Skip to contents

Visually help decide the correct threshold parameter

Source: R/utils.R
threshold_diagnostic.Rd

Plots the information content of the variable to be lumped for various values of the threshold. This is useful for determining the correct value to use, in order to not lose too much information.

Usage

threshold_diagnostic(
  X,
  y = NULL,
  thresholds = NULL,
  lumping_mode = c("auto", "ordinal", "nominal", "hierarchical", "heuristic"),
  preference_graph = NULL,
  clusters = NULL,
  levels = NULL,
  heuristic = c("smart", "largest", "other"),
  outcome_mode = c("auto", "discrete", "continuous"),
  plot = interactive(),
  ...
)

Arguments

X

Factor vector containing the variable to be lumped.

y

Optional. The outcome variable for use with the supervised lumping functions.

thresholds

The values of the threshold to test. Default: ranges from 1 to half the sample size.

lumping_mode

Which type of lumping to do. Default: ordinal if X is ordered, nominal otherwise.

preference_graph

The adjacency matrix of the preference graph. Error if passed and lumping_mode is not "nominal" or "heuristic". Default: a complete graph.

clusters

List of character vectors representing the levels that are allowed to be lumped together. Error if passed and lumping_mode is not "hierarchical".

levels

Character vector specifying the strict ordinal hierarchy of the levels (from lowest to highest). Only used when lumping_mode is "ordinal" and X is not already an ordered factor.

heuristic

Which heuristic to use if lumping_mode is "heuristic". Errors if passed but lumping_mode is not "heuristic". See vignette("metrics") for an explanation of the heuristics. Default: "smart".

outcome_mode

Whether to treat y as discrete or continuous. Default: inferred based on the type of y.

plot

Logical value dictating whether the diagnostic plot should be drawn. Set to FALSE to only obtain the returned data. Default: TRUE in an interactive session, FALSE otherwise.

...

Passed to plot().

Value

Invisibly, a data.frame with columns threshold and information (the mutual information preserved at each tested threshold), for the feasible part of the range.

See also

lump_ordinal(), lump_nominal(), and lump_hierarchical() for the unsupervised lumping functions whose threshold argument this helps choose.

lump_ordinal_supervised(), lump_nominal_supervised(), and lump_hierarchical_supervised() for the supervised counterparts, used when an outcome y is supplied.

Author

Daan Koning

Examples

n <- 100
m <- 10
data <- sample(LETTERS[1:m], n, replace = TRUE)
threshold_diagnostic(data, plot = TRUE)