[ENH, FIX] PCA variance enhancements and consistency improvements

[handwerkerd]

Closes #876.

Changes proposed in this pull request:

• We changed AIC to the default PCA selection criterion in Made AIC the default maPCA option #849 for the API. I noticed it was still defaulting to MDL for the command line interface, so I fixed that
• As discussed in understanding variance and normalized variance explained #876, normalized variance explained was calculated one way for mdl, kic, & aic, and a different way for the low_mem option or if a specific variance explained as selected. Everything now matches the calculation used for mdl, kic, & aic
• The sum of the normalized variances across PCA components is the % variance of the original data that's explained by the PCA according to the PCA function in sklearn. I am now including this value in the info logger. It doesn't match the percent variance explained values that are given as later outputs. Figuring out how to make sense of the difference is a good future goal.
• The PCA function from sklearn has an option to input a pre-specified number of components as an integer. We were unnecessarily blocking off that option (which I wanted to use for testing some stuff). I updated check_tedpca_value to allow integers >=1 and added unit tests for valid and invalid integers.
• doc strings and logging messages were edited in response to the above changes.

Notes:

• Just to this is cleanly documented, the current version of the code calculated varex_norm = varex / varex.sum() for the pre-defined variance-explained option and also for the low-mem option. It uses varex_norm = ppca.explained_variance_ratio_ for mdl, kic, & aic.
• The testing coverage drops slightly because only integration testing covers decomposition/pca.py and tedana.py We can add another integration test for when a fixed integer number of components are predefined, but since that would non-trivially increase the duration of the tests, I didn't want to do that without some discussion. Adding direct tests for pca.py and modularizing tedana.py more would be a better approach, but well beyond the scope of this PR.

[codecov]

Codecov Report

Merging #877 (65645fc) into main (d4406e4) will decrease coverage by 0.05%.
The diff coverage is 77.27%.

@@            Coverage Diff             @@
##             main     #877      +/-   ##
==========================================
- Coverage   93.15%   93.09%   -0.06%     
==========================================
  Files          27       27              
  Lines        2234     2245      +11     
==========================================
+ Hits         2081     2090       +9     
- Misses        153      155       +2     

Impacted Files	Coverage Δ
tedana/workflows/tedana.py	90.00% <ø> (ø)
tedana/decomposition/pca.py	85.26% <61.53%> (-1.26%)	⬇️
tedana/workflows/parser_utils.py	95.83% <100.00%> (+1.09%)	⬆️

---

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 d4406e4...65645fc. Read the comment docs.

[jbteves]

One minor typo, and one place where I'd prefer you use the word instead of the symbol because my brain keeps reading it as a bitwise and since this is code. Otherwise LGTM.

[jbteves]

LGTM

[eurunuela]

LGTM! Thank you @handwerkerd !