Added functionality for extracting parameter values for a model from chain

[torfjelde]

It's currently quite difficult to convert a MCMCChains.Chains object obtained from some DPPL model into something that we can then use in combination with model to do post-inference analysis, etc.

A very standard scenario is effectively the predict method currently present in Turing.jl, where we do some hacky stuff with strings and setting of values for a given varinfo. But now that we have condition and friends, a very natural approach would of course be to:

1. Convert a Chains into an iterator over (Ordered)Dict corresponding to the parameter realizations.
2. Condition on said (Orderd)Dict and sample from the predictive model.

This PR introduces functionality that allows us to do exactly this.

Specifically, it introduces a method values_iterator_from_chain(model, chain) which returns such an iterator for a given model.

NOTE: This does assume the model is static, i.e. we use a single VarInfo to initialize the process.

This is related to #438 as that PR becomes quite trivial with the values_iterator_from_chain introduced in this PR.

Also related: #478

[coveralls]

Pull Request Test Coverage Report for Build 5577657491

• 0 of 56 (0.0%) changed or added relevant lines in 1 file are covered.
• No unchanged relevant lines lost coverage.
• Overall coverage decreased (-1.6%) to 75.102%

---

Changes Missing Coverage	Covered Lines	Changed/Added Lines	%
src/model_utils.jl	0	56	0.0%

Totals
Change from base Build 5564023611:	-1.6%
Covered Lines:	2024
Relevant Lines:	2695

---

💛 - Coveralls

[codecov]

Codecov Report

Patch coverage has no change and project coverage change: -1.60 ⚠️

Comparison is base (3f80199) 76.69% compared to head (86d8fb1) 75.10%.

Additional details and impacted files

@@            Coverage Diff             @@
##           master     #481      +/-   ##
==========================================
- Coverage   76.69%   75.10%   -1.60%     
==========================================
  Files          22       23       +1     
  Lines        2639     2695      +56     
==========================================
  Hits         2024     2024              
- Misses        615      671      +56     

Impacted Files	Coverage Δ
src/DynamicPPL.jl	100.00% <ø> (ø)
src/model_utils.jl	0.00% <0.00%> (ø)

☔ View full report in Codecov by Sentry.
📢 Do you have feedback about the report comment? Let us know in this issue.

[torfjelde]

I believe tests should now be passing 👍 Should be ready for a look-over.

[torfjelde]

(I need to add tests for this PR)

[YongchaoHuang]

Do we need the 'x' as an input argument here?

[torfjelde]

Yup! It's used below for example,, so we need to pass it to every one of these.

[YongchaoHuang]

thanks for the clarification, yes just found it in varname_leaves.

[YongchaoHuang]

varname_in_chain(varinfo, vn, chain) doesn't exist; maybe using varname_in_chain(varinfo, vn, chain,1,1)?

[YongchaoHuang]

the function used here varname_in_chain!(x, vn_parent ∘ l, chain, out, chain_idx, iteration_idx) has the same signature as the function we are defining in Line 78 - will it lead to circular dead lop (I found this when testing it on my machine)?
Maybe revise both as follows?

function vn_in_chain(x, vn_parent, chain, out, chain_idx, iteration_idx)
    sym = Symbol(vn_parent)
    out[vn_parent] = sym ∈ names(chain) && !ismissing(chain[iteration_idx, sym, chain_idx])
    return out
end
function varname_in_chain!(
    x::AbstractArray, vn_parent::VarName{sym}, chain, out, chain_idx, iteration_idx
) where {sym}
    # We use `VarName{sym}()` so that the resulting leaf `vn` only contains the tail of the lens.
    # This way we can use `getlens(vn)` to extract the value from `x` and use `vn_parent ∘ getlens(vn)`
    # to extract the value from the `chain`.
    for vn in varname_leaves(VarName{sym}(), x)
        # Update `out`, possibly in place, and return.
        l = AbstractPPL.getlens(vn)
        println(vn_parent ∘ l)
        print(typeof(vn_parent ∘ l))
        vn_in_chain(x, vn_parent ∘ l, chain, out, chain_idx, iteration_idx)
    end
    return out
end

[YongchaoHuang]

Don't know if my understanding is correct or not - if we have a chain with variable name s in it, and if the VarName s is passed to the following:

for vn in varname_leaves(VarName(:s), [1])
    l = AbstractPPL.getlens(vn)
    println(VarName(:s) ∘ l)
end

it will generate the VarName s[1] which is obviously not present in the chain. This means for variables that take scalar value, the test will fail. I tested this and it did output s[1] => false; but what we want is s => true I guess.

[yebai]

Looks good to me. Let's try to get this PR merged if there is nothing major left. It would pave the way to merge #438