為方便列印除錯

記錄行號和檔名 變數值和型別

    def print_variables(var, line_no):
        import inspect
        frame = inspect.currentframe()
        # __FILE__
        fileName = frame.f_code.co_filename
        # __LINE__
        # fileNo = frame.f_lineno
        print('value: ', var, 'File:', fileName, 'at Line:', line_no)
        print('type: ', type(var), 'File:', fileName, 'at Line:', line_no)

    import inspect
    frame = inspect.currentframe()

    print_variables(val_batches,frame.f_lineno)
    print_variables(val_batches // 5, frame.f_lineno)

列印9宮格

從資料集列印圖片

    def plot_9_img():
        plt.figure(figsize=(10, 10))
        for images, labels in train_dataset.take(1):
            for i in range(9):
                ax = plt.subplot(3, 3, i + 1)
                plt.imshow(images[i].numpy().astype("uint8"))
                plt.title(class_names[labels[i]])
                plt.axis("off")
            plt.show()

    # plot_9_img()

獲得資料集資料夾的名字

class_names = train_dataset.class_names
print(class_names, 'is class_names')

從目錄構建資料集

    train_dataset = image_dataset_from_directory(train_dir,
                                                 shuffle=True,
                                                 batch_size=BATCH_SIZE,
                                                 image_size=IMG_SIZE)
    validation_dataset = image_dataset_from_directory(validation_dir,
                                                      shuffle=True,
                                                      batch_size=BATCH_SIZE,
                                                      image_size=IMG_SIZE)

如果最後不用啟用函式 logit

from_logits=True

    model.compile(optimizer=tf.keras.optimizers.Adam(lr=base_learning_rate),
                  loss=tf.keras.losses.BinaryCrossentropy(from_logits=True),
                  metrics=['accuracy'])

next iter 只迭代一個batch資料集

    IMG_SHAPE = IMG_SIZE + (3,)
    base_model = tf.keras.applications.MobileNetV2(input_shape=IMG_SHAPE,
                                                   include_top=False,
                                                   weights='imagenet')

    image_batch, label_batch = next(iter(train_dataset))
    feature_batch = base_model(image_batch)

    print_variables(feature_batch.shape,frame.f_lineno)

自定義模型

training=False

base_model = tf.keras.applications.MobileNetV2(input_shape=IMG_SHAPE,
                                                   include_top=False,
                                                   weights='imagenet')

    def build_model():
        inputs = tf.keras.Input(shape=(160, 160, 3))
        x = data_augmentation(inputs)
        x = preprocess_input(x)
        x = base_model(x, training=False)
        x = global_average_layer(x)
        x = tf.keras.layers.Dropout(0.2)(x)
        outputs = prediction_layer(x)
        model = tf.keras.Model(inputs, outputs)
        return model

    data_augmentation = tf.keras.Sequential([
        tf.keras.layers.experimental.preprocessing.RandomFlip('horizontal'),
        tf.keras.layers.experimental.preprocessing.RandomRotation(0.2),
    ])


	preprocess_input = tf.keras.applications.mobilenet_v2.preprocess_input

此時可訓練引數很少

Model: "model"
_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
input_2 (InputLayer)         [(None, 160, 160, 3)]     0         
_________________________________________________________________
sequential (Sequential)      (None, 160, 160, 3)       0         
_________________________________________________________________
tf.math.truediv (TFOpLambda) (None, 160, 160, 3)       0         
_________________________________________________________________
tf.math.subtract (TFOpLambda (None, 160, 160, 3)       0         
_________________________________________________________________
mobilenetv2_1.00_160 (Functi (None, 5, 5, 1280)        2257984   
_________________________________________________________________
global_average_pooling2d (Gl (None, 1280)              0         
_________________________________________________________________
dropout (Dropout)            (None, 1280)              0         
_________________________________________________________________
dense (Dense)                (None, 1)                 1281      
=================================================================
Total params: 2,259,265
Trainable params: 1,281
Non-trainable params: 2,257,984
_________________________________________________________________

儲存與恢復模型

	checkpoint_path = 'checkpoints/' + dataset_modelname
    checkpoint = tf.train.Checkpoint(myAwesomeModel=model)
    checkpoint.restore(tf.train.latest_checkpoint(checkpoint_path))
    # 使用tf.train.CheckpointManager管理Checkpoint
    manager = tf.train.CheckpointManager(checkpoint, directory=checkpoint_path, max_to_keep=3)

    history = model.fit(train_ds, epochs=max_epochs, validation_data=test_ds, steps_per_epoch=steps_per_epoch)
    path = manager.save(checkpoint_number=20)
    print("model 儲存在 %s" % path)

