search

kimbjerge / MCC-trap

17 stars 6 forks source link

Wrong data generators are used for the validation #1

Open DiKorsch opened 3 years ago

DiKorsch commented 3 years ago

In lines 79 and 192 of hp_param_training.py you use the train_datagen instead of validation_datagen. Same issue is in hp_param_training-artNets.py.

DiKorsch commented 3 years ago

It totally true, but then you apply data augmentations that are intended for the training stage during your evaluation. I think for the validation_generator you need to create it from the val_datagen. It should not change anything in context of subsets, but should raise your evaluation performance, since you would not apply random augmentation during evaluation.

jakobbn commented 3 years ago

I can see what you meant. I will note that.

Thank you :)