Train / validation split -- does it need to be left as-is or can we utilize the data as we like?

#1

In particular, are we allowed to train on the whole (train, validation) dataset?