How to perform one round of inference before the actual training starts?

[avijit9]

Hi,

I would like to perform one full round of inference before the training starts on the training dataset. What is the easiest and optimal way to do that?

TIA.

[senarvi]

I think the way to do that is to run Trainer.test() before you start training. I would actually like the possibility of performing validation also when training starts, so that when I'm doing transfer learning, I would see the initial state in the validation curves. However, I think you have to perform a separate inference round.