I am facing a weird problem. There is something wrong with my validation block that I am unable to figure out. The convergence is much slower and converges to a much higher metric than if I remove the validation block entirely. Anyone has any idea why?
I would expect a slowdown in the script, if the validation loop is used since it has to be executed.
The latter issue might point towards a data leak. Did you make sure that the training and validation datasets are separate?
I’m also unsure which package you are using with the Trainer class, so I cannot comment on its internals and if they might cause some issues.
Hi @ptrblck. Thank you for the prompt response. My validation dataset is the same as testset currently as I am not tuning any hyperparameters using it. Also, I don’t understand what you mean by the package with the trainer class. I am not using anything different than what is normally used if that helps. Also, the Trainer class is a bit messy and hard to understand. Hence, haven’t posted it here.