[Simplified API] When accuracy reaches a threshold, user should be able to terminate the training process

[daming-lu]

Right now in the event_handler, even accuracy reaches a certain threshold, or loss drops under a threshold, the training process still keeps running. Is there a way to let the user terminate training process earlier?

https://github.com/PaddlePaddle/Paddle/blob/develop/python/paddle/fluid/tests/book/high-level-api/fit_a_line/test_fit_a_line.py#L72-L75

[daming-lu]

I wonder if we can add a returned signal here from event_handler so that it can tell trainer to stop.

[jacquesqiao]

I think Trainer can have an interface Trainer.stop(), and in event_handle we can use it to stop the training process. What do you think?