Replace elementwise_loglikelihoods with pointwise_loglikelihoods (#179)

Author: devmotion

Project.toml

@@ -1,6 +1,6 @@
 name = "DynamicPPL"
 uuid = "366bfd00-2699-11ea-058f-f148b4cae6d8"
-version = "0.9.4"
+version = "0.9.5"
 
 [deps]
 AbstractMCMC = "80f14c24-f653-4e6a-9b94-39d6b0f70001"

src/DynamicPPL.jl

@@ -91,7 +91,7 @@ export  AbstractVarInfo,
 # Convenience functions
         logprior,
         logjoint,
-        elementwise_loglikelihoods,
+        pointwise_loglikelihoods,
 # Convenience macros
         @addlogprob!
 
@@ -122,4 +122,6 @@ include("prob_macro.jl")
 include("compat/ad.jl")
 include("loglikelihoods.jl")
 
+include("deprecations.jl")
+
 end # module

src/deprecations.jl

@@ -0,0 +1,22 @@
+@deprecate getmissing(model) getmissings(model)
+
+# `@deprecate` doesn't work with qualified function names,
+# so we use the following hack
+const _base_in = Base.in
+@deprecate _base_in(vn::VarName, space::Tuple) inspace(vn, space)
+
+@deprecate elementwise_loglikelihoods(
+    model::Model, chain,
+) pointwise_loglikelihoods(
+    model, chain, String,
+)
+@deprecate elementwise_loglikelihoods(
+    model::Model, chain, ::Type{T},
+) where {T} pointwise_loglikelihoods(
+    model, chain, T,
+)
+@deprecate elementwise_loglikelihoods(
+    model::Model, varinfo::AbstractVarInfo,
+) pointwise_loglikelihoods(
+    model, varinfo,
+)

src/loglikelihoods.jl

@@ -1,18 +1,18 @@
 # Context version
-struct ElementwiseLikelihoodContext{A, Ctx} <: AbstractContext
+struct PointwiseLikelihoodContext{A, Ctx} <: AbstractContext
     loglikelihoods::A
     ctx::Ctx
 end
 
-function ElementwiseLikelihoodContext(
+function PointwiseLikelihoodContext(
     likelihoods = Dict{VarName, Vector{Float64}}(),
     ctx::AbstractContext = LikelihoodContext()
 )
-    return ElementwiseLikelihoodContext{typeof(likelihoods),typeof(ctx)}(likelihoods, ctx)
+    return PointwiseLikelihoodContext{typeof(likelihoods),typeof(ctx)}(likelihoods, ctx)
 end
 
 function Base.push!(
-    ctx::ElementwiseLikelihoodContext{Dict{VarName, Vector{Float64}}},
+    ctx::PointwiseLikelihoodContext{Dict{VarName, Vector{Float64}}},
     vn::VarName,
     logp::Real
 )
@@ -22,15 +22,15 @@ function Base.push!(
 end
 
 function Base.push!(
-    ctx::ElementwiseLikelihoodContext{Dict{VarName, Float64}},
+    ctx::PointwiseLikelihoodContext{Dict{VarName, Float64}},
     vn::VarName,
     logp::Real
 )
     ctx.loglikelihoods[vn] = logp
 end
 
 function Base.push!(
-    ctx::ElementwiseLikelihoodContext{Dict{String, Vector{Float64}}},
+    ctx::PointwiseLikelihoodContext{Dict{String, Vector{Float64}}},
     vn::VarName,
     logp::Real
 )
@@ -40,15 +40,15 @@ function Base.push!(
 end
 
 function Base.push!(
-    ctx::ElementwiseLikelihoodContext{Dict{String, Float64}},
+    ctx::PointwiseLikelihoodContext{Dict{String, Float64}},
     vn::VarName,
     logp::Real
 )
     ctx.loglikelihoods[string(vn)] = logp
 end
 
 function Base.push!(
-    ctx::ElementwiseLikelihoodContext{Dict{String, Vector{Float64}}},
+    ctx::PointwiseLikelihoodContext{Dict{String, Vector{Float64}}},
     vn::String,
     logp::Real
 )
@@ -58,26 +58,26 @@ function Base.push!(
 end
 
 function Base.push!(
-    ctx::ElementwiseLikelihoodContext{Dict{String, Float64}},
+    ctx::PointwiseLikelihoodContext{Dict{String, Float64}},
     vn::String,
     logp::Real
 )
     ctx.loglikelihoods[vn] = logp
 end
 
