ordered-prob-metrics: Ordered probability metrics

In yardstick: Tidy Characterizations of Model Performance

Ordered probability metrics

Description

Ordered probability metrics evaluate predictions for ordered factor outcomes where the class probabilities should respect the natural ordering of the levels.

Input requirements

• truth: ordered factor

• estimate / ...: numeric columns containing class probabilities

Available metrics

ranked_prob_score()

Direction: minimize. Range: [0, 1]

See Also

class-metrics for hard classification metrics

prob-metrics for class probability metrics

vignette("metric-types") for an overview of all metric types

Examples

# Example with an ordered factor
set.seed(1)
df <- data.frame(
  truth = ordered(sample(1:3, 20, replace = TRUE)),
  prob_1 = runif(20),
  prob_2 = runif(20),
  prob_3 = runif(20)
)
# Normalize probabilities
df[2:4] <- df[2:4] / rowSums(df[2:4])

ranked_prob_score(df, truth, prob_1:prob_3)

yardstick documentation built on April 8, 2026, 1:06 a.m.