Allow utils._subsampling to accept ndarray

[notZaki]

The current subsampling function only works for 3d arrays. This PR allows it to work for arbitrary number of dimensions and brings us closer to supporting 2D data as inputs (#5).

Notes:

• The function returns a view, so it is possible to mutate the input data
  • It doesn't look like the current code is doing any mutating/in-place operations, but we could return a copy to be safe

[tsalo]

The subsampling is performed across voxels though, right? So would subsampling a flattened SxT array really be equivalent to subsampling an XxYxZxT array?

[notZaki]

Subsampling a flattened array would not be equal to subsampling the original array. This is why the PR doesn't use flattening.

[eurunuela]

I would like to check how the original code does the subsampling for 2D. The original GIFT code supports 1D, 2D and 3D data for maPCA. However, we only took the 3D part for tedana. Maybe we should translate the 1D and 2D options directly from GIFT and then work on optimizing the code.

What do you guys think?

[notZaki]

The original GIFT implementation has additional if-statements to support 1D and 2D data: https://github.com/trendscenter/gift/blob/8f22c6078071f93bd5972ae37df3742f17631e9f/GroupICATv4.0c/icatb/icatb_estimate_dimension.m#L379-L396
In modern matlab, the built-in downsample could've been used as an alternative.

I'm not sure what the advantage would be of translating the GIFT code vs using the proposed slices.

[eurunuela]

I'm not sure what the advantage would be of translating the GIFT code vs using the proposed slices.

The slices approach you proposed is much simpler imo. We could check if both approaches do the same and then stick to the slices one for simplicity.

[codecov-io]

Codecov Report

Merging #20 (95ce55a) into main (67ad715) will decrease coverage by 0.08%.
The diff coverage is 100.00%.

@@            Coverage Diff             @@
##             main      #20      +/-   ##
==========================================
- Coverage   92.09%   92.01%   -0.09%     
==========================================
  Files           3        3              
  Lines         291      288       -3     
==========================================
- Hits          268      265       -3     
  Misses         23       23              

Impacted Files	Coverage Δ
mapca/utils.py	98.57% <100.00%> (-0.03%)	⬇️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 67ad715...95ce55a. Read the comment docs.

[eurunuela]

Thank you @notZaki !

[eurunuela]

@tsalo are you okay with merging this PR?

[tsalo]

Sorry for the massive delay! Merging now.