 
-function tilde_assume(rng, ctx::ElementwiseLikelihoodContext, sampler, right, vn, inds, vi)
+function tilde_assume(rng, ctx::PointwiseLikelihoodContext, sampler, right, vn, inds, vi)
     return tilde_assume(rng, ctx.ctx, sampler, right, vn, inds, vi)
 end
 
-function dot_tilde_assume(rng, ctx::ElementwiseLikelihoodContext, sampler, right, left, vn, inds, vi)
+function dot_tilde_assume(rng, ctx::PointwiseLikelihoodContext, sampler, right, left, vn, inds, vi)
     value, logp = dot_tilde(rng, ctx.ctx, sampler, right, left, vn, inds, vi)
     acclogp!(vi, logp)
     return value
 end
 
 
-function tilde_observe(ctx::ElementwiseLikelihoodContext, sampler, right, left, vname, vinds, vi)
+function tilde_observe(ctx::PointwiseLikelihoodContext, sampler, right, left, vname, vinds, vi)
     # This is slightly unfortunate since it is not completely generic...
     # Ideally we would call `tilde_observe` recursively but then we don't get the
     # loglikelihood value.
@@ -92,7 +92,7 @@ end
 
 
 """
-    elementwise_loglikelihoods(model::Model, chain::Chains, keytype = String)
+    pointwise_loglikelihoods(model::Model, chain::Chains, keytype = String)
 
 Runs `model` on each sample in `chain` returning a `Dict{String, Matrix{Float64}}`
 with keys corresponding to symbols of the observations, and values being matrices
@@ -138,37 +138,37 @@ julia> model = demo(randn(3), randn());
 
 julia> chain = sample(model, MH(), 10);
 
-julia> elementwise_loglikelihoods(model, chain)
+julia> pointwise_loglikelihoods(model, chain)
 Dict{String,Array{Float64,2}} with 4 entries:
   "xs[3]" => [-1.42862; -2.67573; … ; -1.66251; -1.66251]
   "xs[1]" => [-1.42932; -2.68123; … ; -1.66333; -1.66333]
   "xs[2]" => [-1.6724; -0.861339; … ; -1.62359; -1.62359]
   "y"     => [-1.51265; -0.914129; … ; -1.5499; -1.5499]
 
-julia> elementwise_loglikelihoods(model, chain, String)
+julia> pointwise_loglikelihoods(model, chain, String)
 Dict{String,Array{Float64,2}} with 4 entries:
   "xs[3]" => [-1.42862; -2.67573; … ; -1.66251; -1.66251]
   "xs[1]" => [-1.42932; -2.68123; … ; -1.66333; -1.66333]
   "xs[2]" => [-1.6724; -0.861339; … ; -1.62359; -1.62359]
   "y"     => [-1.51265; -0.914129; … ; -1.5499; -1.5499]
 
