Unify `{untyped,typed}_{vector_,}varinfo` constructor functions #879

penelopeysm · 2025-04-01T17:23:16Z

Partly solves #862 (There's still the SimpleVarInfos; but this PR is big enough).

Why

There are a couple of issues that this PR attempts to solve.

First is the ambiguous meaning of TypedVarInfo. Its literal definition is:

Line 99 in eed80e5

const TypedVarInfo = VarInfo{<:NamedTuple}

This was fine in a pre-VarNamedVector world, because UntypedVarInfo meant VarInfo with Metadata, and TypedVarInfo (implicitly) meant VarInfo with NamedTuple of Metadata. However, with VNV now on board, a VarInfo with a NT of VNV also falls under TypedVarInfo. This means that sometimes TypedVarInfo means "VarInfo with NT of anything" (paricularly when used in type signatures), but sometimes it still means "VarInfo with NT of Metadata" (see e.g. the similarly named typed_varinfo function).

Second is the inconsistent interface between different VarInfo constructors. We used to have VarInfo(rng, model, ...) as a very convenient way of constructing a TypedVarInfo (read: a VarInfo with NT of Metadata). None of the other possibilities had equally convenient methods.

What

This PR therefore does the following:

Introduces 4 new functions:
- untyped_varinfo([rng, ]model[, sampler, context]) --> VarInfo with Metadata
- typed_varinfo([rng, ]model[, sampler, context]) --> VarInfo with NT of Metadata
- untyped_vector_varinfo([rng, ]model[, sampler, context]) --> VarInfo with VNV
- typed_vector_varinfo([rng, ]model[, sampler, context]) --> VarInfo with NT of VNV

Note that all four functions are implemented such that if the rng is the same, the values will be the same, regardless of which type of VarInfo is requested. None of these functions are exported.

The metadata argument in VarInfo(rng, model, sampler, context, metadata) no longer works. It was basically being used to specify whether it should be a Metadata or a VarNamedVector inside. If you wanted Metadata (the default) then use DynamicPPL.{untyped,typed}_varinfo; if you wanted VarNamedVector then use DynamicPPL.{untyped,typed}_vector_varinfo.
TypedVarInfo is no longer a type, it has been renamed to NTVarInfo. It is also no longer a function (use typed_varinfo instead).
The type VectorVarInfo = VarInfo{<:VarNamedVector} has been renamed to UntypedVectorVarInfo for consistency. This was not exported (and still isn't).

github-actions · 2025-04-01T17:32:45Z

Benchmark Report for Commit `dbddb2f`

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 |                  9.2 |                 1.6 |
|           Smorgasbord |       201 | forwarddiff |             typed |  false |                730.7 |                33.1 |
|           Smorgasbord |       201 | forwarddiff | simple_namedtuple |   true |                406.7 |                44.7 |
|           Smorgasbord |       201 | forwarddiff |           untyped |   true |               1202.1 |                26.7 |
|           Smorgasbord |       201 | forwarddiff |       simple_dict |   true |               3807.4 |                19.6 |
|           Smorgasbord |       201 | reversediff |             typed |   true |               1400.3 |                29.1 |
|           Smorgasbord |       201 |    mooncake |             typed |   true |                914.5 |                 5.3 |
|    Loop univariate 1k |      1000 |    mooncake |             typed |   true |               5582.4 |                 3.9 |
|       Multivariate 1k |      1000 |    mooncake |             typed |   true |               1058.3 |                 8.4 |
|   Loop univariate 10k |     10000 |    mooncake |             typed |   true |              60953.3 |                 3.6 |
|      Multivariate 10k |     10000 |    mooncake |             typed |   true |               8483.3 |                10.3 |
|               Dynamic |        10 |    mooncake |             typed |   true |                129.9 |                11.9 |
|              Submodel |         1 |    mooncake |             typed |   true |                 24.8 |                 9.3 |
|                   LDA |        12 | reversediff |             typed |   true |                473.8 |                 5.0 |

codecov · 2025-04-01T18:32:13Z

Codecov Report

Attention: Patch coverage is 76.00000% with 30 lines in your changes missing coverage. Please review.

Please upload report for BASE (breaking@cc5e581). Learn more about missing BASE report.

Files with missing lines	Patch %	Lines
src/varinfo.jl	74.57%	30 Missing ⚠️

Additional details and impacted files

@@             Coverage Diff             @@
##             breaking     #879   +/-   ##
===========================================
  Coverage            ?   84.82%           
===========================================
  Files               ?       34           
  Lines               ?     3849           
  Branches            ?        0           
===========================================
  Hits                ?     3265           
  Misses              ?      584           
  Partials            ?        0

☔ 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.

penelopeysm · 2025-04-01T18:59:28Z

src/varinfo.jl

-symbols have been observed. `VarInfo{<:NamedTuple}` is aliased `TypedVarInfo`.
+symbols have been observed. `VarInfo{<:NamedTuple}` is aliased `NTVarInfo`.


