在TF2.0中，`tf.keras.Model.compile`有什么作用？

Question

In TF2.0, it is still necessary to call model.compile after creating a keras model, as seen in the doc .

But with eager mode, there is no need for graph creation, so what is the use of tf.keras.Model.compile in Tensorflow 2.0?

Answer 1

model.compile has nothing to do with graph mode, it defines the loss, optimizer, and metrics, same as with keras and tf.keras in TF 1.x. The graph is just an internal detail that is not really related to model.compile .