search

JohannesBuchner / UltraNest

Fit and compare complex models reliably and rapidly. Advanced nested sampling.
https://johannesbuchner.github.io/UltraNest/
Other
142 stars 30 forks source link

Access chains / intermediary results for run in progess? #111

Closed OGdodge closed 10 months ago

OGdodge commented 10 months ago

Description

Describe what you were trying to get done. Tell us what happened, what went wrong, and what you expected to happen.

Sorry this is more of a question rather than a bug. Apologies if I am missing something obvious, but I've been trying to work this out for a few days. Is it possible to have Ultranest save the chains / results files at a set number of iterations in addition to producing the nicelogger output? I see the points file is produced / updated for runs in progress, but the chains are not.

Ideally I'm aiming to produce intermediate corner / trace plots for the run in progress (either using the inbuilt plotting or anesthetic) as well as take the current best likelihood / a posteriori likelihood point and produce models using it, ie. extract it and run it through a seperate program.

I've tried using run_iter, but haven't got anywhere. I've taken a look at the source code and have noticied I could possible harness _update_result() but I'm not entirely clear on how to use run_iter to trigger something every 100 iterations or so - I tried for iter, result in sampler.run_iter(): if iter % 100 == 0: print('TEST') and similar but couldn't get anything out.

Is this possible and if so what have I missed! Again apologies for a fairly specific request / question.

JohannesBuchner commented 10 months ago

If you only want corner / trace plots, then I think you can copy the code of nicelogger and see how it accesses the live points (points['p'], for example), and run with run(..., viz_callback=mynewfunction)

https://github.com/JohannesBuchner/UltraNest/blob/master/ultranest/viz.py#L85

If you want to not have a callback on every run, keep a counter and skip.

def myfunc():
    myfunc.mycounter += 1
    if myfunc.mycounter % 10 == 0: print("Hello")

myfunc.mycounter = 0
myfunc()
myfunc()
myfunc()
myfunc()

If you really want the full chains rather than the live points, then you can do run and limit the number of iterations with max_iters. API doc: https://johannesbuchner.github.io/UltraNest/ultranest.html#ultranest.integrator.ReactiveNestedSampler You can do that in a loop, successively increasing max_iters, and get the results. Should set max_num_improvement_loops=0.

I am not sure the chains are insightful though, since all of the posterior weight will be concentrated in the maximum likelihood point.