Confusion matrix

A tile heatmap of predicted vs actual classes, shaded by row-normalized rate and annotated with counts, so class imbalance does not hide errors.

geom_confusion_matrix(
  mapping = NULL,
  data = NULL,
  normalize = "row",
  label_size = NULL
)

Arguments

mapping, data

Standard layer overrides. Map the predicted class to x, the actual class to y. Provide counts via size, or pass raw per-observation rows and let the stat count them.

normalize

"row" (default), "col", "all", or "none" — which total the shading is relative to.

label_size

Annotation size in px.

Value

A Layer to add with +.

Examples

d <- data.frame(
  predicted = c("cat", "cat", "dog", "dog", "dog", "cat"),
  actual    = c("cat", "dog", "dog", "dog", "cat", "cat")
)
ggnext(d, aes(predicted, actual)) + geom_confusion_matrix()
#> <ggnext plot>
#>   data:  6 x 2
#>   aes:   x = predicted, y = actual
#>   layers: 1 
#>   coord: cartesian
#>   mode:  static