')\n\nSEED = 2020","metadata":{"execution":{"iopub.status.busy":"2023-12-22T10:30:57.578936Z","iopub.execute_input":"2023-12-22T10:30:57.580047Z","iopub.status.idle":"2023-12-22T10:30:57.607192Z","shell.execute_reply.started":"2023-12-22T10:30:57.579997Z","shell.execute_reply":"2023-12-22T10:30:57.605433Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"# **Загружаем данные**\nЭти данные загружаются из Kaggle и автоматически сегментируются для максимального распараллеливания.","metadata":{}},{"cell_type":"code","source":"def decode_image(image_data):\n    \"\"\"Декодирует изображение в vyjujvthye. vfnhbwe (тензор)\n    Нормализует данные и преобразовывает изображения к указанному размеру\"\"\"\n    image = tf.image.decode_jpeg(image_data, channels=3) # Декодирование изображения в формате JPEG в тензор uint8.\n    image = tf.cast(image, tf.float32) / 255.0  # преобразовать изображение в плавающее в диапазоне [0, 1]\n    image = tf.reshape(image, [*IMAGE_SIZE, 3]) # явный размер, необходимый для TPU\n#     image = tf.keras.applications.inception_resnet_v2.preprocess_input(image)\n    return image\n\ndef read_labeled_tfrecord(example):\n    LABELED_TFREC_FORMAT = {\n        \"image\": tf.io.FixedLenFeature([], tf.string), # tf.string означает байтовую строку\n        \"class\": tf.io.FixedLenFeature([], tf.int64),  # [] означает отдельный элемент\n    }\n    example = tf.io.parse_single_example(example, LABELED_TFREC_FORMAT) # парсим отдельный пример в указанном формате\n    image = decode_image(example['image']) # преобразуем изображение к нужному нам формату\n    label = tf.cast(example['class'], tf.int32)\n    return image, label # возвращает набор данных пар (изображение, метка)\n\ndef read_unlabeled_tfrecord(example):\n    UNLABELED_TFREC_FORMAT = {\n        \"image\": tf.io.FixedLenFeature([], tf.string), # tf.string означает байтовую строку\n        \"id\": tf.io.FixedLenFeature([], tf.string),  # [] означает отдельный элемент\n        # класс отсутствует, задача этого конкурса - предсказать классы цветов для тестового набора данных\n    }\n    example = tf.io.parse_single_example(example, UNLABELED_TFREC_FORMAT)\n    image = decode_image(example['image']) # преобразуем изображение к нужному нам формату\n    idnum = example['id']\n    return image, idnum # returns a dataset of image(s)\n\ndef load_dataset(filenames, labeled=True, ordered=False):\n    \"\"\"Читает из TFRecords. Для оптимальной производительности одновременное чтение из нескольких\n    файлов без учета порядка данных. Порядок не имеет значения, поскольку мы все равно будем перетасовывать данные\"\"\"\n\n    ignore_order = tf.data.Options() # Представляет параметры для tf.data.Dataset.\n    if not ordered:\n        ignore_order.experimental_deterministic = False # отключить порядок, увеличить скорость\n\n    dataset = tf.data.TFRecordDataset(filenames, num_parallel_reads=AUTO) # автоматически чередует чтение из нескольких файлов\n    dataset = dataset.with_options(ignore_order) # использует данные сразу после их поступления, а не в исходном порядке\n    dataset = dataset.map(read_labeled_tfrecord if labeled else read_unlabeled_tfrecord, num_parallel_calls=AUTO)\n    # возвращает набор данных пар (изображение, метка), если метка = Истина, или пар (изображение, идентификатор), если метка = Ложь\n    return dataset\n\ndef data_augment(image, label):\n    # data augmentation. Thanks to the dataset.prefetch(AUTO) statement in the next function (below),\n    # this happens essentially for free on TPU. Data pipeline code is executed on the \"CPU\" part\n    # of the TPU while the TPU itself is computing gradients.