Stopping and resuming training

[Sanne de Roever]

Evgeny, first of all thanks for creating neurolab. I just got my feet wet with NN and am enjoying your work.

I've got some questions regarding stopping and resuming training.

During stats315b at Stanford I learned that reducing the number of epochs can be seen as a form of regularization. What I would like to do is to train a net on a test set for say 10 epochs, then sim on a validation set, and then train for another 10 epochs, the sim on a validation set. After a number of epochs the error on the validation will likely rise: that is the point were I should stop training. Could I accomplish this with neurolab?

[Sanne de Roever]

I forgot to mention that in this case one starts with very small parameters to start with. On each epoch the parameters grow, hence the regularization effect.

[Sanne de Roever]

I working on making this possible. Next to the epoch parameter I'm introducing an interrupt parameter. This parameter interrupts the training so that it can be resumed.