Open rebeccaem opened 10 years ago
Agreed. It wouldn't take a huge effort to put this into the code. The container classes that MH Sampling uses already have methods for writing and reading chain positions from a file (See SequenceOfVectors.{h,C}). Not sure about covariance matrices. The issue is, of course, taking the time to delve into the messy code. Personally, I use multi-level so this isn't an issue for me and I don't want to take the time out to code and test such an effort - but if it is worth your time to implement this, I'm sure we would all appreciate the contribution :) Otherwise, maybe one of the other peeps will get around to it eventually.
Thanks, @eRite . I've talked with @dmcdougall about this, and I think we'll work on it. It would just be really cool to finish a run, and then if you decide you want to get more samples, pick up right where you ended the last time. (I care more about this at the moment than a proper restart capability.)
It would be nice to be able to do two things with MH sampling:
1) When a chain finishes properly, save the current state (position, covariance). Then use this to initialize a future run. 2) Restart the chain in the middle of a run (it seems this is currently available in multi-level).