xgb.ggplot.importance: Plot feature importance

Description

Represents previously calculated feature importance as a bar graph.

xgb.plot.importance() uses base R graphics, while
xgb.ggplot.importance() uses "ggplot".

Usage

xgb.ggplot.importance(
  importance_matrix = NULL,
  top_n = NULL,
  measure = NULL,
  rel_to_first = FALSE,
  n_clusters = seq_len(10),
  ...
)
xgb.plot.importance(
  importance_matrix = NULL,
  top_n = NULL,
  measure = NULL,
  rel_to_first = FALSE,
  left_margin = 10,
  cex = NULL,
  plot = TRUE,
  ...
)

Value

The return value depends on the function:

xgb.plot.importance(): Invisibly, a "data.table" with n_top features sorted by importance. If plot = TRUE, the values are also plotted as barplot.
xgb.ggplot.importance(): A customizable "ggplot" object. E.g., to change the title, set + ggtitle("A GRAPH NAME").

Arguments

importance_matrix: A data.table as returned by xgb.importance().
top_n: Maximal number of top features to include into the plot.
measure: The name of importance measure to plot. When NULL, 'Gain' would be used for trees and 'Weight' would be used for gblinear.
rel_to_first: Whether importance values should be represented as relative to the highest ranked feature, see Details.
n_clusters: A numeric vector containing the min and the max range of the possible number of clusters of bars.
...: Other parameters passed to graphics::barplot() (except horiz, border, cex.names, names.arg, and las). Only used in xgb.plot.importance().
left_margin: Adjust the left margin size to fit feature names. When NULL, the existing par("mar") is used.
cex: Passed as cex.names parameter to graphics::barplot().
plot: Should the barplot be shown? Default is TRUE.

Details

The graph represents each feature as a horizontal bar of length proportional to the importance of a feature. Features are sorted by decreasing importance. It works for both "gblinear" and "gbtree" models.

When rel_to_first = FALSE, the values would be plotted as in importance_matrix. For a "gbtree" model, that would mean being normalized to the total of 1 ("what is feature's importance contribution relative to the whole model?"). For linear models, rel_to_first = FALSE would show actual values of the coefficients. Setting rel_to_first = TRUE allows to see the picture from the perspective of "what is feature's importance contribution relative to the most important feature?"

The "ggplot" backend performs 1-D clustering of the importance values, with bar colors corresponding to different clusters having similar importance values.

Examples

Run this code

data(agaricus.train)

## Keep the number of threads to 2 for examples
nthread <- 2
data.table::setDTthreads(nthread)

model <- xgboost(
  agaricus.train$data, factor(agaricus.train$label),
  nrounds = 2,
  max_depth = 3,
  nthreads = nthread
)

importance_matrix <- xgb.importance(model)
xgb.plot.importance(
  importance_matrix, rel_to_first = TRUE, xlab = "Relative importance"
)

gg <- xgb.ggplot.importance(
  importance_matrix, measure = "Frequency", rel_to_first = TRUE
)
gg
gg + ggplot2::ylab("Frequency")