Some clarification regarding the meaning of n_train, n_val and batch_size

[rdguha1995]

Hey Team,

This might be a pretty basic question. I am training an MLIP using the extxyz data of ~187 structures obtained from AIMD. It is a small molecule in a solvent environment of water.

My question is what does n_train, n_val and batch_size actually mean in terms of the training process. My naive understanding is if n_train is 100 and n_val is 50 with a default batch_size of 5, it will randomly take 100 structures from the training data and batch it into 20 batches of 5. Is that correct?

[Linux-cpp-lisp]

Hi @rdguha1995 ,

Thanks for your interest in our code!

Yes, that is correct, though it is worth mentioning that the splitting into batches is re-randomized at every epoch. The train-validation split is of course only randomized once at the beginning of training in this default case.