自定义层的实现

Implementing custom layers

A recommended approach to creating your custom layer is to extend the tf.keras.Layer class and implement:

• In init, handle all input-agnostic initialization.
  • In build, when you are aware of the shapes of input tensors, perform the remaining initialization steps.
  • In call, perform forward computation.

Please note that you don't need to wait until the build function is executed before setting up your variables; you can also initialize them during the init phase. One key advantage of performing variable setup within the build function is that it a