AbstractPPL 0.11 + change prefixing behaviour

[penelopeysm]

Update to AbstractPPL 0.11.

This version of AbstractPPL also lets us prefix variables in submodels differently (and, imo, more intuitively). Excerpt from the release notes:

using DynamicPPL, Distributions
@model inner() = x ~ Normal()
@model outer() = a ~ to_submodel(inner())

In previous versions, the inner variable x would be saved as a.x.
However, this was represented as a single symbol Symbol("a.x"):

julia> dump(keys(VarInfo(outer()))[1])
VarName{Symbol("a.x"), typeof(identity)}
  optic: identity (function of type typeof(identity))

Now, the inner variable is stored as a field x on the VarName a:

julia> dump(keys(VarInfo(outer()))[1])
VarName{:a, Accessors.PropertyLens{:x}}
  optic: Accessors.PropertyLens{:x} (@o _.x)

[codecov]

Codecov Report

Attention: Patch coverage is 81.81818% with 2 lines in your changes missing coverage. Please review.

Project coverage is 84.85%. Comparing base (324e623) to head (68e8c8e).
Report is 3 commits behind head on breaking.

Files with missing lines	Patch %	Lines
src/utils.jl	60.00%	2 Missing ⚠️

Additional details and impacted files

@@             Coverage Diff              @@
##           breaking     #830      +/-   ##
============================================
- Coverage     84.87%   84.85%   -0.02%     
============================================
  Files            34       34              
  Lines          3815     3817       +2     
============================================
+ Hits           3238     3239       +1     
- Misses          577      578       +1     

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

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[github-actions]

Benchmark Report for Commit 68e8c8e

Computer Information

Julia Version 1.11.4
Commit 8561cc3d68d (2025-03-10 11:36 UTC)
Build Info:
  Official https://julialang.org/ release
Platform Info:
  OS: Linux (x86_64-linux-gnu)
  CPU: 4 × AMD EPYC 7763 64-Core Processor
  WORD_SIZE: 64
  LLVM: libLLVM-16.0.6 (ORCJIT, znver3)
Threads: 1 default, 0 interactive, 1 GC (on 4 virtual cores)

Benchmark Results

|                 Model | Dimension |  AD Backend |      VarInfo Type | Linked | Eval Time / Ref Time | AD Time / Eval Time |
|-----------------------|-----------|-------------|-------------------|--------|----------------------|---------------------|
| Simple assume observe |         1 | forwarddiff |             typed |  false |                 10.1 |                 1.5 |
|           Smorgasbord |       201 | forwarddiff |             typed |  false |                711.9 |                39.2 |
|           Smorgasbord |       201 | forwarddiff | simple_namedtuple |   true |                430.7 |                51.3 |
|           Smorgasbord |       201 | forwarddiff |           untyped |   true |               1244.1 |                29.9 |
|           Smorgasbord |       201 | forwarddiff |       simple_dict |   true |               4001.6 |                20.8 |
|           Smorgasbord |       201 | reversediff |             typed |   true |               1487.2 |                29.0 |
|           Smorgasbord |       201 |    mooncake |             typed |   true |                963.6 |                 5.5 |
|    Loop univariate 1k |      1000 |    mooncake |             typed |   true |               5833.7 |                 4.0 |
|       Multivariate 1k |      1000 |    mooncake |             typed |   true |               1165.9 |                 8.1 |
|   Loop univariate 10k |     10000 |    mooncake |             typed |   true |              64168.0 |                 3.8 |
|      Multivariate 10k |     10000 |    mooncake |             typed |   true |               9510.1 |                11.6 |
|               Dynamic |        10 |    mooncake |             typed |   true |                145.5 |                12.0 |
|              Submodel |         1 |    mooncake |             typed |   true |                 26.9 |                 8.1 |
|                   LDA |        12 | reversediff |             typed |   true |                397.9 |                 7.2 |

[penelopeysm]

I had to change the name of this function because Aqua was complaining about method ambiguities. I'm not entirely sure why but I think it's because AbstractPPL defines prefix(::VarName, ::VarName) and if we also defined prefix(::AbstractContext, ::VarName) here, it's probably something to do with the fact that AbstractContext is an abstract type. I still don't entirely see why this is a problem but changing the name did fix it 😅

[sunxd3]

Do we actually need

DynamicPPL.jl/src/DynamicPPL.jl

Line 25 in cab48c6

import AbstractPPL: predict, prefix

?

[sunxd3]

(I am thinking that maybe we don't need to extend AbstractPPL.prefix? Otherwise, the name prefix_with_context is fine.)

[penelopeysm]

Ehhh, I think that's fair. I think I tried that route as well, and the only thing I didn't entirely like is that it means that in DynamicPPL we have to be explicit every time we write prefix - we would have to write either DynamicPPL.prefix or AbstractPPL.prefix. Do you have a preference?

(Personally, I actually think it's better if they are separate - semantically they are different things. Also, it's a nightmare trying to pull up docstrings for the right function when packages just keep extending it - this is the case with sample for example. But historically we have done a lot of this overloading-functions-from-base-packages thing so extending it was kind of in line with that.)

[sunxd3]

I prefer they have separate names -- not a huge fun of name overloading, I think this is abusing multiple dispatch. So totally agree with you.

[penelopeysm]

Cool, we'll go with DynamicPPL.prefix :)

[sunxd3]

otherwise looks good to me

[sunxd3]

Looks great!

[penelopeysm]

Thanks :)