\n    flag = random.randint(1,3)\n  #  coef_1 = random.randint(70, 90) * 0.01\n  #  coef_2 = random.randint(70, 90) * 0.01\n    if flag == 1:\n        image = tf.image.random_flip_left_right(image, seed=SEED)\n    elif flag == 2:\n        image = tf.image.random_flip_up_down(image, seed=SEED)\n    else:\n        pass\n    return image, label   \n\ndef get_training_dataset():\n    dataset = load_dataset(TRAINING_FILENAMES, labeled=True)\n    dataset = dataset.map(data_augment, num_parallel_calls=AUTO)\n    dataset = dataset.repeat() # набор обучающих данных должен повторяться в течение нескольких эпох\n    dataset = dataset.shuffle(2048)\n    dataset = dataset.batch(BATCH_SIZE)\n    dataset = dataset.prefetch(AUTO) #готовим следующий набор, пока предыдущий обучается\n    return dataset\n\ndef get_validation_dataset():\n    dataset = load_dataset(VALIDATION_FILENAMES, labeled=True, ordered=False)\n    dataset = dataset.batch(BATCH_SIZE)\n    dataset = dataset.cache() # кешируем набор\n    dataset = dataset.prefetch(AUTO) #готовим следующий набор, пока предыдущий обучается\n    return dataset\n\ndef get_test_dataset(ordered=False):\n    dataset = load_dataset(TEST_FILENAMES, labeled=False, ordered=ordered)\n    dataset = dataset.batch(BATCH_SIZE)\n    dataset = dataset.prefetch(AUTO) #готовим следующий набор, пока предыдущий обучается\n    return dataset\n                               \ndef count_data_items(filenames):\n    # the number of data items is written in the name of the .tfrec files, i.e. flowers00-230.tfrec = 230 data items\n    n = [int(re.compile(r\"-([0-9]*)\\.\").search(filename).group(1)) for filename in filenames]\n    return np.sum(n)\n\n# training_dataset = get_training_dataset()\n# validation_dataset = get_validation_dataset()\n\nNUM_TRAINING_IMAGES = count_data_items(TRAINING_FILENAMES)\nNUM_VALIDATION_IMAGES = count_data_items(VALIDATION_FILENAMES)\nNUM_TEST_IMAGES = count_data_items(TEST_FILENAMES)\nSTEPS_PER_EPOCH = NUM_TRAINING_IMAGES // BATCH_SIZE\nprint('Dataset: {} training images, {} validation images, {} unlabeled test images'.format(NUM_TRAINING_IMAGES, NUM_VALIDATION_IMAGES, NUM_TEST_IMAGES))","metadata":{"execution":{"iopub.status.busy":"2023-12-22T10:31:01.735920Z","iopub.execute_input":"2023-12-22T10:31:01.736307Z","iopub.status.idle":"2023-12-22T10:31:01.755116Z","shell.execute_reply.started":"2023-12-22T10:31:01.736274Z","shell.execute_reply":"2023-12-22T10:31:01.754168Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":" **Построить модель на TPU**","metadata":{}},{"cell_type":"code","source":"# функция управляющая изменениями шага обучения в процессе тренировки нейронной сети\nLR_START = 0.00001\nLR_MAX = 0.00005 * strategy.num_replicas_in_sync#0.0001\nLR_MIN = 0.00001\nLR_RAMPUP_EPOCHS = 5\nLR_SUSTAIN_EPOCHS = 0\nLR_EXP_DECAY = .8\n\ndef lrfn(epoch):\n    if epoch < LR_RAMPUP_EPOCHS:\n        lr = (LR_MAX - LR_START) / LR_RAMPUP_EPOCHS * epoch + LR_START\n    elif epoch < LR_RAMPUP_EPOCHS + LR_SUSTAIN_EPOCHS:\n        lr = LR_MAX\n    else:\n        lr = (LR_MAX - LR_MIN) * LR_EXP_DECAY**(epoch - LR_RAMPUP_EPOCHS - LR_SUSTAIN_EPOCHS) + LR_MIN\n    return lr\n    \nlr_callback = tf.keras.callbacks.LearningRateScheduler(lrfn, verbose=True)\n\n# построим график изменения шага обучение в зависимости от эпох\nrng = [i for i in range(EPOCHS)]\ny = [lrfn(x) for x in rng]\nplt.plot(rng, y)\nprint(\"Learning rate schedule: {:.3g} to {:.3g} to {:.3g}\".format(y[0], max(y), y[-1]))","metadata":{"execution":{"iopub.status.busy":"2023-12-22T10:31:24.447635Z","iopub.execute_input":"2023-12-22T10:31:24.448448Z","iopub.status.idle":"2023-12-22T10:31:24.608074Z","shell.execute_reply.started":"2023-12-22T10:31:24.448412Z","shell.execute_reply":"2023-12-22T10:31:24.607233Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"def