Unsurprisingly, the changes are pretty much restricted to this file. The rest of the changes are pretty much just updating VarInfo -> TypedVarInfo.

penelopeysm · 2025-04-01T19:01:13Z

src/varinfo.jl

    logp::Base.RefValue{Tlogp}
    num_produce::Base.RefValue{Int}
 end
-const VectorVarInfo = VarInfo{<:VarNamedVector}
+VarInfo(meta=Metadata()) = VarInfo(meta, Ref{LogProbType}(0.0), Ref(0))


The idea going forward is that VarInfo is purely a low-level constructor for the struct itself, i.e., it is used to construct the VarInfo object from its fields, rather than doing something to instantiate those fields. Thus, one should not expect to be using it except at the very lowest levels of DynamicPPL, e.g. if one is testing basic properties of VarInfo objects.

penelopeysm · 2025-04-01T19:03:15Z

src/varinfo.jl

+function UntypedVarInfo(
+    model::Model,
+    sampler::AbstractSampler=SampleFromPrior(),
+    context::AbstractContext=DefaultContext(),
+)
+    # No rng
+    return UntypedVarInfo(Random.default_rng(), model, sampler, context)
+end
+function UntypedVarInfo(rng::Random.AbstractRNG, model::Model, context::AbstractContext)
+    # No sampler
+    return UntypedVarInfo(rng, model, SampleFromPrior(), context)
+end
+function UntypedVarInfo(model::Model, context::AbstractContext)
+    # No sampler, no rng
+    return UntypedVarInfo(model, SampleFromPrior(), context)
 end


I kinda dislike that i had to copy-paste this 4 times, once for each function. But I don't see a better way around it beyond doing indiscriminate f(args...) = f(g(args...)).

We could make it quite a bit simpler by forbidding f(rng, model, context) and f(model, context), i.e., if we want to specify a context then you also have to specify a sampler.

penelopeysm · 2025-04-01T19:04:22Z

src/varinfo.jl

 """
-function untyped_varinfo(
+function TypedVarInfo(vi::UntypedVarInfo)


This is all shifted from lower in the file, it's not new code

mhauru · 2025-04-02T11:41:33Z

I need to do a proper review, but some thoughts and questions based on just the description of changes:

If {Typed,Untyped}VarInfo is no longer a type, but a function, should its name be lowercase?

The Vector versions are not exported, which I think is good. I view UntypedVectorVarInfo as the king-in-waiting, that will replace UntypedVarInfo once #881 is done. I also dream of a future where all VarInfos are typed, which we could handle using BangBang functions: If new elements need to be added to the NamedTuple we do so by creating a new object, otherwise we operate in-place. In that case I think we would have plain VarInfo be the only user-facing constructor. Do others have thoughts on the future external interface of this stuff?

penelopeysm · 2025-04-02T12:02:42Z

Yup - TypedVarInfo and TypedVectorVarInfo aren't types. I'm not opposed to making them lowercase but I guess for consistency everything should be lowercased so untyped_varinfo, typed_varinfo, untyped_vector_varinfo, typed_vector_varinfo. We did use to have the first two (I deleted them in this PR in favour of the type-lookalikes) but for some reason they weren't used very widely - it wouldn't be too hard to change the names.

(I'd actually prefer if it was md_varinfo, nt_md_varinfo, vnv_varinfo, and nt_vnv_varinfo)

About interface: part of the thing I had in mind with this PR was to keep the interface relatively easy to change in the future. For example replacing MD with VNV is quite straightforward and basically means deleting the existing {Unt,T}ypedVarInfo methods and renaming {Unt,T}ypedVectorVarInfo to them. Personally I view this PR as for the here and now - partly motivated by the need for this in AD testing.

However, if we think that ultimately we just have a single VarInfo() then I could leave that method in instead of deprecating it and then re-introducing it later. Maybe edit the docstring to add a warning that it's an implementation detail as to exactly what form of VarInfo this will return.

mhauru · 2025-04-02T12:49:00Z

I think if this was the final word on VarInfo, we should make the function names lowercase, because it's a pretty clear violation of Julia style conventions. However, I can see an argument for keeping them as they are to have less disruption if/when we change these again.

Personally I view this PR as for the here and now

This sounds fair to me.

However, if we think that ultimately we just have a single VarInfo() then I could leave that method in instead of deprecating it and then re-introducing it later. Maybe edit the docstring to add a warning that it's an implementation detail as to exactly what form of VarInfo this will return.

I like this idea, assuming others are onboard with my vision of a future with only a single VarInfo type.

mhauru

I like this, way more consistent. Some tiny details to note, and the above questions of whether to capitalise and/or deprecate. I could be happy keeping those as they are, just want to make sure we make a conscious decision.

