Question

我正在（试图）编写一个自定义的Keras层，该层实现以下各个组件：

x-> a x + b ReLU（x）

具有a和b可训练的重量。到目前为止，这是我尝试过的事情：

Server Error in '/..' Application.
Runtime Error
    Description: An application error occurred on the server. The current custom error settings for this application prevent the details of the application error from being viewed remotely (for security reasons). It could, however, be viewed by browsers running on the local server machine.

    Details: To enable the details of this specific error message to be viewable on remote machines, please create a <customErrors> tag within a "web.config" configuration file located in the root directory of the current web application. This <customErrors> tag should then have its "mode" attribute set to "Off".


    <!-- Web.Config Configuration File -->

    <configuration>
        <system.web>
            <customErrors mode="Off"/>
        </system.web>
    </configuration>

    Notes: The current error page you are seeing can be replaced by a custom error page by modifying the "defaultRedirect" attribute of the application's <customErrors> configuration tag to point to a custom error page URL.


    <!-- Web.Config Configuration File -->

    <configuration>
        <system.web>
            <customErrors mode="RemoteOnly" defaultRedirect="mycustompage.htm"/>
        </system.web>
    </configuration>

但是，出现错误。我认为问题在于我不知道如何定义可训练的“标量” ...我是否正确地认为了这一点以及如何做到这一点？

编辑/添加：

这是我尝试用ReLU替换为“ Custom_ReLU”的方式构建纯前馈结构的方法：

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

    def __init__(self, units=d):
        super(Custom_ReLU, self).__init__()
        self.units = units

    def build(self, input_shape):
        self.a1 = self.add_weight(shape=[1],
                                initializer = 'random_uniform',
                                trainable=True)
        self.a2 = self.add_weight(shape=[1],
                                initializer = 'random_uniform',
                                trainable=True)

    def call(self,inputs):
        return self.a1*inputs + self.a2*(tf.nn.relu(inputs))

这是错误的摘要：

# Build Vanilla Network
inputs_ffNN = tf.keras.Input(shape=(d,))
x_ffNN = fullyConnected_Dense(d)(inputs_ffNN)
for i in range(Depth):
    x_HTC = Custom_ReLU(x_ffNN)
    x_ffNN = fullyConnected_Dense(d)(x_ffNN)
outputs_ffNN = fullyConnected_Dense(D)(x_ffNN)
ffNN = tf.keras.Model(inputs_ffNN, outputs_ffNN)

Answer 1

我使用您的图层没问题：

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

    def __init__(self):
        super(Custom_ReLU, self).__init__()

        self.a1 = self.add_weight(shape=[1],
                                initializer = 'random_uniform',
                                trainable=True)
        self.a2 = self.add_weight(shape=[1],
                                initializer = 'random_uniform',
                                trainable=True)

    def call(self,inputs):
        return self.a1*inputs + self.a2*(tf.nn.relu(inputs))

用法：

d = 5
inputs_ffNN = tf.keras.Input(shape=(d,))
x_ffNN = tf.keras.layers.Dense(10)(inputs_ffNN)
x_HTC = Custom_ReLU()(x_ffNN)
outputs_ffNN = tf.keras.layers.Dense(1)(x_HTC)

ffNN = tf.keras.Model(inputs_ffNN, outputs_ffNN)
ffNN.compile('adam', 'mse')

ffNN.fit(np.random.uniform(0,1, (10,5)), np.random.uniform(0,1, 10), epochs=10)

完整示例如下：https://colab.research.google.com/drive/1n4jIsY3qEDvtobofQaUPO3ysUW9bQWjs?usp=sharing

自定义Keras层具有可训练的标量

1 个答案: