fit with validation data

[GuillermoJaca]

Feature Description

In the current implementation, we can only retrain the best-selected model in validation data if we give only train data and specify a percentage for validation. However, it would be nice to be able to retrain the best model in the validation data when we pass it as an additional argument.

I think it shouldn't a big change