projsel {hsstan}

Forward selection minimizing KL-divergence in projection

Description

Forward selection minimizing KL-divergence in projection

Usage

projsel(obj, max.iters = 30, start.from = NULL, out.csv = NULL)

Arguments

obj

Object of class hsstan.

max.iters

Maximum number of iterations (number of predictors selected) after which the selection procedure should stop.

start.from

Vector of variable names to be used in the starting submodel. If NULL (default), selection starts from the set of unpenalized covariates if the model contains penalized predictors, otherwise selection starts from the intercept-only model.

out.csv

If not NULL, the name of a CSV file to save the output to.

Value

A data frame of class projsel where each row corresponds to a forward-selected submodel that contains all variables listed up to that row. Attribute start.from reports the predictors in the initial model. The data frame contains the following columns:

var

names of the variables selected.

kl

KL-divergence from the full model to the submodel.

rel.kl.null

relative explanatory power of predictors starting from the intercept-only model.

rel.kl

relative explanatory power of predictors starting from the initial submodel.

elpd

the expected log predictive density of the submodels.

delta.elpd

the difference in elpd from the full model.

Examples

# continued from ?hsstan
sel <- projsel(hs.biom, max.iters=3)
plot(sel)