When training the model, it is possible to set the "validation_split". How does he work? Is it possible to split the sample and use not all data for train
