Projection predictive variable selection for discrete response families with finite support

New preprint by @avehtari and me on an extension of projpred, namely for discrete response families with finite support, e.g., ordinal and nominal (categorical) ones:

Looking forward to merge this into projpred’s master branch soon, but will probably make a small patch release on CRAN first (without this extension), to have a clear cut before the extension comes in. When merging this extension (called augmented-data projection) into master, the latent projection ([2109.04702] Latent space projection predictive inference) will also be included.


Both, augmented-data and latent projection, are now available in projpred’s GitHub version (branch master): GitHub - stan-dev/projpred: Projection predictive variable selection. In case of a brms reference model, an updated brms version is required, available at the brms GitHub site (branch master): GitHub - paul-buerkner/brms: brms R package for Bayesian generalized multivariate non-linear multilevel models using Stan.

1 Like