我正在尝试使用Keras和MNIST数据集在预测模型中冻结特定层的权重，但它不起作用。代码如下所示：

from keras.layers import Dense, Flattenfrom keras.utils import to_categoricalfrom keras.models import Sequential, load_modelfrom keras.datasets import mnistfrom keras.losses import categorical_crossentropyimport numpy as npdef load_data():    (x_train, y_train), (x_test, y_test) = mnist.load_data()    x_train = x_train.astype('float32')    x_test = x_test.astype('float32')    x_train /= 255    x_test /= 255    y_train = to_categorical(y_train, num_classes=10)    y_test = to_categorical(y_test, num_classes=10)    return x_train, y_train, x_test, y_testdef run():    x_train, y_train, x_test, y_test = load_data()    model = Sequential([Flatten(input_shape=(28, 28)),                        Dense(300, name='dense1', activation='relu'),                        Dense(100, name='dense2', activation='relu'),                        Dense(10, name='dense3', activation='softmax')])    model.trainable = True    model.compile(optimizer='Adam',                  metrics=['accuracy'],                  loss=categorical_crossentropy)    print(model.summary())    model.fit(x_train, y_train, epochs=5, verbose=2)    print(model.evaluate(x_test, y_test))    return modeldef freeze(model):    x_train, y_train, x_test, y_test = load_data()    name = 'dense1'    weightsAndBias = model.get_layer(name=name).get_weights()    # 冻结这一层的权重    model.get_layer(name=name).trainable = False    # 记录重新训练前的权重    weights_before = weightsAndBias[0]    # 重新训练    model.fit(x_train, y_train, verbose=2, epochs=1)    weights_after = model.get_layer(name=name).get_weights()[0]    if (weights_before == weights_after).all():        print('权重没有变化!!!')    else:        print('权重变化了!!!!')if __name__ == '__main__':    model = run()    freeze(model)

程序输出的是’权重变化了!!!!’。我无法理解为什么在设置model.get_layer(name=name).trainable = False之后，名为’dense1’的层的权重仍然会发生变化。

回答：

你可以使用以下方式来实现：

model=Sequential()layer=Dense(64,init='glorot_uniform',input_shape=(784,))layer.trainable=Falsemodel.add(layer)layer2=Dense(784, activation='sigmoid',init='glorot_uniform')layer2.trainable=Truemodel.add(layer2)model.compile(loss='relu', optimizer=sgd,metrics = ['mae'])