CalibratedBinaryClassificationMetrics.LogLossReduction Property
Definition
Important
Some information relates to prerelease product that may be substantially modified before it’s released. Microsoft makes no warranties, express or implied, with respect to the information provided here.
Gets the log-loss reduction (also known as relative log-loss, or reduction in information gain - RIG) of the classifier. It gives a measure of how much a model improves on a model that gives random predictions. Log-loss reduction closer to 1 indicates a better model.
public double LogLossReduction { get; }member this.LogLossReduction : doublePublic ReadOnly Property LogLossReduction As DoubleProperty Value
Remarks
The log-loss reduction is scaled relative to a classifier that predicts the prior for every example: $LogLossReduction = \frac{LogLoss(prior) - LogLoss(classifier)}{LogLoss(prior)}$ This metric can be interpreted as the advantage of the classifier over a random prediction. For example, if the RIG equals 0.2, it can be interpreted as "the probability of a correct prediction is 20% better than random guessing".