chainer.cross_entropy

chainer.cross_entropy(dist1, dist2)

Computes Cross entropy.

For two continuous distributions $p(x), q(x)$, it is expressed as

$$H(p,q) = - \int p(x) \log q(x) dx$$

For two discrete distributions $p(x), q(x)$, it is expressed as

$$H(p,q) = - \sum_x p(x) \log q(x)$$

This function call kl_divergence() and entropy() of dist1. Therefore, it is necessary to register KL divergence function with register_kl() decoartor and define entropy() in dist1.

Parameters

• dist1 (Distribution) – Distribution to calculate cross entropy $p$. This is the first (left) operand of the cross entropy.

• dist2 (Distribution) – Distribution to calculate cross entropy $q$. This is the second (right) operand of the cross entropy.

Returns

Output variable representing cross entropy $H(p,q)$.

Return type

Variable