Loss & Training · Easy

Binary Cross-Entropy

Implement stable binary cross entropy with optional weights.

Task

Implement binary_cross_entropy(labels: list[int], probabilities: list[float], sample_weights: list[float] | None = None) -> float. Compute stable binary cross-entropy with optional non-negative sample weights.

Requirements

Clip every probability to [1e-15, 1 - 1e-15] before taking logarithms.
Without weights, return the mean loss across examples.
With weights, divide the weighted loss sum by the sum of weights.
Return 0.0 for empty inputs or when the total sample weight is zero.

Example

labels = [1, 0]
probabilities = [0.9, 0.2]
sample_weights = [1.0, 3.0]

binary_cross_entropy(labels, probabilities, sample_weights)
# 0.1936977924