Pool emcee#66
Conversation
…t works in an interactive node but possibly not after submitted to batch. Needs further investigation. Also added notebook for check of windows in mocks.
andreufont
left a comment
andreufont
left a comment
There was a problem hiding this comment.
Let's talk in person, I'm not sure I understand what is going on here.
|
New version doesn't involve the Sampler class anymore. Let me know if you'd like to update things to use a modified version of the class, or make a new class |
|
Cool, that was fast! I'll take a look at it and try to run it myself. Thanks! |
andreufont
left a comment
andreufont
left a comment
There was a problem hiding this comment.
I think this could work, thanks @marlena6 !
I've added several comments to the code, but mostly for me to understand what we are doing. We can discuss it and try it a bit more before merging, but if you wanted we could also merge it now (it doesn't affect any other code)
|
@marlena6 - I was able to run this at NERSC. I'd still like to discuss the comments I made above, do you mind replying here on GitHub, or discussing in person next week? After that I would merge this PR and open another issue about profiling the sampler at scale (check that the boost is larger than the factor of 5-6 that you mentioned earlier). |
2 participants
Draft pull request to show how emcee with pool is working, it seems to work but messes with the previous Sampler structure so we should discuss how to make both parallelized and non-parallelized options work (two different sampler classes? Optional posterior argument to Sampler class? )