Add parallelization option to CorrelationDecoder and CorrelationDistributionDecoder

[JulioAPeraza]

Closes None.

Currently, CorrelationDecoder takes a long time to be trained. Depending on the number of studies per feature, it takes around 13 min per feature (i.e., per iteration). For large databases like Neurosynth, which has 3228 features, it will take ~29 days. Adding the parallelization option reduces the computation time by a factor of n_cores.

Changes proposed in this pull request:

• Add n_cores parameter to use for parallelization.

[codecov]

Codecov Report

Merging #738 (79a8c95) into main (87c3ce3) will increase coverage by 0.01%.
The diff coverage is 93.75%.

@@            Coverage Diff             @@
##             main     #738      +/-   ##
==========================================
+ Coverage   88.60%   88.61%   +0.01%     
==========================================
  Files          40       40              
  Lines        4449     4455       +6     
==========================================
+ Hits         3942     3948       +6     
  Misses        507      507              

Impacted Files	Coverage Δ
nimare/decode/continuous.py	94.79% <93.54%> (+0.34%)	⬆️
nimare/decode/base.py	91.11% <100.00%> (ø)

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

[tsalo]

LGTM!

[JulioAPeraza]

I would like to run another test before we merge this. The attribute self.images_ was returned empty in one of the tests I ran yesterday.

[tsalo]

Actually, could you also fix the type at:

NiMARE/nimare/decode/base.py

Line 80 in 245b844

LGR.info(f"Retaining {len(features)}/({n_features_orig} features.")

There's an extra opening parenthesis that should be removed, but I don't think it's worth a separate PR.

[JulioAPeraza]

Sure!
Could we also change the vocab in the fetch_neuroquery example to neuroquery6308 in this PR?

[tsalo]

That would be great! Thanks!

[tsalo]

I don't know why the linter only started complaining in this PR, given that you didn't touch the problematic code, but it definitely is correct. The flagged line is definitely too long:

NiMARE/nimare/tests/utils.py

Line 26 in 87c3ce3

"""Create complementary binary images to identify areas of likely significance and nonsignificance.

[JulioAPeraza]

I tried splitting the summary line into two lines and it also failed the Style check test (see the last commit). Do you think we should shorten the summary sentence?

[tsalo]

That's correct. Summary sentences need to fit on one line. I think you can just drop the word "complementary" and it will fit.

[JulioAPeraza]

I think this is ready to merge. I ran a few tests locally and self.images_ now contains the meta-analytic maps.

[tsalo]

LGTM! Feel free to merge.