WIP: Added information gain property to result object.

Added to dynesty, nestle, and ultranest samplers, using their individual outputs. Need to still add pymultinest, polychord, and cpnest.

added 1 commit

• ba590df5 - Added information gain to cpnest sampler's result object

Compare with previous version

Pymultinest and polychord do not output the information gain, nor can I find how/whether it is calculated within the algorithm. Polychord (https://arxiv.org/pdf/1506.00171.pdf, appendix B) does not seem to calculate logZerror using the information gain.

Do I leave those samplers without an information gain output, or go with the approximation logZerror = sqrt(h/nlive)? Ultranest directly uses this conversion.

added 1 commit

• 1c919c85 - Only adds ultranest information gain if num_live_points is designated. Will...

Compare with previous version

added 1 commit

• 57f03df9 - Fixed ultranest bug

Compare with previous version

I'm okay with there being a fall-back for samplers which don't produce this!

Is it possible to run the gaussian_example.py with a few different samplers and see how well their information gains agree?

added Result/output label

changed milestone to %1.0.4

approved this merge request

changed milestone to %1.1.0

mentioned in merge request !907 (merged)

Closing because I had to submit a new MR !907 (merged) from my fork.

closed