vivi_interaction: vivi_interaction

In vivid: Variable Importance and Variable Interaction Displays

vivi_interaction

Description

Compute pairwise interactions only (Friedman's H), without importance.

Usage

vivi_interaction(
  data,
  fit,
  response,
  nmax = 500,
  gridSize = 50,
  predictFun = NULL,
  normalized = FALSE,
  vars = NULL,
  reorder = TRUE
)

Arguments

data	Data frame used for fit.
fit	A supervised ML model understood by condvis2::CVpredict (or supply predictFun).
response	Name of the response column in data.
nmax	Maximum number of rows to consider for grids. Default 500. Use all rows if NULL.
gridSize	Grid size for evaluating partial dependence. Default 50.
predictFun	Optional prediction function ⁠(fit, data, prob = TRUE/FALSE)⁠. If NULL, an internal CVpredict-based function is used.
normalized	Should H be normalised. Default FALSE.
vars	Optional character vector of feature names to restrict the calculation.
reorder	If TRUE, reorder the resulting matrix with vividReorder(). Default TRUE.

Value

Square interaction matrix with zero diagonal.

Examples

# Example 1 — interactions with a linear model
aq <- na.omit(airquality)
fit_lm <- lm(Ozone ~ ., data = aq)
int_mat <- vivi_interaction(data = aq, fit = fit_lm, response = "Ozone")
dim(int_mat); int_mat[1:3, 1:3]
# viviHeatmap(int_mat)  # if you want to visualise interactions only

# Example 2 — classification with ranger

if (requireNamespace("ranger", quietly = TRUE)) {
  library(ranger)
  fit_rf <- ranger(Species ~ ., data = iris,
                   importance = "impurity", probability = TRUE)
  vivi_interaction(data = iris, fit = fit_rf, response = "Species")
}

vivid documentation built on Aug. 24, 2025, 1:09 a.m.