-sum_k y_k log p_k per observation, averaged (ESL eqs 7.7-7.8 give -2 times
the mean log-likelihood).
Usage
morie_esl_cross_entropy(y, p)
Arguments
- y
Indicator matrix (N by K) or one-hot rows.
- p
Predicted probabilities, same shape.
Value
Named list: estimate (mean loss), per_observation, n, K.
References
Hastie, Tibshirani & Friedman (2009), sec. 7.2.
Examples
morie_esl_cross_entropy(diag(2), rbind(c(0.8, 0.2), c(0.3, 0.7)))$estimate
#> [1] 0.2899092