儲存訓練結果

    dataset_modelname = 'flower_photos_mobile'
    history = compile_and_fit_v1(model, dataset_modelname=dataset_modelname, max_epochs=10,
                                 train_ds=ds, test_ds=vs, steps_per_epoch=steps_per_epoch)

    npz_path = 'plot_npz/' + dataset_modelname + '.npz'
    npz_save(npz_path, history=history)
    plot_from_npz_loss(npz_path)

		checkpoint_path = 'checkpoints/' + dataset_modelname

        checkpoint = tf.train.Checkpoint(myAwesomeModel=model)
        # checkpoint.restore(tf.train.latest_checkpoint(checkpoint_path))
        # 使用tf.train.CheckpointManager管理Checkpoint
        manager = tf.train.CheckpointManager(checkpoint, directory=checkpoint_path, max_to_keep=3)
history = model.fit(train_dataset,
                            epochs=initial_epochs,
                            validation_data=validation_dataset)
        path = manager.save(checkpoint_number=initial_epochs)
        print("model 儲存在 %s" % path)

json儲存

import json
        history_json = {}
        history_json['history'] = history.history
        history_json['epoch'] = history.epoch
        with open('plot_npz/'+dataset_modelname+'.json','w') as file:
            file.write(json.dumps(history_json))
            

preprocess from img dir

from tensorflow.keras.preprocessing import image_dataset_from_directory
    _URL = 'https://storage.googleapis.com/mledu-datasets/cats_and_dogs_filtered.zip'
    path_to_zip = tf.keras.utils.get_file('cats_and_dogs.zip', origin=_URL, extract=True)
    PATH = os.path.join(os.path.dirname(path_to_zip), 'cats_and_dogs_filtered')
    print(PATH, 'is PATH')

    train_dir = os.path.join(PATH, 'train')
    validation_dir = os.path.join(PATH, 'validation')

    BATCH_SIZE = 32
    IMG_SIZE = (160, 160)

    train_dataset = image_dataset_from_directory(train_dir,
                                                 shuffle=True,
                                                 seed=10,
                                                 batch_size=BATCH_SIZE,
                                                 image_size=IMG_SIZE)
    validation_dataset = image_dataset_from_directory(validation_dir,
                                                      shuffle=True,
                                                      seed=10,
                                                      batch_size=BATCH_SIZE,
                                                      image_size=IMG_SIZE)

    class_names = train_dataset.class_names
    print(class_names, 'is class_names')

PATH = os.path.join('datasets', 'flower_photos')
    print(PATH, 'is PATH')
    BATCH_SIZE = 32
    IMG_SIZE = (160, 160)
    train_dataset = image_dataset_from_directory(PATH,
                                                 shuffle=True,
                                                 seed=123,
                                                 validation_split=0.2,
                                                 subset="training",
                                                 batch_size=BATCH_SIZE,
                                                 image_size=IMG_SIZE)
    validation_dataset = image_dataset_from_directory(PATH,
                                                      shuffle=True,
                                                      seed=123,
                                                      validation_split=0.2,
                                                      subset="validation",
                                                      batch_size=BATCH_SIZE,
                                                      image_size=IMG_SIZE)
    class_names = train_dataset.class_names
    print(class_names, 'is class_names')

data augmentation

AUTOTUNE = tf.data.experimental.AUTOTUNE

    train_dataset = train_dataset.prefetch(buffer_size=AUTOTUNE)
    validation_dataset = validation_dataset.prefetch(buffer_size=AUTOTUNE)
    test_dataset = test_dataset.prefetch(buffer_size=AUTOTUNE)

    data_augmentation = tf.keras.Sequential([
        tf.keras.layers.experimental.preprocessing.RandomFlip('horizontal'),
        tf.keras.layers.experimental.preprocessing.RandomRotation(0.2),
    ])

    def plot_aug():
        for image, _ in train_dataset.take(1):
            plt.figure(figsize=(10, 10))
            first_image = image[0]
            import inspect
            frame = inspect.currentframe()
            print_variables(first_image, frame.f_lineno)
            print_variables(tf.expand_dims(first_image, 0), frame.f_lineno)
            for i in range(9):
                ax = plt.subplot(3, 3, i + 1)
                augmented_image = data_augmentation(tf.expand_dims(first_image, 0))
                plt.imshow(augmented_image[0] / 255)
                plt.axis('off')
            plt.show()

    # plot_aug()

preprocess_input = tf.keras.applications.mobilenet_v2.preprocess_input
    rescale = tf.keras.layers.experimental.preprocessing.Rescaling(1. / 127.5, offset=-1)
    # Create the base model from the pre-trained model MobileNet V2
    IMG_SHAPE = IMG_SIZE + (3,)
    base_model = tf.keras.applications.MobileNetV2(input_shape=IMG_SHAPE,
                                                   include_top=False,
                                                   weights='imagenet')

image_batch, label_batch = next(iter(train_dataset))
    feature_batch = base_model(image_batch)

    print_variables(feature_batch.shape,frame.f_lineno)
    base_model.trainable = False

    # Let's take a look at the base model architecture
    # base_model.summary()

    global_average_layer = tf.keras.layers.GlobalAveragePooling2D()
    feature_batch_average = global_average_layer(feature_batch)
    print_variables(feature_batch_average.shape,frame.f_lineno)

    prediction_layer = tf.keras.layers.Dense(1)
    prediction_batch = prediction_layer(feature_batch_average)

    print_variables(prediction_batch.shape, frame.f_lineno)