-julia> elementwise_loglikelihoods(model, chain, VarName)
+julia> pointwise_loglikelihoods(model, chain, VarName)
 Dict{VarName,Array{Float64,2}} with 4 entries:
   xs[2] => [-1.6724; -0.861339; … ; -1.62359; -1.62359]
   y     => [-1.51265; -0.914129; … ; -1.5499; -1.5499]
   xs[1] => [-1.42932; -2.68123; … ; -1.66333; -1.66333]
   xs[3] => [-1.42862; -2.67573; … ; -1.66251; -1.66251]
 ```
 """
-function elementwise_loglikelihoods(
+function pointwise_loglikelihoods(
     model::Model,
     chain,
     keytype::Type{T} = String
 ) where {T}
     # Get the data by executing the model once
     spl = SampleFromPrior()
     vi = VarInfo(model)
-    ctx = ElementwiseLikelihoodContext(Dict{T, Vector{Float64}}())
+    ctx = PointwiseLikelihoodContext(Dict{T, Vector{Float64}}())
 
     iters = Iterators.product(1:size(chain, 1), 1:size(chain, 3))
     for (sample_idx, chain_idx) in iters
@@ -188,8 +188,8 @@ function elementwise_loglikelihoods(
     return loglikelihoods
 end
 
-function elementwise_loglikelihoods(model::Model, varinfo::AbstractVarInfo)
-    ctx = ElementwiseLikelihoodContext(Dict{VarName, Float64}())
+function pointwise_loglikelihoods(model::Model, varinfo::AbstractVarInfo)
+    ctx = PointwiseLikelihoodContext(Dict{VarName, Float64}())
     model(varinfo, SampleFromPrior(), ctx)
     return ctx.loglikelihoods
 end

src/model.jl

@@ -177,9 +177,6 @@ Get a tuple of the names of the missing arguments of the `model`.
 """
 getmissings(model::Model{_F,_a,_d,missings}) where {missings,_F,_a,_d} = missings
 
-getmissing(model::Model) = getmissings(model)
-@deprecate getmissing(model) getmissings(model)
-
 """
     nameof(model::Model)
 

src/varname.jl

@@ -82,12 +82,6 @@ inspace(vn, space::Tuple) = vn in space
 inspace(vn::VarName, space::Tuple{}) = true
 inspace(vn::VarName, space::Tuple) = any(_in(vn, s) for s in space)
 
-@noinline function Base.in(vn::VarName, space::Tuple)
-    Base.depwarn("`Base.in(vn::VarName, space::Tuple)` is deprecated, use `inspace(vn, space)` instead.",
-                 nameof(Base.in))
-    return inspace(vn, space)
-end
-
 _in(vn::VarName, s::Symbol) = getsym(vn) == s
 _in(vn::VarName, s::VarName) = subsumes(s, vn)
 

test/Turing/Turing.jl

@@ -67,7 +67,7 @@ end
 # Exports #
 ###########
 # `using` statements for stuff to re-export
-using DynamicPPL: elementwise_loglikelihoods, generated_quantities, logprior, logjoint
+using DynamicPPL: pointwise_loglikelihoods, generated_quantities, logprior, logjoint
 using StatsBase: predict
 
 # Turing essentials - modelling macros and inference algorithms
@@ -122,7 +122,7 @@ export  @model,                 # modelling
         arraydist,
 
         predict,
-        elementwise_loglikelihoods,
+        pointwise_loglikelihoods,
         genereated_quantities,
         logprior,
         logjoint

test/loglikelihoods.jl

@@ -15,7 +15,7 @@ using .Turing
     y = randn();
     model = demo(xs, y);
     chain = sample(model, MH(), MCMCThreads(), 100, 2);
-    var_to_likelihoods = elementwise_loglikelihoods(model, chain)
+    var_to_likelihoods = pointwise_loglikelihoods(model, chain)
     @test haskey(var_to_likelihoods, "xs[1]")
     @test haskey(var_to_likelihoods, "xs[2]")
     @test haskey(var_to_likelihoods, "xs[3]")
@@ -31,7 +31,7 @@ using .Turing
     end
 
     var_info = VarInfo(model)
-    results = DynamicPPL.elementwise_loglikelihoods(model, var_info)
+    results = pointwise_loglikelihoods(model, var_info)
     var_to_likelihoods = Dict(string(vn) => ℓ for (vn, ℓ) in results)
     s, m = var_info[SampleFromPrior()]
     @test logpdf(Normal(m, √s), xs[1]) == var_to_likelihoods["xs[1]"]