Best way to add a validation process

[albas99]

I'm following the text course and using it as a reference here and there for my projects but something I noticed is there's no mention of validation anywhere in the course which at this point is essential in every machine learning/deep learning process.

My question is what's the best way/guideline on how to add validation to the process? Is it best to lump it into the training function or should it be its own separate function that should be called during the loop after the training function? i.e:

for epoch in epochs(N):
    train(...)
    validate(...)

I'm a bit confused about which one to go with and would love some clarification on this.

[LuluW8071]

There is reference for validation test as well in the course at later chapters. And yes it should be separate function and call them later like you proposed. Its the same approach for test or evaluation process as well.

for epoch in epochs(N):
    train(...)
    validate(...)

[mrdbourke]

Hey @albas99 , @LuluW8071 is right.

You can validate within the training loop or after a model has trained.

For example, I validate a model during training as well as after training via manual inspection (e.g. I will run the model across samples manually to see how it performs).

[albas99]

Got it, thanks for the clarification.