tensorflow学习笔计-layers.Lambda

tf.keras.layers.Lambda

函数说明

将任意表达式包装为Layer对象。

tf.keras.layers.Lambda(
    function, output_shape=None, mask=None, arguments=None, **kwargs
)

使用示例

实现方式一

def test_lambda_layer():
    layers_lambda = tf.keras.layers.Lambda(lambda x: x ** 2)
    X = np.linspace(1, 10, 10)
    print(layers_lambda(X))

if __name__ == '__main__':
    test_lambda_layer()

console out:

tf.Tensor([  1.   4.   9.  16.  25.  36.  49.  64.  81. 100.], shape=(10,), dtype=float32)

实现方式二

def test_lambda_layer2():
    class ScaleLayer(tf.keras.layers.Layer):
        def call(self, inputs, *args, **kwargs):
            return inputs ** 2

    layers_lambda = ScaleLayer()
    X = np.linspace(1, 10, 10)
    print(layers_lambda(X))


if __name__ == '__main__':
    test_lambda_layer2()

思考，在Layer引入自变量，模型是否会自动训练

经过多次尝试，该变量会自动训练

def test_lambda_layer3():
    X = np.linspace(1, 10, 10000)
    y = X ** 2

    X = X.reshape(len(X), 1)

    class ScaleLayer(tf.keras.layers.Layer):

        def __init__(self, trainable=True, name=None, dtype=None, dynamic=False, **kwargs):
            super().__init__(trainable, name, dtype, dynamic, **kwargs)
            self.scale = tf.Variable(9.)

        def call(self, inputs, *args, **kwargs):
            return inputs ** self.scale

    layers_lambda = ScaleLayer()
    print(layers_lambda(X))

    model = tf.keras.Sequential()
    model.add(layers_lambda)
    model.compile(optimizer="adam", loss="mse")
    model.fit(X, y, epochs=500)

    print(layers_lambda.scale)


if __name__ == '__main__':
    test_lambda_layer3()