Fix the multinest temporary file transfer

Previously, when running with multinest on condor, there wasn't enough time given between the eviction signal and when the job was getting kicked out for the exit sequence to complete successfully. This meant that because of the link between the temporary directory and the proper path, all progress was being lost. This MR introduces a dumper function that copies over the output from the temporary directory to the proper path every time the output is updated. I have also added a method to track the sampling time across restarts. I have tested this both locally and on condor, and it works as expected.

added Sampling label

mentioned in merge request !798 (merged)

added 2 commits

• 15c47974 - Fix flake8 errors
• 46d3ee27 - Fix tests

Compare with previous version

added 1 commit

• c8604c65 - Fix new flake8 issues

Compare with previous version

approved this merge request

approved this merge request

merged

mentioned in commit abce0f82

@sylvia.biscoveanu for the sampling time tracking, shouldn't the calculate_and_save_sampling_time() method reset the start time to

self.start_time = current_time

at the end, otherwise the total sampling time produces a cumulative total of intervals all based on the original start time. This isn't a problem for standard resumes as the start time it reset in those cases, but for the cases when the dumper is called and the time file is rewritten, this can increase the output run time by a lot.

I found this when trying the implementation in ultranest, where it's equivalent of the dumper function was called a lot more than multinest's one is, so for something that ran for about 40 seconds the total sampling time was coming out at ~1400 seconds (the dumper function was called about 60 times).

mentioned in commit ad8a0d68

changed milestone to %1.0.0