Backend for moves

[Zebedee Nicholls]

In some cases it is helpful to have the move change as the chain is being run (e.g. in their original paper, Nelson et al. discuss how they adapt gamma at each step in Section 4.1.2 https://arxiv.org/pdf/1311.5229.pdf). 

Would there be any interest in adding a backend or similar to the moves class so that it is possible to save this move state over time and reload from disk (in e.g. the case of a node crashing)? I would be happy to sketch out a PR, but wanted to gauge interest here before doing so.

I should also point out that I'm no expert in MCMC so this may be a terrible idea. 

Also, massive props to all the developers for their efforts. You have built an amazing tool and community!

[David W Hogg]

My understanding is that it is hard or even impossible to adjust the emcee move adaptively and still respect the requirement of detailed balance. It’s permitted in a burn-in phase, but the requirements on the final sampling used for inference or probability calculations are pretty strong.

--
You received this message because you are subscribed to the Google Groups "emcee users" group.
To unsubscribe from this group and stop receiving emails from it, send an email to emcee-users...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/emcee-users/338d2876-4559-45ed-883a-c85122992c46n%40googlegroups.com.

--

David W. Hogg — http://cosmo.nyu.edu/hogg/