Validation error is lower than training error

[franva (Winston Fan)]

Hi there,

I am still observing this behavior in Fast AI Version 2.4

And found this post:

Validation error lower than training error

I noticed in lesson 1 generally the validation losses were lower than the training losses. wouldn’t we expect the opposite? shouldn’t the model fit the training data (that it has seen) better than it fits the validation data (which the model hasn’t seen)? Thanks

Is this still the case?

If anyone interested in answering it, I am happy to post more code if needed.

[florianl (Florian)]

sure, that behavior has nothing to do with the fastai version. During training there is augmentation, dropout, etc … active. during validation there is no augmentation and dropout. so it’s easier for the model to give the right predictions during dropout.