uq_postProcessInversion: Extension of docmentation to deal with multiple calls of this function, code corrections and extension

Hello,

after reading my above suggestion for the extension of the documentation at the end of the first point again, I decided to rewrite it by ignoring now the current example
in the documentation but dealing with the use-case I had in mind:

New suggest Text:
If one is applying uq_postProcessInversion more then one time, one needs to take into account that the data in BayesianAnalysis.Results.PostPro are not directly updated but (re)created during every call to uq_postProcessInversin using the information in myBayesianAnalysis.Results.Sample and in myBayesianAnalysis.Results.ForwardModel.

For example, assume that the results of a Bayesian inversion analysis
are investigated

1) In a first step the samples points generated by the first and fifth chains from the sample are removed by

 uq_postProcessInversion(myBayesianAnalysis,...
                'badChains', [1 5]);

2) After some further investigations it is decided that one needs also to increase the burn-in fraction from its default value of 50 % to 70 %. Then one should be aware that using now

uq_postProcessInversion(myBayesianAnalysis,...
 'burnIn',0.7 );  % Warinng: would "undelete" removed chains

would somehow “undelete” the removed chains, since the contents of ayesianAnalysis.Results.PostPro are not updated but recreated.
To ensure that the resuls in BayesianAnalysis.Results.PostPro are computed without the samples points generated by the first and fifth chains from the original
MCMC-sample and also without the results from the initial 70 % of the steps in the MCMC algorithm one needs the command

uq_postProcessInversion(myBayesianAnalysis,...
                'badChains', [1 5], 'burnIn',0.7 ); 

3) To investigate afterwards the effect of decreasing the burn-in fraction to 60 %, , one could use the command

 uq_postProcessInversion(myBayesianAnalysis,...
                'badChains', [1 5], 'burnIn',0.6 ); 

Greetings
Olaf

1 Like