plotIndividualContribution: Decompose the predicted value based on the given features

In schalkdaniel/compboost: Efficient Component-Wise Boosting Implementation

Decompose the predicted value based on the given features

Description

This function visualizes the contribution of each feature regarding the predicted value. By default, multiple base learners defined on one feature are aggregated. If you want to show the contribution of single base learner, then set aggregate = FALSE.

Usage

plotIndividualContribution(
  cboost,
  newdata,
  aggregate = TRUE,
  colbreaks = c(-Inf, 0, Inf),
  collabels = c("negative", "positive"),
  nround = 2L,
  offset = TRUE
)

Arguments

cboost	(Compboost) A trained Compboost object.
newdata	(data.frame()) Data frame containing exactly one row holding the new observations.
aggregate	(logical(1L)) Number of colored base learners added to the legend.
colbreaks	(numeric()) Breaks to visualize/highlight different predicted values. Default creates different colors for positive and negative score values. If set to NULL no coloring is applied.
collabels	(character(length(colbreaks) - 1)) Labels for the color breaks. If set to NULL intervals are used as labels.
nround	(integer(1L)) Digit passed to round for labels (default is nround = 2L).
offset	(logical(1L)) Flag to indicate whether the offset should be added to the figure or not.

Value

ggplot object containing the graphic.

Examples

dat = mtcars
fnum = c("cyl", "disp", "hp", "drat", "wt", "qsec")
fcat = c("vs", "am", "gear", "carb")
for (fn in fcat) dat[[fn]] = as.factor(dat[[fn]])

cboost = Compboost$new(data = dat, target = "mpg",
  loss = LossQuadratic$new())

for (fn in fnum) cboost$addComponents(fn, df = 3)
for (fn in fcat) cboost$addBaselearner(fn, "ridge", BaselearnerCategoricalRidge)
cboost$train(500L)
cbreaks = c(-Inf, -0.1, 0.1, Inf)
clabs   = c("bad", "middle", "good")
plotIndividualContribution(cboost, dat[10, ], colbreaks = cbreaks,
  collabels = clabs)
plotIndividualContribution(cboost, dat[10, ], offset = FALSE,
  colbreaks = cbreaks, collabels = clabs)

schalkdaniel/compboost documentation built on April 15, 2023, 9:03 p.m.