uq_postProcessInversion: Problem: burnIn fraction does not lead to consistent number of posterior sample points

[klux]

Dear UQWorld community,

I encountered a similar issue as mentioned in the Feature request (already closed), this time for the fraction instead of the integer value.

uq_postProcessInversion: Problem, if burnIn is an integer value

Dear @paulremo, dear all, I think I have found an inconsistency between the manual for Bayesian inversion and the code in uq_postProcessInversion.m if one is calling this function with an integer value for the burnIn parameter: Denoting this value by burnInIntValue it holds that: According to the documentation, page 57, burnInIntValue sample should be removed (i.e. not copied to the contents of Results.PostProc) as burn in. In the lines 346–349 of uq_postProcessInversion.m it holds that t…

In the manual for Bayesian inversion on page 55, it says for ‘burnIn’:

Specifies the fraction of sample points that are discarded as burn-in

However, it seems to me that the burn-in fraction times the chosen number of steps is the first step to be used in the post-processed sample. Is that correct?

Concrete example: In my code, I use 400 steps and 100 chains and the default burn-in of 0.5. When I call

PostSample2D = reshape(permute(myBayesianAnalysis.Results.PostProc.PostSample, [2 1 3]), size(PostSample3D, 2), );

as suggested in Looking for documentation/example of using UQLab results of Bayesian inference for forward UQ or further Bayesian inference to access the posterior sample, I obtain 2x20,100 posterior sample. But for a default burn-in of 0.5, you would expect a 2x20,000 posterior sample.

So, is there a msimatch between the implementation and the documentation?

I am using UQLabCore, version 1.3.0. In case, the issue has already been solved in the new version, I apologize for bringing the topic up again.

Thanks in advance for your clarification.

Best regards
Kerstin