Minor refactors

[Red-Portal]

This PR adds stereographic slice sampling by ¹. This sampler is a multivariate slice sampler and does not operate over an extended state-space, which means it is properly reversible with respect to the model parameters. Therefore, this can be used as a Gibbs sampler component. This complements GibbsPolarSlice, which works well on heavy-tailed multivariate targets but operates on an extended state space. Some preliminary experiments suggest that GibbsPolarSlice works modestly better than StereographicSlice. However, the performance could be improved in case the target distribution is conditioned and re-centered through adaptation as done in the original paper¹.

After some simple empirical evaluation, it seems like using RandPermGibbs with univariate slice samplers still work much better as a component Gibbs sampler. I decided not to add stereographic slice sampling.

Minor refactors,

Footnotes

1. Bell, Cameron, Krzystof Łatuszyński, and Gareth O. Roberts. "Adaptive stereographic MCMC." arXiv preprint arXiv:2408.11780 (2024). ↩ ↩²

[codecov]

Codecov Report

Attention: Patch coverage is 0% with 57 lines in your changes missing coverage. Please review.

Project coverage is 81.02%. Comparing base (233eeaa) to head (97f6e4a).

Files with missing lines	Patch %	Lines
src/multivariate/stereographic.jl	0.00%	57 Missing ⚠️

❗ There is a different number of reports uploaded between BASE (233eeaa) and HEAD (97f6e4a). Click for more details.

HEAD has 1 upload less than BASE

Flag	BASE (233eeaa)	HEAD (97f6e4a)
	4	3

Additional details and impacted files

@@             Coverage Diff             @@
##             main      #25       +/-   ##
===========================================
- Coverage   94.06%   81.02%   -13.05%     
===========================================
  Files          10       11        +1     
  Lines         354      411       +57     
===========================================
  Hits          333      333               
- Misses         21       78       +57     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.