0.22.0

What's Changed

This release contains a few different themes of work.
The first theme is alignment with external packages like GingerALE
and the MKDAtoolbox, which resulted in a couple of small changes:

• how statistics are calculated/how null values are thresholded. Instead of doing +0.5 and lowering to the floor, doing typical rounding less biased.
• taking into account sample_size in MKDADensity
• allow for explicit seed arguments for reproducibility of results

There will be more writeups about how the performance of NiMARE compares to other meta-analysis packages.

The second theme is corrections/fixes, big thanks to @SahilKumar75 and @const7 for their contributions!

The third theme is machine learning, where Google Summer of Code student @lyraluoyu implemented NiMARE's machine learning module #1148. Combined with our large studyset releases from neurostore (https://neurostore.org/api/neurostore-studyset-releases/), this opens up a number of interesting questions/comparisons that were previously not testable.

• [FIX] Reject coordinates outside mask image bounds by @SahilKumar75 in #1135
• [FIX] Fix inclusive thresholds in decoder preprocessing by @const7 in #1137
• [FIX] NaN-safe FDR correction, and a one-tailed z for BrainMap forward inference by @jdkent in #1139
• [FIX] Correct the Neurosynth decoder's uniformity test by @jdkent in #1138
• [FIX] ensure traversing slices of a studyset maintains the view by @jdkent in #1140
• [FIX] Build Monte Carlo null bin edges from the bin centres by @jdkent in #1144
• [FIX] Fix ALE histogram binning to use nearest bin instead of floor by @jdkent in #1147
• [FIX] Correct KDA's approximate-null binning by @jdkent in #1145
• [REF] Use _round2 for KDA's approximate-null tie-break by @jdkent in #1150
• [FIX] Round millimetre coordinates to the nearest voxel (MKDA/KDA parity with Neurosynth) by @jdkent in #1142
• [ENH] Add an autodiff fallback for CBMR's closed-form information matrix by @jdkent in #1151
• [ENH] mkda sample size weighting by @jdkent in #1141
• [FIX] Load m2r2's mdinclude extension instead of importing MdInclude by @jdkent in #1156
• [ENH] Allow random processes to be seeded for reproducibility by @jdkent in #1152
• [ENH] Add fetch_neurostore and deprecate fetch_neurosynth by @jdkent in #1160
• [DOC] Make it obvious how to reach the NiMARE team by @jdkent in #1159
• [FIX] Validate Monte Carlo iteration counts by @SahilKumar75 in #1154
• [ENH] 001 ma feature dataset by @jdkent in #1148
• [MAINT] Bump pymare version in setup.cfg by @jdkent in #1163

New Contributors

• @SahilKumar75 made their first contribution in #1135

Full Changelog: 0.21.0...0.22.0