In addition to parameter calibration, can the Bayesian inference module in UQLab be used to make predictions?
It is not exactly clear to me what you mean by “making prediction”. Are you interested in the predictive distribution? If yes, I suggest you to first have a look at the manual:
Let us know if this is not what you’re interested in or if you are stuck with the implementation.
In UQLab - Bayesian inversion user manual, there are model predictions as shown in the picture. How can I use it to make predictions in a real case?
Dear @Mr121Tony ,
if you are interested in making predictions in a “real case”, you have to generate a
BayesOpts object as described in Section 3.1 of the manual describing the “real case” and use this as input for
uq_createAnalysis. Using the
.Results.PostProc.PostSample-component of the struct resulting from this this function call one can compute predictions, which I prefer to denote as performing forward UQ.
Following the suggestion by Paul-Remo in https://uqworld.org/t/looking-for-documentation-example-of-using-uqlab-results-of-bayesian-inference-for-forward-uq-or-further-bayesian-inference/690/2?u=olaf.klein I extract 2D-arrays containing sample-vectors of
the values of the different parameters combined with a value for the computed discrepancy. Afterwards I use these sample-vectors to perform Forward UQ
similar to the method the posterior predictive samples are computed in UQLab:
for each sample-vector I computed the value for my model for the given set of
parameter values and afterwards noise was added by computing a noise sample
for a normal distributed error according to the corresponding discrepancy value to
get a sample for the posterior predictive density.
@olaf.klein, Thank you. I will try to solve this problem.