HISTORY.md

src/varinfo.jl

mhauru · 2025-04-07T11:50:33Z

src/varinfo.jl

+    TypedVarInfo([rng, ]model[, sampler, context, metadata])

-Return an untyped varinfo object for the given `model` and `context`.
+Return a VarInfo object for the given `model` and `context`, which has a NamedTuple of
+`Metadata` structs as its metadata field.

 # Arguments
- `model::Model`: The model for which to create the varinfo object.
- `context::AbstractContext`: The context in which to evaluate the model. Default: `SamplingContext()`.
- `metadata::Union{Metadata,VarNamedVector}`: The metadata to use for the varinfo object.
-    Default: `Metadata()`.
+- `rng::Random.AbstractRNG`: The random number generator to use during model evaluation
+- `model::Model`: The model for which to create the varinfo object
+- `sampler::AbstractSampler`: The sampler to use for the model. Defaults to `SampleFromPrior()`.
+- `context::AbstractContext`: The context in which to evaluate the model. Defaults to `DefaultContext()`.


Is there a reason to have this docstring here, rather than attached to the method it documents?

No, probably just me still trying to figure out exactly when Documenter works and exactly when it doesn't. Changed now

penelopeysm · 2025-04-08T10:13:01Z

@mhauru What do you think of this overall approach?

Lowercase the functions
Don't export any of the functions (not even typed / untyped)
Don't deprecate VarInfo, but put a gigantic warning in the docstring saying that the internal implementation of this type may change when you call this constructor and if you need specific types, use the unexported functions

mhauru · 2025-04-08T13:28:56Z

I think I'm happy with that. I hesitated with the middle point, but I guess the lowercase versions weren't exported before either. Is the idea to avoid breakage if we one day move to only having one VarInfo type?

penelopeysm · 2025-04-08T13:30:38Z

Exactly (or in the case where people relied on VarInfo returning, specifically, a TypedVarInfo, at least we warned them).

test/model.jl

penelopeysm · 2025-04-08T16:36:41Z

Bonus: The diff is much more readable now that VarInfo(...) isn't deprecated 😅

mhauru

Thanks @penelopeysm!

* Release 0.36 * AbstractPPL 0.11 + change prefixing behaviour (#830) * AbstractPPL 0.11; change prefixing behaviour * Use DynamicPPL.prefix rather than overloading * Remove VarInfo(VarInfo, params) (#870) * Unify `{untyped,typed}_{vector_,}varinfo` constructor functions (#879) * Unify {Untyped,Typed}{Vector,}VarInfo constructors * Update invocations * NTVarInfo * Fix tests * More fixes * Fixes * Fixes * Fixes * Use lowercase functions, don't deprecate VarInfo * Rewrite VarInfo docstring * Fix methods * Fix methods (really) * Link varinfo by default in AD testing utilities; make test suite run on linked varinfos (#890) * Link VarInfo by default * Tweak interface * Fix tests * Fix interface so that callers can inspect results * Document * Fix tests * Fix changelog * Test linked varinfos Closes #891 * Fix docstring + use AbstractFloat * Fix `condition` and `fix` in submodels (#892) * Fix conditioning in submodels * Simplify contextual_isassumption * Add documentation * Fix some tests * Add tests; fix a bunch of nested submodel issues * Fix fix as well * Fix doctests * Add unit tests for new functions * Add changelog entry * Update changelog Co-authored-by: Hong Ge <[email protected]> * Finish docs * Add a test for conditioning submodel via arguments * Clean new tests up a bit * Fix for VarNames with non-identity lenses * Apply suggestions from code review Co-authored-by: Markus Hauru <[email protected]> * Apply suggestions from code review * Make PrefixContext contain a varname rather than symbol (#896) --------- Co-authored-by: Hong Ge <[email protected]> Co-authored-by: Markus Hauru <[email protected]> --------- Co-authored-by: Markus Hauru <[email protected]> Co-authored-by: Hong Ge <[email protected]> Co-authored-by: Markus Hauru <[email protected]>

Unify {Untyped,Typed}{Vector,}VarInfo constructors

0413fbe

penelopeysm added 2 commits April 1, 2025 18:38

Update invocations

7b33103

NTVarInfo

2ed9a59

penelopeysm changed the title ~~Unify {Untyped,Typed}{Vector,}VarInfo constructors~~ Unify {Untyped,Typed}{Vector,}VarInfo 'constructors' Apr 1, 2025

