Reliability of the UQ Results

Hello!

i have a question about the reliability of the results of Orthogonal Matching Pursuit Method (OMP). UQLab is a great tool but for practise, i programmed the OMP Method and solved a cfd problem with 50 uncertainties. My goal is to find the mean and standard deviation of the QoI.

I used the epsilon_loo (1.20 in UQLab Manual) for validation but i am a bit unsure how to choose the less possible number of samples for each case that delivers a reliable result.
According to MC simulations mean should be at around: 7.3636198298 and st. dev: 0.3203129694
I have the following diagramms coming from my programmed OMP Method for PCE Order k=2 and 3. In x axis there are the of mean and st. dev. of the QoI and Number of Samples (or evaluations) in y axis:

1.Here the epsilon_loo is very low but the results aren’t stable as i add samples. Should i maybe add Samples until i see the results (and mainly the st.dev) converge? how is the standard way of doing this? because the espilon_loo is apparently not the best index for reliability.

Thank you very much!

Hi @dimitris_p

I’m wondering if your smaller experimental designs (EDs) are a subset of the larger ones. That is, do you create a completely new ED for each new ED size?

Best regards
Styfen

Hello @styfen.schaer

Yes, each larger ED contains all the samples of the previous smaller ED, so they are subset of the larger ones.

Best regards
Dimitrios