Major refactoring of Diagnostics module

[JulioAPeraza]

This is part of the refactoring issue #772, to reduce some of the duplicated code as was pointed out here #765.

Changes proposed in this pull request:

• Create a new base class Diagnostics, to reduce duplicated code.
• In addition, we plan to change the AttributeError to a Warning when using pairwise Estimators and return only the clusters table as suggested here Add CBMA workflow #761 by @jdkent.

[JulioAPeraza]

@jdkent To avoid the breaking change, do you think it would be a good idea if instead, we return an empty contribution table when the estimator is not supported, similar to what we do when the number of clusters is 0:

NiMARE/nimare/diagnostics.py

Lines 167 to 170 in 08cbfad

	if clusters_table.shape[0] == 0:
	LGR.warning("No clusters found")
	contribution_table = pd.DataFrame(columns=["id"], data=rows)
	return contribution_table, clusters_table, label_maps

In that case, we won't need to have the extra flag return_clusters_table_only.

[codecov]

Codecov Report

Patch coverage: 100.00% and project coverage change: +0.07 🎉

Comparison is base (08cbfad) 88.39% compared to head (758b08f) 88.46%.

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #776      +/-   ##
==========================================
+ Coverage   88.39%   88.46%   +0.07%     
==========================================
  Files          40       40              
  Lines        4791     4761      -30     
==========================================
- Hits         4235     4212      -23     
+ Misses        556      549       -7     

Impacted Files	Coverage Δ
nimare/diagnostics.py	99.15% <100.00%> (+5.23%)	⬆️

... and 2 files with indirect coverage changes

Help us with your feedback. Take ten seconds to tell us how you rate us. Have a feature suggestion? Share it here.

☔ View full report in Codecov by Sentry.
📢 Do you have feedback about the report comment? Let us know in this issue.

[jdkent]

yeah, returning an empty contribution table sounds okay with an explicit warning. Not introducing another argument feels good.

I'm feeling between two options:

• return empty contribution table
• return None

An empty clusters table makes sense because there was nothing incompatible in the code, it was just an attribute of the data, and we wouldn't want to change the output type just because nothing survived the clustering algorithm. The actual answer is that there are no clusters in the table

In this current case, we know which combinations of Estimators/Diagnostics beforehand don't work, so if we return an empty table, are we saying "no studies contributed to any of the clusters"? (No we are not saying that). So I have a slight preference to return None for the contribution table for these predetermined combinations because None signals to me that "we could not make a reasonable output for some reason, but someday in the future a reasonable output could exist here".

[JulioAPeraza]

Thanks!
I've added the suggestions and documented the new changes. I also added a new test to improve coverage.

[jdkent]

Looked it over, changes look good and you test it with one of the pairwise estimators, LGTM.

[JulioAPeraza]

thanks!