[breaking] v0.41

HISTORY.md

@@ -1,3 +1,7 @@
+# v0.41
+
+Removed the `varinfo` keyword argument from `DynamicPPL.TestUtils.AD.run_ad` and replaced the `varinfo` field in `ADResult` with `ldf::LogDensityFunction`.
+
 # 0.40.7
 
 Expanded DynamicPPLMooncakeExt precompilation workload.

Project.toml

@@ -1,6 +1,6 @@
 name = "DynamicPPL"
 uuid = "366bfd00-2699-11ea-058f-f148b4cae6d8"
-version = "0.40.7"
+version = "0.41"
 
 [deps]
 ADTypes = "47edcb42-4c32-4615-8424-f2b9edc5f35b"

benchmarks/Project.toml

@@ -24,7 +24,7 @@ DynamicPPL = {path = "../"}
 ADTypes = "1.14.0"
 Chairmarks = "1.3.1"
 Distributions = "0.25.117"
-DynamicPPL = "0.40"
+DynamicPPL = "0.41"
 Enzyme = "0.13"
 ForwardDiff = "1"
 JSON = "1.3.0"

docs/Project.toml

@@ -32,7 +32,7 @@ Distributions = "0.25"
 Documenter = "1"
 DocumenterInterLinks = "1"
 DocumenterMermaid = "0.1, 0.2"
-DynamicPPL = "0.40"
+DynamicPPL = "0.41"
 FillArrays = "0.13, 1"
 ForwardDiff = "0.10, 1"
 LogDensityProblems = "2"

src/test_utils/ad.jl

@@ -8,16 +8,9 @@ using DynamicPPL:
     DynamicPPL,
     Model,
     LogDensityFunction,
-    VarInfo,
-    AbstractVarInfo,
     AbstractTransformStrategy,
     LinkAll,
     getlogjoint_internal,
-    to_vector_params,
-    get_vector_values,
-    unflatten!!,
-    OnlyAccsVarInfo,
-    VectorValueAccumulator,
     InitFromPrior
 
 using LinearAlgebra: norm
@@ -151,8 +144,9 @@ struct ADResult{Tparams<:AbstractFloat,Tresult<:AbstractFloat,Ttol<:AbstractFloa
     model::Model
     "The function used to extract the log density from the model"
     getlogdensity::Function
-    "The VarInfo that was used"
-    varinfo::Union{Nothing,AbstractVarInfo}
+
+    "The LogDensityFunction that was used"
+    ldf::LogDensityFunction
     "The values at which the model was evaluated"
     params::Vector{Tparams}
     "The AD backend that was tested"
@@ -201,7 +195,6 @@ end
         rtol::AbstractFloat=sqrt(eps()),
         getlogdensity::Function=getlogjoint_internal,
         rng::Random.AbstractRNG=Random.default_rng(),
-        varinfo::Union{Nothing,AbstractVarInfo}=nothing,
         transform_strategy::AbstractTransformStrategy=LinkAll(),
         params::Union{Nothing,Vector{<:AbstractFloat}}=nothing,
         verbose=true,
@@ -235,15 +228,12 @@ Everything else is optional, and can be categorised into several groups:
 
    You can control whether this transformation happens or not by passing the
    `transform_strategy` keyword argument. The default is `LinkAll()`, which means that all
-   parameters will be transformed to unconstrained space. However, if you want to evaluate
-   in the original space, you can use `UnlinkAll()`; you can also specify mixed linking
-   strategies if desired (see [the DynamicPPL documentation](@ref transform-strategies) for
-   more information.
+   parameters will be transformed to unconstrained space. This is the most relevant setting
+   for testing AD.
 
-   Instead of passing the `transform_strategy`, you can also directly pass the `varinfo`
-   keyword argument, which expects a VarInfo object that has been generated with the desired
-   transformation. If both `varinfo` and `transform_strategy` are passed, then `varinfo`
-   takes precedence, and `transform_strategy` is ignored.
+   However, if you want to evaluate in the original space, you can use `UnlinkAll()`; you
+   can also specify mixed linking strategies if desired (see [the DynamicPPL
+   documentation](@ref transform-strategies) for more information). 
 
 1. _How to specify the parameters to be used for evaluation._
 
@@ -252,12 +242,6 @@ Everything else is optional, and can be categorised into several groups:
    prior of the model. If you want to seed the parameter generation, you can pass the `rng`
    keyword argument, which will then be used to generate the parameters.
 
-   Note that these only reflect the parameters used for _evaluating_ the gradient. If you
-   also want to control the parameters used for _preparing_ the gradient, then you need to
-   manually set these parameters inside the `varinfo` keyword argument, for example using
-   `vi = DynamicPPL.unflatten!!(vi, prep_params)`. You could then evaluate the gradient at a
-   different set of parameters using the `params` keyword argument.
-
 1. _Which type of logp is being calculated._
 
    By default, `run_ad` evaluates the 'internal log joint density' of the model,
@@ -334,7 +318,6 @@ function run_ad(
     rtol::AbstractFloat=sqrt(eps()),
     getlogdensity::Function=getlogjoint_internal,
     rng::AbstractRNG=default_rng(),
-    varinfo::Union{Nothing,AbstractVarInfo}=nothing,
     transform_strategy::AbstractTransformStrategy=LinkAll(),
     params::Union{Nothing,Vector{<:AbstractFloat}}=nothing,
     verbose=true,
@@ -347,19 +330,11 @@ function run_ad(
     verbose && @info "Running AD on $(model.f) with $(adtype)\n"
 
     # Generate initial parameters
-    vvals = if isnothing(varinfo)
-        accs = OnlyAccsVarInfo(VectorValueAccumulator())
-        _, accs = DynamicPPL.init!!(rng, model, accs, InitFromPrior(), transform_strategy)
-        get_vector_values(accs)
-    elseif varinfo isa VarInfo
-        varinfo.values
-    elseif varinfo isa OnlyAccsVarInfo
-        get_vector_values(varinfo)
-    end
-    ldf = LogDensityFunction(model, getlogdensity, vvals; adtype=adtype)
+    ldf = LogDensityFunction(model, getlogdensity, transform_strategy; adtype=adtype)
     if isnothing(params)
-        params = to_vector_params(vvals, ldf)
+        params = rand(rng, ldf, InitFromPrior())
     end
+
     params = [p for p in params]  # Concretise
     verbose && println("       params : $(params)")
 
@@ -380,7 +355,7 @@ function run_ad(
             grad_true = test.grad
         elseif test isa WithBackend
             ldf_reference = LogDensityFunction(
-                model, getlogdensity, vvals; adtype=test.adtype
+                model, getlogdensity, transform_strategy; adtype=test.adtype
             )
             value_true, grad_true = logdensity_and_gradient(ldf_reference, params)
             # collect(): https://github.com/JuliaDiff/DifferentiationInterface.jl/issues/754
@@ -417,7 +392,7 @@ function run_ad(
     return ADResult(
         model,
         getlogdensity,
-        varinfo,
+        ldf,
         params,
         adtype,
         atol,