search

JohannesBuchner / PyMultiNest

Pythonic Bayesian inference and visualization for the MultiNest Nested Sampling Algorithm and PyCuba's cubature algorithms.
http://johannesbuchner.github.io/PyMultiNest/
Other
191 stars 87 forks source link

Importance nested sampling and multimodal should be **False** by default #220

Open AdamOrmondroyd opened 1 year ago

AdamOrmondroyd commented 1 year ago

I was finding that changing multimodal was making no difference to the output. MultiNest cannot have both importance nested sampling and multimodal both turned on, so if both are True, then multimodal is changed to False. The default, multimodal=True and importance_nested_sampling=True, is therefore incompatible with MultiNest. To match MultiNest they should both default to False.

https://github.com/farhanferoz/MultiNest/blob/cc616c17ed440fbff8d35937997c86256ed4d744/MultiNest_v3.12/nested.F90#L168-L170

AdamOrmondroyd commented 1 year ago

I just remembered this PR. I understand that this is a breaking change, but it does not work as is. Perhaps it would be better to raise a warning if both INS and mmodal are turned on?

JohannesBuchner commented 1 year ago

if IS=1, I expect MultiNest still does multiple ellipsoids, or not? I guess multimodal refers to splitting the run into several independent ones when modes are encountered?

To match MultiNest they should both default to False.

I don't understand this part. IIRC MultiNest does not have defaults.

How did this PR decide which to set to False, and why should multimodal be False if IS=False?

AdamOrmondroyd commented 1 year ago

if IS=1, I expect MultiNest still does multiple ellipsoids, or not? I guess multimodal refers to splitting the run into several independent ones when modes are encountered?

Some notes I have from a conversation with Mike Hobson last October say "multimodal=False -> live points CANNOT move between clusters". I'll have to reimmerse myself to remember the issues I was seeing.

How did this PR decide which to set to False, and why should multimodal be False if IS=False?

I can't remember why I set both to False. However, if you follow the link, on line 170 of nested.f90, multimodal is set to false if IS is true. I think PyMultiNest should reflect this.

JohannesBuchner commented 1 year ago

maybe importance_nested_sampling=False is enough here.

AdamOrmondroyd commented 1 year ago

What do you think about adding a warning if both are set to True?

JohannesBuchner commented 1 year ago

Explaining what multinest does internally goes beyond the scope of pymultinest.

IS and non-IS MultiNest are two quite different beasts. The stdout of multinest says which one is being performed.