search

dflemin3 / approxposterior

A Python package for approximate Bayesian inference and optimization using Gaussian processes
https://dflemin3.github.io/approxposterior/
MIT License
41 stars 9 forks source link

Add MultiNest for posterior retreival #49

Open dflemin3 opened 5 years ago

dflemin3 commented 5 years ago

I should implement MultiNest (specifically PyMultiNest) as nested sampling typically converges more quickly than MCMC and can handle multi-modal posterior distributions, whereas MCMC cannot. It should be pretty straight-forward given how the samplers both required an LnPrior and an LnLikelihood function. I'll need to make sure to properly handle the kwargs in the mcmcUtils validation functions.

dflemin3 commented 4 years ago

I should consider this issue in the broader context of implementing functionality to permit different sampling strategies and codes, e.g. approxposterior with pymc3, pystan, etc. Such a change will require some generalizations, but should be feasible for the 1.0 release.