Page 1 of 1

GetDist: mcsamples.py warning

Posted: December 31 2018
by Dahee Lee
Hi.
When I use GetDist, I got this warning for my parameter V0.

[WARNING] [mcsamples.py:1149] auto bandwidth for V0 very small or failed (h=0.0003142188258111446,N_eff=7341.548063183055). Using fallback (h=0.0011550058718305806)

Result is here.
V0.png
V0.png (7.68 KiB) Viewed 5244 times
What does that mean? And how can I solve this warning?

Sorry for my trivial questions.
Any help would be appreciated.

Re: GetDist: mcsamples.py warning

Posted: December 31 2018
by Antony Lewis
It means the distribution is very spiky or not converged.
Have you set the parameter ranges for V0 correctly (e.g. thereby giving a fake spike across V0=0)?

Re: GetDist: mcsamples.py warning

Posted: January 28 2019
by Dahee Lee
I set the range 0 to 100(because V0 is positive value), and there is no fake spike.
When the number of sample is small(below ~50000), there is no warning and compression.
But the sample number is larger, distribution(1d plot) is more compressed.
Result has worst e-value: R-1 = 0.00828, R-1 of V0 is 0.0233.

In addition, is it possible that position of peak is different from best-fit value in distribution(1d plot)?
In that, best-fit value is about 1.9 in GetdistGUI, but the peak is about 0.8.
Any help would be appreciated.

Re: GetDist: mcsamples.py warning

Posted: January 29 2019
by Antony Lewis
What is the actual code used to make the plot? Are you setting the getdist ranges (not plot limits) explicitly? The global best fit does not have to agree with marginalized parameter peaks.