Cannot modify max_epochs and rerun fit()

[ferrouswheel]

It'd be nice to be able to run network.fit(X, Y) with a network until network.max_epochs is reached, then change max_epochs to a higher number and call fit again to have training continue.

This would be possible if the condition in train_loop used num_epochs_past as the initial value for epoch rather than 0:

num_epochs_past = len(self.train_history_)
epoch = num_epochs_past # epoch was previously assigned to 0

while epoch < epochs:
    ...

Of course, it was probably intended that calling max_epochs applies independently for each call to fit() and if that's the case it'd be good for future documentation to mention this. In the mean time, they'll be this github issue that people can find by google ;-)

[BenjaminBossan]

Yes, this behavior may be a little bit unintuitive for some but I don't think it needs to be changed.

Btw., you can pass the number of epochs you want to train directly to the fit call instead of setting max_epochs.