2648 Enhance RandLambda to execute deterministic transforms for random part of dataset

[Nic-Ma]

part of #2648 .

Description

This PR enhanced the RandLambdad transform and also added array version RandLambda transform to execute transform for random part of dataset.

Status

Ready

Types of changes

• Non-breaking change (fix or new feature that would not break existing functionality).
• Breaking change (fix or new feature that would cause existing functionality to change).
• New tests added to cover the changes.
• Integration tests passed locally by running ./runtests.sh -f -u --net --coverage.
• Quick tests passed locally by running ./runtests.sh --quick --unittests.
• In-line docstrings updated.
• Documentation updated, tested make html command in the docs/ folder.

[rijobro]

Doesn't this imply that we should just implement RandResize, RandCenterSpatialCropd and ThresholdIntensityd?

[Nic-Ma]

Hi @rijobro ,

The basic idea of this RandCompose is to randomly execute transforms on the data, it's random prob "YES or NO", not like our other random size / random value transforms.
And with this RandCompose, no need to implement the "random" version transform for every deterministic transform anymore.

Thanks.

[wyli]

My question is why this implemented as a special compose method, in the example prob=0.2 for Resized, prob=0.3 for CenterSpatialCropd, and prob0.4 for ThresholdIntensityd are all independent events. Instead, I think each of them could be implemented with a transform wrapper inheriting the RandomizableTransform interface https://github.com/Project-MONAI/MONAI/blob/dev/monai/transforms/transform.py#L236

[Nic-Ma]

Hi @wyli ,

Thanks for your suggestion.
Actually, at the beginning, I planned to enhance the RandLambdad transform to support prob, but I feel maybe using RandCompose can help simplify a list of wrapper of transforms.
Anyway, if you guys prefer to wrapper, let me change back to enhance the RandLambdad.

Thanks.

[rijobro]

I think enhancing RandLambda would be more appropriate. Even though most users will combine transforms in a chain, we shouldn't assume that they will all of the time.

[Nic-Ma]

Hi @rijobro and @wyli ,

I have updated this PR to enhance RandLambdad and also added the array version RandLambda transform.
Could you please help review it again?

Thanks in advance.

[wyli]

Thanks, would it be useful to make this invertible?

[wyli]

Was partly discussed here #2004

[Nic-Ma]

/black

[Nic-Ma]

Hi @wyli @rijobro ,

I added inverse operation to the PR, could you please help review it again?

Thanks in advance.

[Nic-Ma]

/black