GetDist : parameter statistics table

Use of Cobaya. camb, CLASS, cosmomc, compilers, etc.
Post Reply
Junxian Li
Posts: 4
Joined: July 24 2023
Affiliation: Sun Yat-Sen University

GetDist : parameter statistics table

Post by Junxian Li »

With getdist-gui, in the sample likelihood constraints panel, I got a very strange upper and lower bound.

The lower bound value is bigger than best fit value, while the upper bound value is a abnormal small value.

It is nothing to do with the model.

The chains are converged with |R-1|<0.01. The contour and marginalized constraints stats are quite normal. see red box below.

Bestfit values are also quite normal.

No idea why the boundary values are strange. see yellow box below.
1727151734078.jpg
1727151734078.jpg (418.11 KiB) Viewed 1495 times
Junxian Li
Posts: 4
Joined: July 24 2023
Affiliation: Sun Yat-Sen University

Re: GetDist : parameter statistics table

Post by Junxian Li »

Marginalized constraints: These values are obtained by marginalizing the posterior probability distribution, representing the uncertainty of the parameters. The calculation considers the entire posterior distribution of the parameter space, making it more stable and accurate.
Best fit: This is the point estimate derived from maximum likelihood estimation, which may exhibit some instability due to sampling at a single point. When sampling is insufficient or the likelihood distribution shape at that point is not ideal, anomalous cases of lower1 and upper1 may occur.
Antony Lewis
Posts: 1980
Joined: September 23 2004
Affiliation: University of Sussex
Contact:

Re: GetDist : parameter statistics table

Post by Antony Lewis »

Does look a bit odd, please share the chains in a github issue, thanks.
Junxian Li
Posts: 4
Joined: July 24 2023
Affiliation: Sun Yat-Sen University

Re: GetDist : parameter statistics table

Post by Junxian Li »

Dear Antony, thanks for your reply. Github issue has been created, see https://github.com/cmbant/getdist/issues/119.
Post Reply