Closed pzhren closed 5 years ago
Hi, thanks for asking. I didn't experience any nan problem. Maybe you have a high learning rate.
The metric function I got it from here https://github.com/gsig/charades-algorithms/blob/master/pytorch/utils/map.py
Hi, Please note that for testing Charades, you only evaluate the entire test set, at the end of each training epoch. So, you can modify the code and define a new callback (for testing at the end of each epoch) and pass it to model.fit_generator.
Nour
How to solve the nan problem in the test