I am using OpenNMT-py v.1.0.0.rc1 image-to-markup model, and as per the documentation we need to provide train, test and validation dataset. But When I am looking in the model, I am seeing any extra step taken to tune hyperparameters (beam size, dropout, Adam’s parameters, etc.) like grid search, random forest, etc. May I request you to tell me why we are specifically providing separate validation dataset?
Thank you in advance!