Derive feature groups from the factors behind a dummy encoding

Builds the groups list for group_components() from the original, un-encoded data: every factor column of data whose encoded level columns are found in object$x becomes one group. By default, level columns are expected under the model.matrix() naming convention (paste0(feature, levels), e.g. season with level "Winter" becomes "seasonWinter"), which covers both one-hot (~ f - 1) and treatment (~ f) coding – levels without a matching column (such as a dropped reference level) are simply skipped.

Usage

dummy_groups(
  object,
  data,
  naming = function(feature, levels) paste0(feature, levels)
)

Arguments

object

(glex) Object of class glex.

data

(data.frame) The data before dummy encoding; its factor and character columns define the candidate groups. Other columns are ignored.

naming

(function(feature, levels)) Maps a factor name and its levels to the encoded column names. Defaults to the model.matrix() convention paste0(feature, levels).

Value

Named list of encoded column names, one element per matched factor, suitable as the groups argument of group_components().

Details

If the data was encoded with a different scheme, pass naming: a function of (feature, levels) returning the encoded column names, e.g. function(feature, levels) paste(feature, levels, sep = "_"). It must reproduce the column names the model was trained with, i.e. those found in object$x.

Examples

library(xgboost)
set.seed(1)
n <- 200
d <- data.frame(
  f = factor(sample(c("a", "b", "c"), n, replace = TRUE)),
  x1 = rnorm(n)
)
y <- c(a = 0, b = 2, c = -1)[d$f] + d$x1 + rnorm(n)
x <- model.matrix(~ f + x1 - 1, d)

xg <- xgboost(x, y, nrounds = 10, max_depth = 3, nthreads = 1)
gl <- glex(xg, x)

dummy_groups(gl, d)
#> $f
#> [1] "fa" "fb" "fc"
#> 
grouped <- group_components(gl, dummy_groups(gl, d))
names(grouped$m)
#> [1] "f"    "f:x1" "x1"