Calculate logloss for evaluating predictions

Calculate logloss or cross-entropy for a set of predictions.

logloss(prediction, outcome, tol = .Machine$double.neg.eps)

Arguments

prediction	A vector of estimated probabilities.
outcome	A vector of observed outcomes.
tol	Numerical tolerance. Can also be used to threshold errors for really bad predictions, or when you don't want a model to be penalized too strongly in the presence of high dispersion. Default is .Machine$double.neg.eps.

Value

a numeric vector

Examples

preds <- c(0.9, 0.1, 0.8, 0.5)
outcomes <- c(1, 0, 1, 0)
logloss(preds, outcomes)
#> [1] 0.2817529

# Thresholding large errors for bad predictions
preds <- c(0.000001)
outcomes <- c(1)
logloss(preds, outcomes)
#> [1] 13.81551
logloss(preds, outcomes, 0.01)
#> [1] 4.60517