Extreme Value Analysis with Pymc3

I am trying to model a GEV distribution with three parameters (shape, location, and scale), but couldn’t find any example or tutorial to do it. Right now, I am just going with the traditional approach of defining a transition model and the MCMC Metropolis-Hastings; but it seems to be very slow. Any leads?

We would need a bit more information. It sounds like you want to implement your own distribution? There are definitely tutorials on there for that, let me know if you have trouble locating them.

Can you run NUTS on your transition model? That’s usually much faster.

Do you use the Gumbel distribution?