get_model(use_model, weights):\n    # noisy-student\n    base_model = use_model(weights=weights, \n                      include_top=False, pooling='avg',\n                      input_shape=(*IMAGE_SIZE, 3))\n    x = base_model.output\n    predictions = Dense(104, activation='softmax')(x)\n    return Model(inputs=base_model.input, outputs=predictions)","metadata":{"execution":{"iopub.status.busy":"2023-12-22T10:33:01.024736Z","iopub.execute_input":"2023-12-22T10:33:01.025552Z","iopub.status.idle":"2023-12-22T10:33:01.030493Z","shell.execute_reply.started":"2023-12-22T10:33:01.025516Z","shell.execute_reply":"2023-12-22T10:33:01.029590Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"# **DenseNet201**","metadata":{}},{"cell_type":"code","source":"with strategy.scope():    \n    dense = get_model(DenseNet201, 'imagenet')\n        \ndense.compile(\n    optimizer='nadam',\n    loss = 'sparse_categorical_crossentropy',\n    metrics=['sparse_categorical_accuracy']\n)","metadata":{"execution":{"iopub.status.busy":"2023-12-22T09:07:13.774879Z","iopub.execute_input":"2023-12-22T09:07:13.775255Z","iopub.status.idle":"2023-12-22T09:08:11.868173Z","shell.execute_reply.started":"2023-12-22T09:07:13.775220Z","shell.execute_reply":"2023-12-22T09:08:11.867179Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"history = dense.fit(get_training_dataset(), \n          steps_per_epoch=STEPS_PER_EPOCH, \n          epochs=EPOCHS, \n          callbacks=[lr_callback, ModelCheckpoint(filepath='dense.h5', monitor='val_loss',\n                                  save_best_only=True)],validation_data=get_validation_dataset(), workers = 3)","metadata":{"execution":{"iopub.status.busy":"2023-12-22T09:08:11.869572Z","iopub.execute_input":"2023-12-22T09:08:11.869832Z","iopub.status.idle":"2023-12-22T09:55:56.096606Z","shell.execute_reply.started":"2023-12-22T09:08:11.869794Z","shell.execute_reply":"2023-12-22T09:55:56.095353Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"plt.plot(history.history['sparse_categorical_accuracy'], \n         label='Оценка точности на обучающем наборе')\nplt.plot(history.history['val_sparse_categorical_accuracy'], \n         label='Оценка точности на проверочном наборе')\nplt.xlabel('Эпоха обучения')\nplt.ylabel('Оценка точности')\nplt.legend()\nplt.show()","metadata":{"execution":{"iopub.status.busy":"2023-12-22T09:56:35.700377Z","iopub.execute_input":"2023-12-22T09:56:35.700791Z","iopub.status.idle":"2023-12-22T09:56:35.854092Z","shell.execute_reply.started":"2023-12-22T09:56:35.700754Z","shell.execute_reply":"2023-12-22T09:56:35.853114Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"plt.plot(history.history['loss'], \n         label='Оценка потерь на обучающем наборе')\nplt.plot(history.history['val_loss'], \n         label='Оценка потерь на проверочном наборе')\nplt.xlabel('Эпоха обучения')\nplt.ylabel('Оценка потерь')\nplt.legend()\nplt.show()","metadata":{"execution":{"iopub.status.busy":"2023-12-22T09:56:39.516503Z","iopub.execute_input":"2023-12-22T09:56:39.517053Z","iopub.status.idle":"2023-12-22T09:56:39.658178Z","shell.execute_reply.started":"2023-12-22T09:56:39.517018Z","shell.execute_reply":"2023-12-22T09:56:39.657272Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"# **EfficientNetB7**","metadata":{}},{"cell_type":"code","source":"with strategy.scope():    \n    efficientnet = get_model(EfficientNetB7, 'noisy-student')\n        \nefficientnet.compile(\n    optimizer='adam',\n    loss = 'sparse_categorical_crossentropy',\n    