In Encog framework, can FinishTraining() method be called repeatedly before saving it to disk?

Question

Consider the senario when a neural-network is being trained on a large data set over a long period of time. Is it correct to call FinishTraining() method before temporarily saving the neural-network to disk for training later? Or should we call the FinishTraining() only once, after all the training has completed at the very end?

Encog machine learning framework has the option to save a neural-network to disk and continue the training later using the TriningContinuation class in Encog. Does calling the FinishTraining() method before saving the neural-network to disk negatively affect the training of the neural-network in any way?

Answer 1

Yes it can. The main purpose of the FinishTraining method is to flush any temporary updates to the weights. You could call it at the end of each iteration, without any harm except performance. Many of the trainers in Encog do not even implement the finishTraining method, as they do not have need of it.