Summary statistics for arbitrary variables within Turing model

Hi -
Is there a way to get summary stats on theta defined in the model below?

@model schools(y, s) = begin
N = length(y)
eta = Array{Float64}(undef, N)
theta = Array{Float64}(undef, N)
mu ~ Uniform(-100, 100)
tau ~ Uniform(0, 100)
eta ~ MvNormal(fill(0, N), 1)
theta = mu .+ (tau .* eta)

for i = 1:N
    y[i] ~ Normal(theta[i], s[i])

mu, tau, eta


Can workout from the chain data on mu, tau and eta. Wondering if there is a way to accomplish it within Turing.


Nope, not right now. It’s a feature target we’re shooting for, though. I think there’s some hacks people have come up to do this with but I can’t find them.

Thank you.