Data Science & ML
Advanced

Categorical Cross-Entropy

Loss for multi-class classification with one-hot labels.

Formula

L=−∑i=1Kyilog⁡(y^i)L = -\sum_{i=1}^{K} y_i \log(\hat{y}_i)

Variables

y_iTrue (one-hot)
\hat{y}_iPredicted prob
KClasses

Example

True class prob 0.7: loss = -log(0.7) = 0.357

Did You Know?

Paired with softmax outputs in most neural classifiers.

lossclassification