Bayesian Calibration -identical params in workdirs-

[ala]

Hi,
I am trying to calibrate an openseespy model, I have defined all the needs, the procedure ends but i see parameters in the various run are all the same, all the params in the workdirs are the same, the dakotaTab.out contains all identical parameters and results for the various runs, would you please help me solve it.
Thanks in advance for the help

[fmk]

we will need to see the input file to start solving this one. If running locally it will be in ~/Documents/quoFEM/LocalWorkDir/tmp.SimCebter/templatedir. It is either dakota.json or scInput.json depending on version of quoFEM. We will probably need more, but if you can post this we can see what else is needed/

[ala]

Thank you for your fast answer, here I append the file

[AakashBS]

Hello,

Can you please share the "logFileTMCMC.txt" located in "~/Documents/quoFEM/LocalWorkDir/tmp.SimCenter"?

Best regards,
Aakash

[ala]

Hi, here is the file.
Best regards

[AakashBS]

Thank you for sharing the file. Please increase the number of samples requested to 100 and try again.

[ala]

I have already done a test varing the number of samples, it does not matter.

[AakashBS]

Can you please share your model script, in a private message if you would prefer, so that I can reproduce the issue and try a few solutions?

[ala]

I am afraid I should send the whole ambient. The matter is that I don't read any warning or error and the output files are produced but the params files (params.in) the procedure produces in each of the directories (workdir.#) are exactly the same. The problem generates before the model script is executed. It sems to me the procedure dedicated to produce random values input  params file (different for each workdir) does not work properly in this case, I don't know, pheraps due to an ambient configuration variable, although I have tryed to set everything at my best (not sufficient).

[fmk]

Aakash, what are the circumstances for which the params.in is going to be the same in each workdir.

[AakashBS]

The most common reason for this to happen is that the number of requested samples is small in comparison to the number of parameters to be estimated. This can make the sample covariance matrix ill-conditioned, which causes difficulties when generating new sample points. As a result, during subsequent stages of the TMCMC algorithm, new sample points might not be accepted and each workdir will end up with the same parameter value.

However, since increasing the sample size to a larger number did not resolve the problem, I am not sure what could be the cause of the issue.. It would help if I can reproduce the issue, to diagnose the source of the problem. Sharing the model script or the contents of one of the workdirs will let me recreate the analysis on my machine and help identify the cause.

[AakashBS]

You can also try increasing the standard deviation to have a wider prior, as 10% coefficient of variation results in a narrow, informative prior. If the predictions from the model evaluated at the samples drawn from the prior are all too far away from the data, this can cause numerical difficulties during the algorithm run.

Using a log-normal prior instead of a normal prior will ensure that the input parameter values are admissible, if the parameters should always have positive values.

[ala]

Thank very much you for the helpful suggestions, I will let you know