Bump DynamicPPL compat to 0.30

[penelopeysm]

Do not merge – seeing where CI breaks

Tests fixed via TuringLang/DynamicPPL.jl#699.

Closes #2377

Closes #2378

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 86.39%. Comparing base (08331c5) to head (d5d0eea).
Report is 2 commits behind head on master.

Additional details and impacted files

@@           Coverage Diff           @@
##           master    #2376   +/-   ##
=======================================
  Coverage   86.39%   86.39%           
=======================================
  Files          22       22           
  Lines        1573     1573           
=======================================
  Hits         1359     1359           
  Misses        214      214           

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[penelopeysm]

Two of the failing tests:

using Turing
import ReverseDiff

alg = HMC(0.01, 5; adtype=AutoReverseDiff())

@model function vdemo6()
    x = Vector{Real}(undef, 10)
    @. x ~ InverseGamma(2, 3)
end
sample(vdemo6(), alg, 10)

@model function vdemo7()
    x = Array{Real}(undef, N, N)
    @. x ~ [InverseGamma(2, 3) for i in 1:N]
end
sample(vdemo7(), alg, 1000)

---

Bisects to TuringLang/DynamicPPL.jl@c38e65f, i.e. TuringLang/DynamicPPL.jl#555

---

It seems to only happen with InverseGamma, if it's replaced with Normal() or Beta(2, 2) this doesn't error.

---

from_maybe_linked_internal shows up in the stacktrace, so I tracked the result of of from_maybe_linked_internal_transform(vi, vn, dist) for a variety of dists:

Dist	DPPL 0.29	DPPL 0.30
InverseGamma(2, 3)	Base.Fix1{typeof(broadcast), typeof(exp)}(broadcast, exp) ∘ DynamicPPL.FromVec{Tuple{}}(())	DynamicPPL.UnwrapSingletonTransform{Tuple{}}(()) ∘ (Base.Fix1{typeof(broadcast), typeof(exp)}(broadcast, exp) ∘ DynamicPPL.ReshapeTransform{Tuple{Int64}, Tuple{}}((1,), ()))
Normal()	identity ∘ DynamicPPL.FromVec{Tuple{}}(())	DynamicPPL.UnwrapSingletonTransform{Tuple{}}(()) ∘ (identity ∘ DynamicPPL.ReshapeTransform{Tuple{Int64}, Tuple{}}((1,), ()))
Beta(2, 2)	Inverse{Bijectors.Logit{Float64, Float64}}(Bijectors.Logit{Float64, Float64}(0.0, 1.0)) ∘ DynamicPPL.FromVec{Tuple{}}(())	DynamicPPL.UnwrapSingletonTransform{Tuple{}}(()) ∘ (Inverse{Bijectors.Logit{Float64, Float64}}(Bijectors.Logit{Float64, Float64}(0.0, 1.0)) ∘ DynamicPPL.ReshapeTransform{Tuple{Int64}, Tuple{}}((1,), ()))

The top-right is the only one that fails.

---

I commented out these lines:

https://github.com/TuringLang/DynamicPPL.jl/blob/bd9f465d174a631c862645739af6748ac0abbefd/src/utils.jl#L378-L384

and when rerun, from_maybe_linked_internal_transform(vi, vn, InverseGamma(2, 3)) gives

Base.Fix1{typeof(broadcast), typeof(exp)}(broadcast, exp) ∘ DynamicPPL.ReshapeTransform{Tuple{Int64}, Tuple{}}((1,), ()))

But it still errors. So the problem is with ReshapeTransform rather than UnwrapSingletonTransform – but apparently only when it's composed with Base.Fix1{typeof(broadcast), typeof(exp)}(broadcast, exp).

[coveralls]

Pull Request Test Coverage Report for Build 11520126094

Warning: This coverage report may be inaccurate.

This pull request's base commit is no longer the HEAD commit of its target branch. This means it includes changes from outside the original pull request, including, potentially, unrelated coverage changes.

• For more information on this, see Tracking coverage changes with pull request builds.
• To avoid this issue with future PRs, see these Recommended CI Configurations.
• For a quick fix, rebase this PR at GitHub. Your next report should be accurate.

Details

• 0 of 0 changed or added relevant lines in 0 files are covered.
• No unchanged relevant lines lost coverage.
• Overall coverage remained the same at 86.395%

---

Totals
Change from base Build 11497545739:	0.0%
Covered Lines:	1359
Relevant Lines:	1573

---

💛 - Coveralls

[penelopeysm]

MWE that doesn't involve DynamicPPL at all:

import ReverseDiff
f(x) = exp.(reshape(vec(x), ()))
f([1.0]) # this is fine
ReverseDiff.gradient(f, [1.0]) # this errors

Here exp.() is a simplified version of invlink_transform(InverseGamma(...)) and reshape(vec(x), ()) a simplified version of DynamicPPL.ReshapeTransform(()).

Reported upstream JuliaDiff/ReverseDiff.jl#265

[yebai]

Thanks @penelopeysm!