Accepts a square count matrix, a named list of count matrices, long frequency
records (reference, predicted, n, optional panel), or paired reference
and predicted label vectors. Model packages can add support for metrics
objects through confusion_input(). Matrix columns are aligned by label identity;
their physical order need not match rows. All panels share one class order.
Usage
draw_confusion(
x,
y = NULL,
classes = NULL,
...,
theme = NULL,
width = NULL,
height = NULL,
element_id = NULL,
filename = NULL
)Arguments
- x
Matrix, table, data frame, named list, metrics object, or vector: Counts or reference labels.
- y
Optional vector: Predicted labels when x contains reference labels.
- classes
Optional Character vector: Ordered class labels shared by every panel; unset preserves first appearance.
- ...
Additional named settings for
setup_ConfusionConfig().- theme
Optional Theme: Chart theme. Color fades and marginal backgrounds follow the active theme unless overridden in the config.
- width, height
Optional Numeric or Character: Widget dimensions.
- element_id
Optional Character: HTML element identifier.
- filename
Optional Character: Static output path, currently SVG.
Statistical semantics
Color intensity is the fraction within each reference row; labels show raw counts. Sensitivity, specificity, PPV, and NPV are one-versus-rest rates for each named class. Balanced accuracy is mean class recall, including every declared class. Zero denominators produce NA, never zero; consequently balanced accuracy is NA if any declared class has no reference observations. Summaries of pooled resamples describe pooled predictions, not mean fold performance. Repeated observations are counted each time they appear.