saving the state of the network for best validation value

[leconteur]

I want to test the best network I obtained during the fit on new data and with other scoring function. However, the _train_layer function does not seem to keep the best network so far in memory. This prevents me from using the n_stable argument as the network starts to overfit and the value I get is possibly far from the best I could obtain. Is there a way to reset the network to the parameters of the best iteration?

[alexjc]

There's no built-in way to do this yet but it sounds like a great feature!