How to correct errors related to likelihood function in Bayesian inference

UQ with UQLab Community Q&A and How To
1

I’m trying to apply UQ Lab Bayesian inference to my problem.
Summary of UQ setting for my problem is like below:

  1. Input : velocity (1 X 56)
  2. Parameter: 3
  3. Measurement data : pressure drop (1 X 56)
  4. Model output : caculated pressure drop(# of chain X 56)
  5. MCMC model setting - solver : MCMC, sampler: AIES, steps: 500, nchain : 100
    There was a warning related to likelihood function :

Unable to perform assignment because the size of the left side is 100-by-56 and the size of the right side is 1-by-56.

Error in uq_inversion_likelihood (line 64)
** modelRunsCurr(:,ModelIDCurr == jj) = forwardModel(jj).evaluation(:,OutputIDCurrModel);**

Error in uq_initialize_uq_inversion>@(x)uq_inversion_likelihood(x,Internal,‘LogLikelihood’) (line 497)
** Internal.LogLikelihood = @(x) uq_inversion_likelihood(x,Internal,‘LogLikelihood’);**

Error in uq_AIES (line 67)
** [logLKLHD_Curr, forwardModel_Curr] = LogLikelihood(reshape(Seed,nDim,nChains).’);**

Error in uq_inversion (line 91)
** Results.LogLikeliEval] = uq_AIES(…**

Error in uq_initialize_uq_inversion (line 752)
uq_runAnalysis(CurrentAnalysis);

Error in UQ_1209 (line 126)
myBayesianAnalysisUnknownDisc = uq_createAnalysis(BayesOpts);

I tried this with one parameter ( the other setting was all same). No warnings popped up and had good estimates.
But, after I the number of parameters, this warning came out and couldn’t fix it. Is there any points I need to correct?

2

Hi, @Yesol,
did you solve this problem already?
I think your problem is related to the input of the samples and output matrices.
The input sample matrix must be of dimension Ns x M and the output matrix Ns x Nout, where Ns is the number of samples, M is the number of variables in your model, and Nout is the number of different outputs you want to analyze.
Let me know
Best
Gian

1 Like
3

Yes, I solved the problem.
As you said, dimension was the problem!