Confusion matrix heatmap — confusion_matrix

Cross-tabulates predicted against actual classes and displays the counts as a heatmap. The input can be a fitted binomial glm (with a probability threshold) or a pair of vectors of actual and predicted classes.

Usage

confusion_matrix_plot(
  x,
  predicted = NULL,
  threshold = 0.5,
  normalise = c("none", "row", "col"),
  title = NULL
)

Arguments

x: A binomial glm, or the vector of actual classes.
predicted: When x is an actual-class vector, the matching predicted classes.
threshold: When x is a glm, the probability threshold for the positive class. As well as a number in [0, 1], you may pass the string "youden" to reuse the Youden's J optimal threshold (the same operating point roc_curve_plot() marks), so the confusion matrix and the ROC curve agree on the cut-off.
normalise: One of "none", "row" (by actual class) or "col" (by predicted class); controls the fill shading and the cell annotation.
title: Plot title.

Value

A ggplot2::ggplot object. The threshold actually used is stored in attr(plot, "threshold").

Examples

gfit <- glm(accuracy ~ word_frequency + RT + condition,
            data = lexical_decision, family = binomial)
confusion_matrix_plot(gfit, threshold = 0.5)


# Reuse the Youden-optimal operating point.
confusion_matrix_plot(gfit, threshold = "youden")