search

brinckmann / montepython_public

Public repository for the Monte Python Code
MIT License
93 stars 81 forks source link

Calculating evidence from Montepython #376

Open Amlan1996 opened 4 months ago

Amlan1996 commented 4 months ago

Hi,

Can I calculate evidence from Monte Python run outputs under the Metropolis-Hastings algorithm? I was running with the CMB+Pantheon_Plus data set. After the convergence, I got all the chain files but couldn't find concise information on the log-likelihood or chi2, which I can use to integrate over my priors.

Please let me know if there is any way to do this.

Best regards, Amlan

brinckmann commented 4 months ago

Hi Amlan,

MontePython doesn't compute the evidence for Metropolis-Hastings. It does provide the -loglkl in the chains, it's the second number with the first number being the multiplicity.

I know of two external codes that can compute the Bayesian evidence for you by post-processing your chains: MCEvidence code https://github.com/yabebalFantaye/MCEvidence + release paper https://arxiv.org/abs/1704.03472 harmonic code https://github.com/astro-informatics/harmonic + release paper https://arxiv.org/abs/2111.12720

I have not tried either code. Good luck!

Best, Thejs

Amlan1996 commented 4 months ago

Dear Thejs,

Thanks for the information on these external codes.

I was also checking the documentation and learned that nested sampling in Montepython might automatically calculate the evidence. Since I have never used it, can I rerun everything using the same .param, .best fit, and .covmat files to run under this nested sampling? Is the command also the same for analysing the chains similar to metropolis hastings?

Please let me know.

Best regards, Amlan