Let's $p_\theta(x)$ is parameterized PDF, $P_\theta(x)$ is CDF of $p_\theta(x)$ and we know the distribution.
$q(x)$ is also PDF and $Q(x)$ is CDF of $q(x)$ but we only get the sample of score function(I mean unnormalized value $q(x)={1 \over z}q^{'}(x)$)
how to reduce the distance between $P_\theta(x)$ and $Q(x)$ using the samples($q^{'}(x)$)?
I thought quantile regression but it needs Q(x)