Skip to content

Add information gain result

This is finishing off !840 (closed), which @kaylee.desoto had started a while back. @gregory.ashton, you had asked to compare the information gain for a few samplers using the gaussian_example. Here are the results:

Information gain
dynesty: 3.3349504450440577
cpnest: 3.3390112702606984
nestle: 3.2969096440032217
ultranest: 3.048505964684068
log Evidence
dynesty: -280.84150357873546
cpnest: -275.9488938702244
nestle: -275.89352893247315
ultranest: -287.5094497652268

They seem to agree relatively well compared to the variation in the log evidence, but I'm not really sure of the level of agreement I should be looking for. What do you think? Otherwise, I have tested that these changes save the information gain in the result object and the result file both for samplers that return this property and for those that don't where it defaults to nan, so I think it's ready to go.

Merge request reports