Hello all, I have a question about the circular cases such as when we want to calculate the mean of std of angles.

If the angles are unwrapped (rather than starting from 0 and finishing at 360, it continuously increases from zero to infinity) do we need those circulate statistics or we can use regular concepts of mean and standard deviation?

Also, in general, does UQlab cover circular statistics as well?

Thanks in advance

Any ideas/comments are greatly appreciated.

Any ideas/comments are greatly appreciated. Can UQlab deal with circular statistics?

Hello all, I am still struggling with this problem and I appreciate any helps. Let me ask my question in another way:

Considering my quantity of interest to be angle/phase. Can I use those values to create PCE with uqlab like any other non-circular parameters? or I should apply other circular statistics-related changes to the uqlab code itself?

Hi @Aep93 ,

No, UQLab does not offer special support for circular statistics. And no, unwrapping the angles does not solve the problem of having nonunique numerical values for the same quantity.

It should be possible to compute a PCE for inputs that are angles, why not, as long as you know their distribution, and use an interval in which the quantities are unique (e.g., [-\pi, \pi] or [0, 2\pi]). The fact that the quantities are angles/circular just give you the additional information that the surrogate model should in principle be periodic (M^\text{PCE}(\alpha) = M^\text{PCE}(\alpha + k 2\pi ) for all k \in \mathbb{Z}).

Please, let us know how you deal with your problem and whether computing a PCE using angles works out for you!