Page 1 of 1

Cobaya: NaN Rminus1_cl_level

Posted: July 01 2020
by Agne Semenaite
I am running cobaya with mcmc sampler with des shear likelihood and all default settings, as provided by the cosmo-generator. However, when I open the progress file I see that the Rminus1_cl_level is NaN throughout. I have also tried running des shear with planck and get the same result.

As a result, my chains never converge, however, everything else seems ok. I get reasonable Rminus1 on means, I can see no NaN values in the chains, no error messages in the terminal output, the inferred parameter values and contours provided by getdist also look reasonable.

I am not even sure how to begin to troubleshoot this - perhaps someone has had a similar experience? Thanks!

Re: Cobaya: NaN Rminus1_cl_level

Posted: July 01 2020
by Antony Lewis
Perhaps it just hasn't reached the point in convergence where the R-1 tolerance in the means is reached and it starts to also check the R-1 for the convergence of the confidence limits?

Re: Cobaya: NaN Rminus1_cl_level

Posted: July 01 2020
by Agne Semenaite
Oh right - I had not realised the first time the r-1 on confidence limtis is calculated was after the means tolerance is reached - thanks so much! I will try to look into why the chains are running for so long with the default parameters instead.