CheckPoint doesn't record the current epoch.

[TheAutumnOfRice]

If I set a CheckPoint callback and a max_epochs parameter, the max_epochs will not be the true MAX epochs, but checkpoint's epoch PLUS max_epochs.

For example, if I set the max_epochs=400 and load the checkpoint at epoch=125, the fit-loop will start at epoch=125 and end at epoch=526, which is not expected.

It is noticed that the function NeuralNet.fit_loop has a parameter called epochs, the default value is None, namely, max_epochs. So the training will be looped for max_epochs times no matter what the current epoch is. I think it might be better if the default value is set to:

max_epochs-net.history[-1,'epoch']

Or, modify the for-loop in the fit_loop function:

# net.py, line 786
# for _ in range(epochs):
for _ in range(self.history[-1,'epoch'],epochs):

Version: 0.10.0

[ottonemo]

Thanks for the report. I agree that the current behavior of max_epochs is confusing in this regard. I think this is very similar to #674

If it is all the same to you, I would rather continue the discussion in the thread #674?

[TheAutumnOfRice]

@ottonemo Thanks for your reply! I didn't notice that thread. Yes it is the same problem.