Adds a pool to the reconstruction

Adds a pool option to the reconstruction of the marginalized parameters

This also adds a generic npool interface to multithreading which will work with both dynesty and ptemcee current implementations via the equivalent kwargs.

added 1 commit

• 89523f33 - Apply suggestion to bilby/gw/conversion.py

Compare with previous version

added 1 commit

• 21588a2c - Apply suggestion to bilby/gw/conversion.py

Compare with previous version

resolved all threads

added 1 commit

• d376e6e5 - Fix accidental change to the variable ame

Compare with previous version

resolved all threads

I should also add that I ran this on the marginalized_likelihood example with npool=3 and found about a factor 3 speedup in the reconstruction :)

added 1 commit

• 8890eb14 - Revert change from map -> imap

Compare with previous version

@colm.talbot I reverted your proposed change from map to map. I tested the proposal and by itself, it errors out. This is because imap returns a generator. You can remedy this by wrapping it in a list(). However, I noticed in the documentation for imap it says Equivalent of map()-- can be MUCH slower thanPool.map(). So, I did a test run and found that map takes 03:23, imap (with a list wrap) took 05:47. For this reason I think we should keep with the map.

resolved all threads

approved this merge request

mentioned in merge request bilby_pipe!325 (merged)

unapproved this merge request

There are a few docstrings that should be updated to include the new npool argument.

added 1 commit

• de0d44dd - Add doc strings and add npool to cpnest

Compare with previous version

resolved all threads

Okay, I've run a few more involved checks using the marginalized_likelihood.py example. Here is the posterior running with npool=1 and npool=4

run	sampling time	reconstruction time
npool=1	36s	10m 15s
npool=4	14s	2m 18s

And this is the posterior confirming that things get generated correctly.

Finally, I ran with a reference frequency of 35Hz and double-checked that the result.posterior['reference_frequency'] is indeed 35Hz. (This was a bug that afflicted pbilby when we implemented parallelization of the reconstruction).

resolved all threads

added 1 commit

• 2c7b5ce8 - Add stdout to tqdm

Compare with previous version

changed milestone to %0.6.8

approved this merge request

approved this merge request

mentioned in commit 775d682a

merged