metrics=['sparse_categorical_accuracy'])","metadata":{"execution":{"iopub.status.busy":"2023-12-22T10:33:29.880502Z","iopub.execute_input":"2023-12-22T10:33:29.880906Z","iopub.status.idle":"2023-12-22T10:34:26.458081Z","shell.execute_reply.started":"2023-12-22T10:33:29.880868Z","shell.execute_reply":"2023-12-22T10:34:26.456711Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"history = efficientnet.fit(get_training_dataset(), \n          steps_per_epoch=STEPS_PER_EPOCH, \n          epochs=EPOCHS, \n          callbacks=[lr_callback, ModelCheckpoint(filepath='efnetb7.h5', monitor='val_loss',\n                                  save_best_only=True)],validation_data=get_validation_dataset(),workers = 3)","metadata":{"execution":{"iopub.status.busy":"2023-12-22T10:34:33.712188Z","iopub.execute_input":"2023-12-22T10:34:33.712590Z","iopub.status.idle":"2023-12-22T11:38:25.051346Z","shell.execute_reply.started":"2023-12-22T10:34:33.712556Z","shell.execute_reply":"2023-12-22T11:38:25.049980Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"plt.plot(history.history['sparse_categorical_accuracy'], \n         label='Оценка точности на обучающем наборе')\nplt.plot(history.history['val_sparse_categorical_accuracy'], \n         label='Оценка точности на проверочном наборе')\nplt.xlabel('Эпоха обучения')\nplt.ylabel('Оценка точности')\nplt.legend()\nplt.show()","metadata":{"execution":{"iopub.status.busy":"2023-12-22T11:39:02.085334Z","iopub.execute_input":"2023-12-22T11:39:02.085685Z","iopub.status.idle":"2023-12-22T11:39:02.236459Z","shell.execute_reply.started":"2023-12-22T11:39:02.085653Z","shell.execute_reply":"2023-12-22T11:39:02.235501Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"plt.plot(history.history['loss'], \n         label='Оценка потерь на обучающем наборе')\nplt.plot(history.history['val_loss'], \n         label='Оценка потерь на проверочном наборе')\nplt.xlabel('Эпоха обучения')\nplt.ylabel('Оценка потерь')\nplt.legend()\nplt.show()","metadata":{"execution":{"iopub.status.busy":"2023-12-22T11:39:05.341009Z","iopub.execute_input":"2023-12-22T11:39:05.341374Z","iopub.status.idle":"2023-12-22T11:39:05.499068Z","shell.execute_reply.started":"2023-12-22T11:39:05.341343Z","shell.execute_reply":"2023-12-22T11:39:05.498026Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"# **Vgg19**","metadata":{}},{"cell_type":"code","source":"with strategy.scope():    \n    vgg = get_model(VGG19, 'imagenet')\n        \nvgg.compile(\n    optimizer='adam',\n    loss = 'sparse_categorical_crossentropy',\n    metrics=['sparse_categorical_accuracy'])","metadata":{"execution":{"iopub.status.busy":"2023-12-22T12:39:01.472186Z","iopub.execute_input":"2023-12-22T12:39:01.473453Z","iopub.status.idle":"2023-12-22T12:39:06.960018Z","shell.execute_reply.started":"2023-12-22T12:39:01.473405Z","shell.execute_reply":"2023-12-22T12:39:06.958886Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"history = vgg.fit(get_training_dataset(), \n          steps_per_epoch=STEPS_PER_EPOCH, \n          epochs=EPOCHS, \n          callbacks=[lr_callback, ModelCheckpoint(filepath='vgg.h5', monitor='val_loss',\n                                  save_best_only=True)],validation_data=get_validation_dataset(),workers = 3)","metadata":{"execution":{"iopub.status.busy":"2023-12-22T12:39:23.480503Z","iopub.execute_input":"2023-12-22T12:39:23.480932Z","iopub.status.idle":"2023-12-22T13:13:00.061093Z","shell.execute_reply.started":"2023-12-22T12:39:23.480895Z","shell.execute_reply":"2023-12-22T13:13:00.059856Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"plt.plot(history.history['sparse_categorical_accuracy'], \n         label='Оценка точности на обучающем наборе')\nplt.plot(history.history['val_sparse_categorical_accuracy'], \n         label='Оценка точности на проверочном