Question concerning fix and single dataset types


could someone please clarify the purpose of the single_ds and fix_ds?

I’m not able to find explanations in the documentation.

From trying to read the code, I do the conclusion, that single_ds is used for reconstructing item from tensor back, and basically it is a validation dataset with batch_size equal to 1.

For fix_ds, it looks as well as a copy of validation dataset (or a validation ds), without the transformation being applied to it.

It my understanding correct?

Many thanks in advance for the clarification

Those are just internals. single_ds is where we put single items when doing predictions on an element and fix_ds is the training set with validation transforms.

Thank you for the clarification. Are there any plans to update this in the documentation?

Any PR is accepted. Like I said, those are internals, which is why they aren’t documented currently.