Expected Calibration Error for Binary Classification — ece_confidence

Calculates Expected Calibration Error (ECE) or Adaptive ECE for a binary classifier, reproducing the behaviour of pyhealth.metrics.calibration.ece_confidence_binary.

Usage

ece_confidence_binary(prob, label, bins = 20L, adaptive = FALSE)

Arguments

prob: Numeric vector or two-column matrix \ containing predicted probabilities for the positive class (only the first column is used if a matrix is supplied).
label: Numeric vector or two-column matrix of true labels \ encoded as 0/1 (only the first column is used if a matrix is supplied).
bins: Integer. Number of bins (default 20).
adaptive: Logical. If FALSE (default) equal-width bins \ spanning \(0,1\) are used; if TRUE each bin contains the \ same number of samples (equal-size bins).

Value

A single numeric value – the (adaptive) ECE.

Examples

set.seed(1)
p  <- runif(1e4)          # predicted probabilities
y  <- rbinom(1e4, 1, 0.4) # ground-truth labels
ece_confidence_binary(p, y)             # standard ECE
#> [1] 0.2618338
ece_confidence_binary(p, y, adaptive=TRUE)  # adaptive ECE
#> [1] 0.261506