наборе')\nplt.xlabel('Эпоха обучения')\nplt.ylabel('Оценка точности')\nplt.legend()\nplt.show()","metadata":{"execution":{"iopub.status.busy":"2023-12-22T13:13:38.962572Z","iopub.execute_input":"2023-12-22T13:13:38.963713Z","iopub.status.idle":"2023-12-22T13:13:39.124853Z","shell.execute_reply.started":"2023-12-22T13:13:38.963674Z","shell.execute_reply":"2023-12-22T13:13:39.123849Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"plt.plot(history.history['loss'], \n         label='Оценка потерь на обучающем наборе')\nplt.plot(history.history['val_loss'], \n         label='Оценка потерь на проверочном наборе')\nplt.xlabel('Эпоха обучения')\nplt.ylabel('Оценка потерь')\nplt.legend()\nplt.show()","metadata":{"execution":{"iopub.status.busy":"2023-12-22T13:13:41.530074Z","iopub.execute_input":"2023-12-22T13:13:41.530455Z","iopub.status.idle":"2023-12-22T13:13:41.697016Z","shell.execute_reply.started":"2023-12-22T13:13:41.530420Z","shell.execute_reply":"2023-12-22T13:13:41.695944Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"# **Сравнение моделей**","metadata":{}},{"cell_type":"code","source":"eval_dataset = get_validation_dataset()\n\neval_results = {\n    'DenseNet201': dense.evaluate(eval_dataset)[1],\n    'EfficientNetB7': efficientnet.evaluate(eval_dataset)[1],\n    'Vgg19': vgg.evaluate(eval_dataset)[1],\n}\n\nfig, ax = plt.subplots()\nax.bar(list(eval_results.keys()), list(eval_results.values()))\nax.set_ylabel('sparse_categorical_accuracy')\nax.set_title('Метрика моделей на валидационных данных')\nplt.show()","metadata":{"execution":{"iopub.status.busy":"2023-12-22T13:14:12.137397Z","iopub.execute_input":"2023-12-22T13:14:12.137760Z","iopub.status.idle":"2023-12-22T13:14:54.468716Z","shell.execute_reply.started":"2023-12-22T13:14:12.137727Z","shell.execute_reply":"2023-12-22T13:14:54.467848Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"Лучший результат показала EfficientNetB7","metadata":{}},{"cell_type":"markdown","source":"# **Тест**","metadata":{}},{"cell_type":"code","source":"model = tf.keras.models.load_model('efnetb7.h5')","metadata":{"execution":{"iopub.status.busy":"2023-12-22T11:40:04.359518Z","iopub.execute_input":"2023-12-22T11:40:04.359953Z","iopub.status.idle":"2023-12-22T11:40:13.611427Z","shell.execute_reply.started":"2023-12-22T11:40:04.359916Z","shell.execute_reply":"2023-12-22T11:40:13.610159Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"# Поскольку мы разделяем набор данных и выполняем итерацию отдельно для изображений и идентификаторов, порядок имеет значение.\ntest_ds = get_test_dataset(ordered=True) \n\nprint('Вычисляем предсказания...')\ntest_images_ds = test_ds.map(lambda image, idnum: image)\nprobabilities = model.predict(test_images_ds)\npredictions = np.argmax(probabilities, axis=-1)\nprint(predictions)\n\nprint('Создание файла submission.csv...')\ntest_ids_ds = test_ds.map(lambda image, idnum: idnum).unbatch()\ntest_ids = next(iter(test_ids_ds.batch(NUM_TEST_IMAGES))).numpy().astype('U') # все в одной партии\nnp.savetxt('submission.csv', np.rec.fromarrays([test_ids, predictions]), fmt=['%s', '%d'], delimiter=',', header='id,label', comments='')","metadata":{"execution":{"iopub.status.busy":"2023-12-22T11:40:25.623854Z","iopub.execute_input":"2023-12-22T11:40:25.624305Z","iopub.status.idle":"2023-12-22T12:35:36.122770Z","shell.execute_reply.started":"2023-12-22T11:40:25.624268Z","shell.execute_reply":"2023-12-22T12:35:36.121550Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"![image.png](attachment:8873624f-71f0-4dc2-9b9d-b5f18c2fdbfd.png)","metadata":{},"attachments":{"8873624f-71f0-4dc2-9b9d-b5f18c2fdbfd.png":{"image/png":"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"}}}]}