search

stan-dev / projpred

Projection predictive variable selection
https://mc-stan.org/projpred/
Other
110 stars 26 forks source link

Clarify doc for refit_prj #489

Closed avehtari closed 5 months ago

avehtari commented 6 months ago

cv_varsel doc says:

refit_prj: For the evaluation part, should the submodels along the predictor ranking be fitted again (TRUE) or should their fits from the search part be re-used (FALSE)?

I think this sentence would need to be clarified to avoid possible confusion with refitting meaning running MCMC. Maybe rewrite as

refit_prj: For the evaluation part, should the submodels along the predictor ranking be projected again using ndraws_pred draws or nclusters_pred clusters (TRUE) or should their projections from the search part, which used ndraws draws or nclusters clusters, be re-used (FALSE)?

fweber144 commented 6 months ago

I agree. Just want to add that there might be more places (in the docs and/or vignettes) where terminology such as "submodel fit" or "fitting a submodel" is used. Those places would have to be adapted, too.

fweber144 commented 6 months ago

For

should the submodels along the predictor ranking be projected again

something like

should the projections onto the submodels along the predictor ranking be performed again

might be a little more appropriate, as we are projecting the reference model onto the submodels.