问题
I'm using Tensorflow 1.4 Estimator and Dataset APIs for distributed training in Google Cloud Platform. I want to implement early stopping to prevent overfitting during the training, and looked at early stopping hooks documentation below:
- https://www.tensorflow.org/api_docs/python/tf/estimator/experimental/make_early_stopping_hook https://www.tensorflow.org/api_docs/python/tf/estimator/experimental/stop_if_no_decrease_hook
But, none of these hooks support distributed training; so the question stands:
How do I implement early stopping with Estimator API that also supports distributed training? If that's not possible, what do you recommend? How do I prevent overfitting?
There is a similar question [here], but that doesn't address the distributed training concern.
来源：https://stackoverflow.com/questions/57776215/how-do-i-implement-early-stopping-with-estimator-api-for-distributed-training