mc_sbrier: Calculate the multiclass scaled Brier Score for predicted...

In gweissman/gmish: Miscellaneous Functions for Predictive Modeling

Calculate the multiclass scaled Brier Score for predicted probabilities against a binary outcome.

Description

Calculate the multiclass scaled Brier Score for predicted probabilities against a binary outcome.

Usage

mc_sbrier(preds, obs)

Arguments

preds	A data.frame or matrix of predicted probabilities with one column per class.
obs	A data.frame or matrix containing the observed binary outcomes (0 or 1) with one column per class. The order of the columns should match the order for preds.

Value

The multiclass scaled Brier Score given by

1 - \frac{BS}{BS_{max}}

, where the multiclass Brier score is given by

BS = \frac{1}{N} \sum^K \sum^N (y - \hat y)^2

for N observations and K classes

Examples

# Generate some predictions
predictions <- data.frame(p1 = runif(1000), p2 = runif(1000), p3 = runif(1000))
# Generate some binary outcomes
observations <- data.frame(o1 = sample(0:1, size = 1000, replace = TRUE),
o2 = sample(0:1, size = 1000, replace = TRUE), o3 = sample(0:1, size = 1000, replace = TRUE))
# Calculate the multiclass Brier Score
mc_sbrier(predictions, observations)

gweissman/gmish documentation built on Feb. 16, 2025, 8:38 a.m.