python - 使用 Keras 回调保存最好的 val_loss

Question

我根据此处讨论的内容为 mnist 数据集设置了一个去噪自动编码器: https://blog.keras.io/building-autoencoders-in-keras.html

我正在尝试查看输入图像的重建如何随时间变化；我注意到有时 DAE 尖峰的损失(用于训练和验证)，例如从损失 ~0.12 到 ~3.0。为了避免在训练过程中使用这些“失误”，我尝试使用 Keras 的回调，保存最佳权重(明智的 val_loss)并在每个训练“段”(在我的例子中为 10 个时期)后加载它们。

但是，我收到一条错误消息:

File "noise_e_mini.py", line 71, in <module>
callbacks=([checkpointer]))
File "/usr/local/lib/python2.7/dist-packages/keras/engine/training.py", line 1650, in fit
validation_steps=validation_steps)
File "/usr/local/lib/python2.7/dist-packages/keras/engine/training.py", line 1145, in _fit_loop
callbacks.set_model(callback_model)
File "/usr/local/lib/python2.7/dist-packages/keras/callbacks.py", line 48, in set_model
callback.set_model(model)
AttributeError: 'tuple' object has no attribute 'set_model'

我的代码是:

from keras.layers import Input, Dense
from keras.models import Model
from keras import regularizers
from keras.callbacks import ModelCheckpoint
input_img = Input(shape=(784,))

filepath_for_w='denoise_by_AE_weights_1.h5'


def autoencoder_block(input,l1_score_encode,l1_score_decode):


    #    encoder:
    encoded = Dense(256, activation='relu',activity_regularizer=regularizers.l1(l1_score_encode))(input_img)
    encoded = Dense(128, activation='relu',activity_regularizer=regularizers.l1(l1_score_encode))(encoded)
    encoded = Dense(64, activation='relu',activity_regularizer=regularizers.l1(l1_score_encode))(encoded)
    encoded = Dense(32, activation='relu',activity_regularizer=regularizers.l1(l1_score_encode))(encoded)

    encoder = Model (input=input_img, output=encoded)

    #    decoder:
    connection_layer= Input(shape=(32,))
    decoded = Dense(64, activation='relu',activity_regularizer=regularizers.l1(l1_score_decode))(connection_layer)
    decoded = Dense(128, activation='relu',activity_regularizer=regularizers.l1(l1_score_decode))(decoded)
    decoded = Dense(256, activation='relu',activity_regularizer=regularizers.l1(l1_score_decode))(decoded)
    decoded = Dense(784, activation='sigmoid',activity_regularizer=regularizers.l1(l1_score_decode))(decoded)

    decoder = Model (input=connection_layer , output=decoded)

    crunched = encoder(input_img)
    final = decoder(crunched)

    autoencoder = Model(input=input_img, output=final)
    autoencoder.compile(optimizer='adadelta', loss='binary_crossentropy')
    return (autoencoder)



from keras.datasets import mnist
import numpy as np
(x_train, y_train), (x_test, y_test) = mnist.load_data()


x_train = x_train.astype('float32') / 255.
x_test = x_test.astype('float32') / 255.
x_train = x_train.reshape((len(x_train), np.prod(x_train.shape[1:])))
x_test = x_test.reshape((len(x_test), np.prod(x_test.shape[1:])))
print x_train.shape
print x_test.shape

noise_factor = 0.5

x_test_noisy = x_test + noise_factor * np.random.normal(loc=0.0, scale=1.0, size=x_test.shape) 
x_test_noisy = np.clip(x_test_noisy, 0., 1.)



autoencoder=autoencoder_block(input_img,0,0)

for i in range (10):

    x_train_noisy = x_train + noise_factor * np.random.normal(loc=0.0, scale=1.0, size=x_train.shape) 
    x_train_noisy = np.clip(x_train_noisy, 0., 1.)
    checkpointer=ModelCheckpoint(filepath_for_w, monitor='val_loss', verbose=0, save_best_only=True, save_weights_only=True, mode='auto', period=1),

    autoencoder.fit(x_train_noisy, x_train,
                epochs=10,
                batch_size=256,            
                shuffle=True,
                validation_data=(x_test_noisy, x_test),
                callbacks=([checkpointer]))
    autoencoder.load_weights(filepath_for_w)  # load weights from the best in the run

    decoded_imgs = autoencoder.predict(x_test_noisy) # save results for this stage for presentation
    np.save('decoded'+str(i)+'.npy',decoded_imgs)    ####

np.save('tested.npy',x_test_noisy)
np.save ('true_catagories.npy',y_test)
np.save('original.npy',x_test)


autoencoder.save('denoise_by_AE_model_1.h5')

我做错了什么？非常感谢:)

Answer 1

你的问题可能在这一行之内

callbacks=([checkpointer]))

您需要删除括号，因为回调需要列表，而不是元组，尝试:

callbacks=[checkpointer]

我还注意到您的检查点以逗号结尾，您也应该将其删除。

checkpointer=ModelCheckpoint(filepath_for_w, monitor='val_loss', verbose=0, save_best_only=True, save_weights_only=True, mode='auto', period=1),