Proposed rwalk change

• Don't add the initial point to the chain (prevents bad estimated of the act)
• Break strictly after maxmcmc and return a random draw from the prior

Closes #477 (closed)

added 1 commit

• e33b1a4e - Break if maxmcmc reached

Compare with previous version

added 9 commits

• e33b1a4e...8bc78e9c - 8 commits from branch master
• bfb7f21c - Merge branch 'master' into proposed-rwalk

Compare with previous version

unmarked as a Work In Progress

I've removed the WIP as I think this is now working well. To test it, I've run on GW150914 for the GWTC1 setup.

Here is the stats plot from @isobel.romero-shaw origin run (using 0.6.7: (CLEAN) 1591b646))

Meanwhile this is the stats plot for this MR

Note that there are fewer large peaks where the sampler bottlenecks. The run times were similar and the number of calls were similar. However, in cases where the sampler can get "stuck" often, this MR will significantly improve performance.

To check that it is "safe" - in the sense of not introducing a bias. Here is a summary page comparing the LI rejection samples (latest), the raw Bilby output (labelled Bilby_Isobel), the downsamples Bilby data used in the paper (Biby_dat), and this MR Bilby-MR782-MR774

• The max JS test against Isobel's original run was 0.00022 - i.e. this seems to be producing very similar results to Isobels' run.
• The max JS test against the LI samples was 0.00213 for mass_2, however against the bilby samples the same JS test gives 0.00021 so I'm not sure what is going on there.

approved this merge request

added Sampling label

I ran on GW151226

• Summary page

Example of the run image from my run: Example of the run image from @isobel.romero-shaw original run:

added 47 commits

• bfb7f21c...03b41f8c - 45 commits from branch master
• 32f737bb - Merge branch 'master' into proposed-rwalk
• 22a69e5a - Merge branch 'proposed-rwalk' of git.ligo.org:Monash/tupak into proposed-rwalk

Compare with previous version

approved this merge request

added 2 commits

• 6fcc497f - Merge branch 'master' into proposed-rwalk
• b1f48230 - Merge branch 'proposed-rwalk' of git.ligo.org:lscsoft/bilby into proposed-rwalk

Compare with previous version

Update: the summary page above was comparing a time marginalized run against Isobel's run which is not time marginalized. Here is the better comparison (both bilby runs are not time marginalized

https://ldas-jobs.ligo.caltech.edu/~gregory.ashton/bilby_review/bilby0.6.7-MR782-MR774-ec5973ee-CLEAN_bilby_pipe0.3.11-b97834f-CLEAN/GW151226/comparison/home.html

added 1 commit

• 1d0de624 - Add lines to ensure termination after nact * act steps

Compare with previous version

approved this merge request

approved this merge request

merged

mentioned in commit 09bf4887

changed milestone to %0.6.9

mentioned in merge request parallel_bilby!55 (merged)