Complex number for the output quantity of interest

All About UQ UQ Discussion Forum
1

Hi,
I was running my model using different sampling methods. The other sampling methods are good except LHS sampling returns complex numbers in my output quantity of interest. Can I know what can be the problem here? Hence when I use distribution fitter app, I am unable to use the data.
Thank you for your help!
Visha

2

I forgot to add that in my model case, the complex number only appears when I ran on 10000 samples. When I tried running on 1000 samples, the resulting moments of the distribution function was fine.

3

Dear Vishashalu,

Have you verified that the input you created is correct? It sounds to me like you have a problem with your model. If you are creating a large number of input samples, you are probably getting inputs that your model is not properly designed for. This might only be the case with a large sample size, as you then get inputs from the tail of the distribution.

Best regards,
Styfen

4

Hi!
Well, this problem only appears when I choose LHS type sampling method, whereas the MC and Sobol or Halton with a 10000 sample runs seems good to me for my model. So does that mean my input created is wrong?
Thanks!
Best regards,
Visha

5

Dear Visha,

I can only guess about the problem. But did you verify that your input is correct when using LHS?

Best regards,
Styfen

6

Dear Styfen,
Yes of course you could only guess:), what I also did not mention here is that when I ran sample numbers lets say 100 and 1000 using LHS sampling method, the results were good. However when I ran on 10000 samples then the resulting mean of the distribution has the complex number.
So I am not sure why is this the case.
Best regards,
Visha

7

Dear Styfen,
Looks like I did mention it to you about only the 10000 sample case run. I think my inputs are correct although I am not entirely sure if I defined them well. But with what I have introduced as the inputs, the MC, sobol and halton sampling methods are working just fine hence why I wasnt sure the LHS sample is giving me these results. Hmm.
Best regards,
Visha

8

Does that mean your input contains complex numbers or the output of your model?
According to your first post, it is the output. And then I would really suggest the following:

  • Check if your sampled input is correct. Plot the PDF of it, for example.
  • Check again that you have defined everything correctly and that the code does what it is supposed to do.
  • Check which inputs lead to incorrect output.
  • Check your model for errors - why does it give the unwanted output, but only for these inputs?

Best regards,
Styfen

9

Dear Styfen,
Thats a good advise! I will do that right away and if anything will get back to you.
Yes the output mean value of the distribution contains complex number not the input.
Thanks!
best regards,
Visha