varimp.diag: Diagnostic variable selection plot

In cjcarlson/embarcadero: Species distribution models with BART

Diagnostic variable selection plot

Description

When BART is run with a small number of trees it tends to up-select variables that contribute the most meaningfully, and under-selects variables that have no relevance or contribute only marginally. As the number of trees, this pattern becomes less visible, as it overfits to less useful variables. Plotting this is a useful way of identifying which variables should be dropped - those that have the most visible increase relative to number of trees are performing the poorest.

Usage

varimp.diag(x.data, y.data, ri.data = NULL, iter = 50, quiet = FALSE)

Arguments

x.data	A data frame of covariates
y.data	A vector of outcomes (1/0)
iter	How many BART models to run for each of (10, 20, 50, 100, 150, 200) tree models

cjcarlson/embarcadero documentation built on Sept. 9, 2023, 10:47 p.m.