prob macro - pre-allocation syntax

[mohamed82008]

@trappmartin brought up an issue in TuringLang/Turing.jl#997 about the syntax for pre-allocating varinfo. After the PR is merged, we can do:

using Turing

@model demo(y, ::Type{T} = Float64) where {T} = begin
    x = Vector{T}(undef, 10)
    x .~ Normal()
    y .~ Normal.(x, 1.0)
end
xval = rand(10)
varinfo = Turing.VarInfo(demo(xval))
logprob"y = rand(10) | model = demo, x = xval, varinfo = varinfo" 

Pre-allocating is useful if we need to change xval or y in a loop and re-evaluate. Pre-allocating varinfo can be useful in these cases. However, as @trappmartin indicated, the syntax seems to imply we are conditioning on varinfo which we are not. One solution proposed in the PR was to use ; instead of , before varinfo. Another one that I prefer is to not expose this in the macro at all, and only allow it in the functional API under the hood.

[cpfiffer]

Another one that I prefer is to not expose this in the macro at all, and only allow it in the functional API under the hood.

Yeah, I'd agree with that. The macro is really more for people who want to query their data quickly, while the API might be more for power users who want to root around in performance-land.

[yebai]

Closed in favour of #356