= Score-function gradient estimator
{title2=$\nabla_\lambda\mathbb E_{q_\lambda}h=\mathbb E_{q_\lambda}[h\nabla_\lambda\log q_\lambda]$}
For a parameter-independent integrand $h$, the displayed <score-function gradient estimator> follows by differentiating the sampling <probability density function> or <probability mass function>, using $\nabla q=q\nabla\log q$. It requires conditions permitting <differentiation under the integral sign>, or the corresponding sum, and control of support changes. It applies to discrete choices without pretending that their sampled values are differentiable paths. If $h=h_\lambda$, add $\mathbb E_{q_\lambda}[\nabla_\lambda h_\lambda]$.
Back to article page
