Add catch for empty log_evidence_err array

I ran into an issue when trying to use the ResultList object where all the log evidence errors in my individual runs were infinite so I wasn't able to use the combine() method. I've just added a catch that if this array is empty, return nan as the new result.log_evidence_err. I could also see the value in returning the standard deviation of the evidences from the individual runs. What do the rest of you think?

Looks good! I'd be open to using the std. dev. when there aren't errors available for the individual runs. Although I don't feel strongly.

changed milestone to %0.5.2

approved this merge request

approved this merge request

I'd be happy for the standard deviation to be added too.

It looks like @sylvia.biscoveanu it is your call if you want to add the std dev :) (or just merge)

added 19 commits

• 68707a93...b383e97b - 17 commits from branch lscsoft:master
• 52669a34 - add standard deviation of evidence array
• 77906f27 - Merge remote-tracking branch 'lscsoft/master' into combined_result_log_err

Compare with previous version

added 1 commit

• 1aede781 - Undo standard deviation

Compare with previous version

Ok, I decided to keep the np.nan because I realized that the standard deviation should probably be the np.log(np.std(evidences)) not np.std(log_evidences), and I'm worried about things blowing up if I just do np.log(np.std(np.exp(log_evidences))), and I don't think I can use logsumexp because I have to subtract the mean evidence and then square the difference, and I can't put that whole term in the exp part... Anyway I think this is ready to be merged now. Pardon my digression

approved this merge request

approved this merge request

approved this merge request

merged

mentioned in commit bfd7510e