Unifying Pre- and Postprocessing

[FynnBe]

No description provided.

[constantinpape]

The problem is that the reference_tensor in scale_range and scale_mean_variance only makes sense in post-processing. So it is a bit awkward to unify pre-and-postprocessing and then have special cases for it in each function.

[constantinpape]

@FynnBe let me know when you have made all the changes and I should have a look again.

[FynnBe]

@FynnBe let me know when you have made all the changes and I should have a look again.

I think it's all up to date. I left binarize and sigmoid in the common processing's, as they might not be as useful for the input, but wouldn't have any other kwargs...

[constantinpape]

I think you forgot scale_mean_variance in preprocessing.

[constantinpape]

Also refere to the preprocessing here.

[FynnBe]

should scale_mean_variance exist for preprocessing? It is different from the zero_mean_unit_variance.

[constantinpape]

I think for scale_mean_variance we only need to know the reference tensor, correct?
Then it is all correct like it is right now, because we actually don't need to know any other kwargs here.

[FynnBe]

I think you forgot scale_mean_variance in preprocessing.

zero_mean_unit_variance is in the Pre- and Postprocessing section...

[constantinpape]

zero_mean_unit_variance is in the Pre- and Postprocessing section...

But then same as preprocessing in the kwargs is ambiguous and we need to refer to zero_mean_unit_variance there.

[FynnBe]

But then same as preprocessing in the kwargs is ambiguous and we need to refer to zero_mean_unit_variance there.

same as preprocessing is now only specified for scale_range

[FynnBe]

actually for scale_mean_variance a mode=fixed would also makes sense, but one can achieve the same with zero_mean_unit_variance + scale_linear. Therefore I would leave it at that for now.

[constantinpape]

Ok, I think I understand it know and it should all be correct:
For scale_mean_variance we only need to know the reference_tensor as kwarg, so we actually don't need to refer to any other kwargs.

[constantinpape]

I think for scale_mean_variance we only need to know the reference tensor, correct?
Then it is all correct like it is right now, because we actually don't need to know any other kwargs here.

[FynnBe]

Ok, I think I understand it know and it should all be correct:
For scale_mean_variance we only need to know the reference_tensor as kwarg, so we actually don't need to refer to any other kwargs.

The mode (per_sample or per_dataset) is required as well to determine how to determine mean and variance from the reference.

[constantinpape]

The mode (per_sample or per_dataset) is required as well to determine how to determine mean and variance from the reference.

True, but we have that already.
I just wanted to point out that we don't need to refer to any other kwargs.

This is good to go now and I will merge.