Brier Score#
Brier Score is a strictly proper scoring metric that is equivalent to applying mean squared error to the probabilities of a probabilistic estimator.
Note
In order to maintain the convention of maximizing validation scores, this metric outputs the negative of the original score.
Estimator Compatibility: Probabilistic Classifier
Score Range: -2 to 0
Parameters#
This metric does not have any parameters.
Example#
use Rubix\ML\CrossValidation\Metrics\BrierScore;
$metric = new BrierScore();
References#
-
G. W. Brier. (1950). Verification of Forecasts Expresses in Terms of Probability. ↩