Do you think it is worth adding the Generalized Pareto Distribution (GPD) to Stan? I am doing some Extreme Value Theory analyses and for some Peak Over Threshold problems, a GPD would be a good fit. I read the excellent case study by Aki and I wonder if that effort could be included in the standard library so that to use it with other interfaces (brms and the like).
2 Likes
Would you like to take a crack at adding it? First step is to create an issue in the math library.
I’m not looking at a reference to it atm, are the derivatives known or easy to compute? That will provide an additional speed benefit.
I can definitely open an issue. I am not competent to work on it directly, I fear. But, as far as I know, most of the work has been done by Aki already.
EDIT: Issue has been opened here
The functions all seem to be fairly basic, and shouldn’t be too tricky to implement with gradients. I can take a crack at this over the next week or two
4 Likes
I was wondering if you had the opportunity to look at the functions. I am very much looking forward to testing it!
Ah thanks for prompting me on this! It fell off the to-do list, but will get onto it over the next few days.
1 Like
(hehe a very very friendly) ping!