torch.distributions.ExponentialFamily.entropy
ExponentialFamily.entropy(): TensorComputes entropy using Bregman divergence formula.
For exponential family distributions: H = -E[log p(x)] = A(η) - η·E[T(x)] + E[log h(x)]
Since E[T(x)] = ∇A(η), this simplifies to: H = A(η) - η·∇A(η) + E[log h(x)]
This default implementation uses autograd to compute ∇A(η).
Returns
Tensor– Entropy of the distribution