QU fitting running for hours on Modelled data. about rm-tools HOT 3 OPEN

Hi @Jaswanth-Jza,

I think think this is a problem with the underlying multinest library. I've even found it can slow down further with more cores provided to it.

I've got PR #56 open which replaces this library with Bilby for far greater flexibility. It can dynamically switch between different fitting libraries. It uses Dynesty by default. Would you be willing to test this version of QU fitting?

It requires some minor changes to the model files, but allows for much greater flexibility and performance.

Cheers,
Alec

from rm-tools.

Hey @AlecThomson ,

Yes, I would definitely be willing to test the version of QU fitting with Dynesty!

With best wishes,
Jaswanth

from rm-tools.

Just a quick note @Jaswanth-Jza, Bilby allows you to choose a variety of samplers (see bibly docs, so long as you have the required ones installed (dynesty comes by default). I've also tested a bit using UltraNest, which seems to run slightly faster than Dynesty (but that may vary for different models or data). You can chose the sampler with the --sampler argument on the command line. You can even switch back to PyMultinest, if you really want to.

I forgot to mention, I've got examples of the new model files along with the PR #56. Particularly if you're using multi-component models, I'd encourage you to look at m4.py or m11.py. I've added in (thanks to bilby) the constraint that RM1 > RM2 and that frac_p_1 + frac_p_2 < 1. This prevents both multimodal results from labelling degeneracies, and unphysically high polarisation fractions.

from rm-tools.