ImageClassifier workflow : gen_valid uses transform instead of transform_test

paris-saclay-cds / ramp-workflow

Toolkit for building predictive workflows on top of pydata (pandas, scikit-learn, pytorch, keras, etc.).

https://paris-saclay-cds.github.io/ramp-docs/

BSD 3-Clause "New" or "Revised" License

68 stars 42 forks source link

Open victor-estrade opened 7 years ago

victor-estrade commented 7 years ago

Hi,

I noticed that in the BatchGeneratorBuilder class the gen_valid comes from the same _get_generator function as the gen_train.

This means that the valid set of images will go through the same preprocessing/augmentation as the training images.

Is it the wanted behaviour ? Or should it use the transform_test_img ?

jorisvandenbossche commented 7 years ago

transform_test_img is currently only used for the testing data, and not for validation data. But we could use the same approach there?