Hi there, I’m using UQ[py]Lab for inferring marginals and (Gaussian) copula for 23 variables. I’ve defined the inference options following the Statistical Inference manual (https://storage.googleapis.com/uqpylab-doc-html/UserManual_Inference.html). For one variable, I used ‘auto’ for inferring marginal ‘Type’, and I got Student as the best distribution (see screenshot attached). However, Student’s t distribution is NOT listed as a candidate among the 13 univariate distributions according to the Input Modeling manual (https://storage.googleapis.com/uqpylab-doc-html/UserManual_Input.html). The obtained parameter (e-14) for the Student distribution is essentially zero while the moments are nan. Inference results for other marginals and the copula seem ok to me.
By any chance, can anyone help explain why? If this is a bug, how can I prevent this from happening again? I’m thinking that maybe I should explicitly specify ‘Type’ as a list of the 13 distributions instead of using ‘auto’; but I’m not sure whether this will help since I don’t know what went wrong. Thanks in advance!
I’m a new user, so I could only include one media in the above post. To provide more information, here I’m showing also the density histogram of the samples for the variable.
Dear @Xinyue_Wang
The supported distributions for inference are not exactly the 13 listed in the input module, but the ones listed below. I agree that this should probably be mentioned in the manual.
constant
uniform
gaussian
weibull
exponential
lognormal
gamma
gumbel
gumbelmin
logistic
beta
student
laplace
rayleigh
In order to find out the problems causing the nan values, it would be very helpful if you could provide the data.
Best regards
Styfen
Hi Styfen,
Thank you for your kind reply and clarification. I now understand why Student t came up. Since I didn’t get any response, I have decided to explicitly specify a list (excluding Student t) for ‘Type’ inference instead of using ‘auto’. This workaround worked well for me.
I initially included Triangular distribution in the list but got error, so I had to exclude it as well. But now I understand that I was misled by the manual (the manual says Triangular is among the 13 candidates). Though in reality, Triangular is not consdiered by the ‘auto’, based on your reply.
Reagarding data to reproduce the error or nan values, I’m sorry that I couldn’t reproduce the same data as I could 12 days ago. I have made a lot of changes to my code and it would take a lot of time to go back. However, based on the fact that other dictribution types worked fine for the same data, I guess that the error didn’t come from the data. I wish I could provide more useful info to help us figure this out, but I’m sorry that I could not.
Thank you again,
Xinyue
