假设我有这样的模型：
def mask_layer(tensor):return layers.Multiply()([tensor, tf.ones([1, 128])])def get_model():inp_1 = keras.Input(shape=(64, 101, 1), name="input")x = layers.Conv2D(256, kernel_size=(3, 3), kernel_regularizer=l2(1e-6), strides=(3, 3), padding="same")(inp_1)x = layers.LeakyReLU(alpha=0.3)(x)x = layers.Conv2D(128, kernel_size=(3, 3), kernel_regularizer=l2(1e-6), strides=(3, 3), padding="same")(x)x = layers.LeakyReLU(alpha=0.3)(x)x = layers.Flatten()(x)x = layers.Dense(512)(x)x = layers.LeakyReLU(alpha=0.3)(x)x = layers.Dense(256)(x)x = layers.LeakyReLU(alpha=0.3)(x)x= layers.Dense(128, name="output1")(x)mask = layers.Lambda(mask_layer, name="lambda_layer")(x)out2 = layers.Dense(40000, name="output2")(mask)model = keras.Model(inp_1, [mask, output2], name="2_out_model")model.compile(optimizer=tf.keras.optimizers.Adam(learning_rate=0.001), loss="mean_squared_error")plot_model(model, to_file='model.png', show_shapes=True, show_layer_names=True)model.summary()return model然后，我训练我的网络：
model = get_model()es = tf.keras.callbacks.EarlyStopping(monitor='val_loss', mode='min', verbose=1, patience=50)history = model.fit(X_train, [Y_train, Z_train], validation_data=(X_val, [Y_val, Z_val]), epochs=500, batch_size=32, callbacks=[es])test_loss, _, _ = model.evaluate(X_test, [Y_test, Z_test], verbose=1)我想用另一组训练数据重新训练已经训练过的网络，但要改变Lambda层的定义，假设这次函数返回的是：
return layers.Multiply()([tensor, tf.ones([1, 128])*1.2])我是否需要重新调用”get_model()”函数（因为我重新定义了一个层）然后再次训练？这样做不会有重新初始化模型权重的风险吗？提前感谢你 🙂
回答：
你的Lambda层不是可训练的层，所以你可以安全地将训练好的权重移动到另一个模型（具有相同的结构）中，但更改你的Lambda层
以下是示例：
def mask_layer1(tensor): return layers.Multiply()([tensor, tf.ones([1, 128])])def mask_layer2(tensor): return layers.Multiply()([tensor, tf.ones([1, 128])*1.2])def get_model(mask_kind): inp = keras.Input(shape=(64, 101, 1), name="input") x = layers.Conv2D(256, kernel_size=(3, 3), kernel_regularizer=l2(1e-6), strides=(3, 3), padding="same")(inp) x = layers.LeakyReLU(alpha=0.3)(x) x = layers.Conv2D(128, kernel_size=(3, 3), kernel_regularizer=l2(1e-6), strides=(3, 3), padding="same")(x) x = layers.LeakyReLU(alpha=0.3)(x) x = layers.Flatten()(x) x = layers.Dense(512)(x) x = layers.LeakyReLU(alpha=0.3)(x) x = layers.Dense(256)(x) x = layers.LeakyReLU(alpha=0.3)(x) x = layers.Dense(128, name="output1")(x) if mask_kind == 1: mask = layers.Lambda(mask_layer1, name="lambda_layer")(x) elif mask_kind == 2: mask = layers.Lambda(mask_layer2, name="lambda_layer")(x) else: mask = x out = layers.Dense(40000, name="output2")(mask) model = keras.Model(inp, [mask, out], name="2_out_model") model.compile(optimizer=tf.keras.optimizers.Adam(learning_rate=0.001), loss="mean_squared_error") return modelmodel1 = get_model(mask_kind=1)model1.fit(...)model2 = get_model(mask_kind=2)model2.set_weights(model1.get_weights())model2.fit(...)