Corner plot shows different errors than the best fitted values

[cosmicbabs]

Hi!
The corner plot shows the same mean the fitter gave but not the same errors. I tried to figure out what the issue in the code was, but I couldn't find it.

[jvines]

Hi!

This is expected, as the corner plot was always considered a neat visualization add-on and not a final report of parameters. For the actual errorbars we use the Highest Density Regions of the final posterior distribution, which in some cases (that are not uncommon!) can be multimodal and thus the usual way one would obtain the credible intervals might not be representative of the distribution. For the corner plot, I realize the calculation is done the 'usual way', which is basically getting the upper and lower percentiles corresponding to 1 sigma. I recon this should be an easy thing to change, but changing it would make the run time during the plots slower (although I'm sure it wouldn't be an unmanageable overhead).

Let me know if this is absolutely crucial to you and I can change it.

Cheers

[cosmicbabs]

Hey! It is not absolutely necessary, but it would be great if the values matched in case I wanted to use the image of the corner plot in the paper. Is it possible just to add manually the values from the best fit to the corner plot? or to read them from the file? Not calculate them again?

[jvines]

Yeah, I just need to change a couple of lines and it should match. I'll work on it ;)

[taner1tanriverdi]

Hi!
which one i can use in paper to give errors, best_fit_average.dat or best_fit_sample.dat ?
corner gives different errors. but mean is same as mentioned before.