add verbose to dde.model.train and dde.model.compile

[KangyuWeng]

Purpose

This pull request aims to add a verbose parameter to the dde.model.train() and dde.model.compile() methods. This addition allows users to control the verbosity of these methods, which is important in federated learning.

Changes Made

1. Set the optional values of verbose to be 0 or 1.

• This is the same with the verbose in other parts of deepxde and make it possible for future extensions of adding verbose=2.

2. Added the verbose parameter to train() and compile().

• The verbose parameter is added to both dde.model.train() and dde.model.compile().
• Default is verbose=1, without any changes to the original methods.
• Setting verbose=0 will silent all outputs.

3. Read verbose parameter in the decorator utils.timing.

• When verbose=0 is passed to utils.timing, it will no longer print the timing information.
• Default is verbose=1 which prevents affecting other methods using the decorator utils.timing.

4. Added verbose parameter in model._test()

• When verbose=0, there should be no output, including test loss. So, display.training_display() in model._test() is disabled when verbose = 0.

5. Disable display.training_display() in TensorFlow v1 compatibility mode.

• This is modified to be disabled when verbose=0.

Test

Example used: https://deepxde.readthedocs.io/en/latest/demos/pinn_forward/ode.system.html.

Backend tested: pytorch, jax, paddle.

Problem: The new verbose feature works well overall. But there are thousands of output like "I1106 13:10:16.057708 3995094848 kernel_dispatch.h:102] Get BackendSet from tensor" when using paddle as the backend. This problem also occurs when using the original deepxde, before any changes.

Notes

Maybe we could disable timeit.default_timer() in the utils.timing() when verbose=0.