WAIC, or LOO etc in Turing?

Turing’s support for model comparison could use some extending. If you’re interested in opening a PR over in the MCMCChains repo (which we would think of as being the place to store model comparison stuff) I would be happy to help out.

I know that Seth Axen recently extended Python’s excellent arviz package to work with Turing (and all MCMCChains compatible chain objects). arviz supports WAIC and some other stuff. The package is still a WIP so I don’t know how much functionality there is on the statistics side.

1 Like