3732 Update TTA module based on latest features

[Nic-Ma]

Fixes #3732 .

Description

This PR updated the TTA module with newer MONAI features based on the discussion with @dongyang0122 and @holgerroth .

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 --disttests.
• In-line docstrings updated.
• Documentation updated, tested make html command in the docs/ folder.

[Nic-Ma]

/black

[Nic-Ma]

/build

[Nic-Ma]

/build

[Nic-Ma]

/build

[Nic-Ma]

/build

[Nic-Ma]

/build

[rijobro]

Hi @Nic-Ma, I've created a PR to add to your PR: Nic-Ma#362.

1. Improve tests
2. Reduce data type conversion.

The test currently fails, I will try to debug on Monday, but I feel it is important that it works before we merge. We seem to get different results for:

1. numpy and torch inputs, and
2. using the post_func instead of inferrer_fn=post_trans(model(x)).

[Nic-Ma]

Hi @rijobro ,

Plan sounds good to me!

Thanks for your help here.

[wyli]

Thanks, it looks good to me.

#3733 (comment)

[rijobro]

@Nic-Ma my PR is ready: Nic-Ma#362.

Think you will need to merge it for the unit tests to work.

[Nic-Ma]

@Nic-Ma my PR is ready: Nic-Ma#362.

Think you will need to merge it for the unit tests to work.

Hi @rijobro ,

Thanks for your enhancement, started CI tests in your PR and put some minor comments in it.

[Nic-Ma]

/black

[Nic-Ma]

/build

[wyli]

/build

[Nic-Ma]

/build

[wyli]

Thanks, it looks good to me. Minor issues (not introduced by this PR) are 1. Runtime error could be a warning to allow for randomness defined in the network such as dropout. 2 np.std has a ddof parameters but this is currently hardcoded cc @rijobro

[Nic-Ma]

Hi @wyli ,

Thanks for your review.

1. I changed RuntimeError to Warning and updated the unit tests in the latest commit.
2. About the std arg, PyTorch also has another arg unbiased:
   https://pytorch.org/docs/stable/generated/torch.std.html
   I think maybe we don't need to add these additional args so far? If we got some requests, let's add it then?

What do you think?

Thanks.

[wyli]

/build

[wyli]

/build