Use sparse array in ALE, ALESubtraction, SCALE, KDA, and MKDADensity

[JulioAPeraza]

Closes #692.

Changes proposed in this pull request:

• Create the MA maps for all studies in compute_ale_ma and return a 4D sparse array.
• Return sparse array by ALEKernel._transform and KernelTransformer.transform.
• Add support for ma_maps 4D sparse array in ALE and CBMAEstimator. This will involve adding support for sparse array in ALE._compute_summarystat_est, ALE._determine_histogram_bins, and ALE._compute_null_approximate.

[tsalo]

I noticed a couple of commented-out lines that will to be removed before this is merged. I had a few comments/suggestions as well. Thanks for this!

[tsalo]

I'm not sure how well this generalizes to other masker types, but I've opened an issue requesting a new attribute for masker objects with this info: nilearn/nilearn#3296

[JulioAPeraza]

Having that attribute will be helpful. I think it may save us memory and computation time.

[tsalo]

I'd like to compare this to https://github.com/tsalo/NiMARE/blob/49191a0a4b10a2ac522828c9e558f279d5c6d6ea/nimare/meta/utils.py#L400-L438 before we merge.

[codecov]

Codecov Report

Merging #725 (010827e) into main (ac57801) will increase coverage by 0.48%.
The diff coverage is 98.23%.

@@            Coverage Diff             @@
##             main     #725      +/-   ##
==========================================
+ Coverage   85.34%   85.83%   +0.48%     
==========================================
  Files          41       41              
  Lines        4539     4646     +107     
==========================================
+ Hits         3874     3988     +114     
+ Misses        665      658       -7     

Impacted Files	Coverage Δ
nimare/meta/utils.py	61.95% <95.40%> (+7.65%)	⬆️
nimare/generate.py	100.00% <100.00%> (ø)
nimare/meta/cbma/ale.py	96.71% <100.00%> (+1.78%)	⬆️
nimare/meta/cbma/base.py	97.10% <100.00%> (+1.46%)	⬆️
nimare/meta/cbma/mkda.py	98.78% <100.00%> (+1.69%)	⬆️
nimare/meta/kernel.py	78.80% <100.00%> (-1.32%)	⬇️
nimare/utils.py	94.55% <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 ac57801...010827e. Read the comment docs.

[jdkent]

Looking great! I have a couple places where docstrings should likely be updated, and I'm testing locally to if some of the functions are ever called with np.array since the switch to sparse arrays.

[jdkent]

these lines no longer need to be in the code:

NiMARE/nimare/tests/test_meta_ale.py

Lines 48 to 53 in ac57801

	# If there is a memory limit along with pre-generated images, then we should still see the
	# logger message.
	meta = ale.ALE(kernel__sample_size=20)
	with caplog.at_level(logging.DEBUG, logger="nimare.meta.cbma.base"):
	meta.fit(dset)
	assert "Loading pre-generated MA maps" in caplog.text

since we have removed memory limit

[tsalo]

I think the core changes are ready. There are just a few docstring changes that I think need to be made, as well as a small improvement to one of the CBMA tests (i.e., covering the new masker type check).

[tsalo]

Everything looks good to me, thanks!