How do you extract a sampled variable from a submodel?

[samanklesaria]

Say I've defined a model model1 that samples the variable β and returns nothing (perhaps through TuringGLM). Say I want to use the β variable sampled by model1 from within another model model2. The nearest I can get is the following:

function model2()
    a ~ to_submodel(model1)
    println($(__varinfo__[VarName(Symbol("a.β"))]))
end

Is there a cleaner way of doing this?

[sunxd3]

We had some discussion about this at #2485. (You are very welcome to chime in, btw.)

@penelopeysm and @mhauru would know better, but I think we don't yet have a complete interface for what you want at the moment, am I correct?

[penelopeysm]

The easiest way is to change what the submodel returns, such that the return value contains the beta parameter -- unfortunately if the model you're using is from a library then I guess that's not an option. In the absence of that, yeah, the best way is to use __varinfo__(@varname(var"a.β")). Heads up: in Turing 0.38.0 (#2535) this will change slightly and you will have to use __varinfo__(@varname(a.β)).