Fix tests

5cc1287

penelopeysm added 2 commits April 1, 2025 19:35

More fixes

d8c360f

Fixes

460fb2f

penelopeysm commented Apr 1, 2025

View reviewed changes

penelopeysm added 2 commits April 1, 2025 20:56

Fixes

1f7b201

Fixes

416d729

penelopeysm marked this pull request as ready for review April 2, 2025 00:14

penelopeysm requested a review from mhauru April 2, 2025 09:23

mhauru reviewed Apr 7, 2025

View reviewed changes

penelopeysm self-assigned this Apr 8, 2025

penelopeysm mentioned this pull request Apr 8, 2025

Release 0.36 #829

Merged

8 tasks

penelopeysm commented Apr 8, 2025

View reviewed changes

test/model.jl Outdated Show resolved Hide resolved

penelopeysm changed the title ~~Unify {Untyped,Typed}{Vector,}VarInfo 'constructors'~~ Unify {untyped,typed}_{vector_,}varinfo constructor functions Apr 8, 2025

penelopeysm force-pushed the py/varinfo branch 3 times, most recently from c8f9073 to 04a2009 Compare April 8, 2025 16:07

penelopeysm added 2 commits April 8, 2025 17:35

Use lowercase functions, don't deprecate VarInfo

04e6cb8

Rewrite VarInfo docstring

52a017a

penelopeysm force-pushed the py/varinfo branch from 04a2009 to 52a017a Compare April 8, 2025 16:35

penelopeysm requested a review from mhauru April 8, 2025 16:36

penelopeysm added 2 commits April 8, 2025 19:12

Fix methods

46a9d42

Fix methods (really)

dbddb2f

mhauru approved these changes Apr 9, 2025

View reviewed changes

penelopeysm merged commit b9c368b into breaking Apr 9, 2025
17 of 18 checks passed

penelopeysm deleted the py/varinfo branch April 9, 2025 10:35

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Unify `{untyped,typed}_{vector_,}varinfo` constructor functions #879

Unify `{untyped,typed}_{vector_,}varinfo` constructor functions #879

penelopeysm commented Apr 1, 2025 •

edited

Loading

github-actions bot commented Apr 1, 2025 •

edited

Loading

codecov bot commented Apr 1, 2025 •

edited

Loading

penelopeysm Apr 1, 2025

penelopeysm Apr 1, 2025 •

edited

Loading

penelopeysm Apr 1, 2025

penelopeysm Apr 1, 2025

mhauru commented Apr 2, 2025

penelopeysm commented Apr 2, 2025 •

edited

Loading

mhauru commented Apr 2, 2025

mhauru left a comment

mhauru Apr 7, 2025

penelopeysm Apr 8, 2025

penelopeysm commented Apr 8, 2025

mhauru commented Apr 8, 2025

penelopeysm commented Apr 8, 2025 •

edited

Loading

penelopeysm commented Apr 8, 2025

mhauru left a comment

		symbols have been observed. `VarInfo{<:NamedTuple}` is aliased `TypedVarInfo`.
		symbols have been observed. `VarInfo{<:NamedTuple}` is aliased `NTVarInfo`.

Unify {untyped,typed}_{vector_,}varinfo constructor functions #879

Unify {untyped,typed}_{vector_,}varinfo constructor functions #879

Conversation

penelopeysm commented Apr 1, 2025 • edited Loading

Why

What

github-actions bot commented Apr 1, 2025 • edited Loading

Benchmark Report for Commit dbddb2f

Computer Information

Benchmark Results

codecov bot commented Apr 1, 2025 • edited Loading

Codecov Report

penelopeysm Apr 1, 2025

Choose a reason for hiding this comment

penelopeysm Apr 1, 2025 • edited Loading

Choose a reason for hiding this comment

penelopeysm Apr 1, 2025

Choose a reason for hiding this comment

penelopeysm Apr 1, 2025

Choose a reason for hiding this comment

mhauru commented Apr 2, 2025

penelopeysm commented Apr 2, 2025 • edited Loading

mhauru commented Apr 2, 2025

mhauru left a comment

Choose a reason for hiding this comment

mhauru Apr 7, 2025

Choose a reason for hiding this comment

penelopeysm Apr 8, 2025

Choose a reason for hiding this comment

penelopeysm commented Apr 8, 2025

mhauru commented Apr 8, 2025

penelopeysm commented Apr 8, 2025 • edited Loading

penelopeysm commented Apr 8, 2025

mhauru left a comment

Choose a reason for hiding this comment

Unify `{untyped,typed}_{vector_,}varinfo` constructor functions #879

Unify `{untyped,typed}_{vector_,}varinfo` constructor functions #879

penelopeysm commented Apr 1, 2025 •

edited

Loading

github-actions bot commented Apr 1, 2025 •

edited

Loading

Benchmark Report for Commit `dbddb2f`

codecov bot commented Apr 1, 2025 •

edited

Loading

penelopeysm Apr 1, 2025 •

edited

Loading

penelopeysm commented Apr 2, 2025 •

edited

Loading

penelopeysm commented Apr 8, 2025 •

edited

Loading