fix rename adbackend to adtype

README.md

@@ -43,7 +43,7 @@ function LogDensityProblems.capabilities(::Type{<:NormalLogNormal})
 end
 ```
 
-Since the support of `x` is constrained to be positive, and VI is best done in the unconstrained Euclidean space, we need to use a *bijector* to transform `x` into unconstrained Euclidean space. We will use the [`Bijectors.jl`](https://github.com/TuringLang/Bijectors.jl) package for this purpose. 
+Since the support of `x` is constrained to be positive and VI is best done in the unconstrained Euclidean space, we need to use a *bijector* to transform `x` into unconstrained Euclidean space. We will use the [`Bijectors.jl`](https://github.com/TuringLang/Bijectors.jl) package for this purpose. 
 This corresponds to the automatic differentiation variational inference (ADVI) formulation[^KTRGB2017].
 ```julia
 using Bijectors
@@ -99,7 +99,7 @@ q, stats, _ = AdvancedVI.optimize(
     elbo,
     q_transformed,
     max_iter;
-    adbackend = ADTypes.AutoForwardDiff(),
+    adtype    = ADTypes.AutoForwardDiff(),
     optimizer = Optimisers.Adam(1e-3)
 )
 

docs/src/elbo/repgradelbo.md

@@ -189,7 +189,7 @@ _, stats_cfe, _ = AdvancedVI.optimize(
     q0_trans,
     max_iter;
     show_progress = false,
-    adbackend     = AutoForwardDiff(),
+    adtype        = AutoForwardDiff(),
     optimizer     = Optimisers.Adam(1e-3)
 ); 
 
@@ -199,7 +199,7 @@ _, stats_stl, _ = AdvancedVI.optimize(
     q0_trans,
     max_iter;
     show_progress = false,
-    adbackend     = AutoForwardDiff(),
+    adtype        = AutoForwardDiff(),
     optimizer     = Optimisers.Adam(1e-3)
 ); 
 
@@ -264,7 +264,7 @@ _, stats_qmc, _ = AdvancedVI.optimize(
     q0_trans,
     max_iter;
     show_progress = false,
-    adbackend     = AutoForwardDiff(),
+    adtype        = AutoForwardDiff(),
     optimizer     = Optimisers.Adam(1e-3)
 ); 
 

docs/src/examples.md

@@ -102,7 +102,7 @@ q_trans, stats, _ = AdvancedVI.optimize(
     q0_trans,
     n_max_iter;
     show_progress = false,
-    adbackend     = AutoForwardDiff(),
+    adtype        = AutoForwardDiff(),
     optimizer     = Optimisers.Adam(1e-3)
 ); 
 nothing

src/AdvancedVI.jl

@@ -93,14 +93,14 @@ export estimate_objective
 
 
 """
-    estimate_gradient!(rng, obj, adbackend, out, prob, λ, restructure, obj_state)
+    estimate_gradient!(rng, obj, adtype, out, prob, λ, restructure, obj_state)
 
 Estimate (possibly stochastic) gradients of the variational objective `obj` targeting `prob` with respect to the variational parameters `λ`
 
 # Arguments
 - `rng::Random.AbstractRNG`: Random number generator.
 - `obj::AbstractVariationalObjective`: Variational objective.
-- `adbackend::ADTypes.AbstractADType`: Automatic differentiation backend. 
+- `adtype::ADTypes.AbstractADType`: Automatic differentiation backend. 
 - `out::DiffResults.MutableDiffResult`: Buffer containing the objective value and gradient estimates. 
 - `prob`: The target log-joint likelihood implementing the `LogDensityProblem` interface.
 - `λ`: Variational parameters to evaluate the gradient on.

src/objectives/elbo/repgradelbo.jl

@@ -95,10 +95,10 @@ estimate_objective(obj::RepGradELBO, q, prob; n_samples::Int = obj.n_samples) =
     estimate_objective(Random.default_rng(), obj, q, prob; n_samples)
 
 function estimate_gradient!(
-    rng      ::Random.AbstractRNG,
-    obj      ::RepGradELBO,
-    adbackend::ADTypes.AbstractADType,
-    out      ::DiffResults.MutableDiffResult,
+    rng   ::Random.AbstractRNG,
+    obj   ::RepGradELBO,
+    adtype::ADTypes.AbstractADType,
+    out   ::DiffResults.MutableDiffResult,
     prob,
     λ,
     restructure,
@@ -112,7 +112,7 @@ function estimate_gradient!(
         elbo = energy + entropy
         -elbo
     end
-    value_and_gradient!(adbackend, f, λ, out)
+    value_and_gradient!(adtype, f, λ, out)
 
     nelbo = DiffResults.value(out)
     stat  = (elbo=-nelbo,)

src/optimize.jl

@@ -16,7 +16,7 @@ The variational approximation can be constructed by passing the variational para
 - `objargs...`: Arguments to be passed to `objective`.
 
 # Keyword Arguments
-- `adbackend::ADtypes.AbstractADType`: Automatic differentiation backend. 
+- `adtype::ADtypes.AbstractADType`: Automatic differentiation backend. 
 - `optimizer::Optimisers.AbstractRule`: Optimizer used for inference. (Default: `Adam`.)
 - `rng::AbstractRNG`: Random number generator. (Default: `Random.default_rng()`.)
 - `show_progress::Bool`: Whether to show the progress bar. (Default: `true`.)
@@ -54,7 +54,7 @@ function optimize(
     params_init,
     max_iter     ::Int,
     objargs...;
-    adbackend    ::ADTypes.AbstractADType, 
+    adtype       ::ADTypes.AbstractADType, 
     optimizer    ::Optimisers.AbstractRule = Optimisers.Adam(),
     show_progress::Bool                    = true,
     state_init   ::NamedTuple              = NamedTuple(),
@@ -77,7 +77,7 @@ function optimize(
         stat = (iteration=t,)
 
         grad_buf, obj_st, stat′ = estimate_gradient!(
-            rng, objective, adbackend, grad_buf, problem,
+            rng, objective, adtype, grad_buf, problem,
             λ, restructure,  obj_st, objargs...
         )
         stat = merge(stat, stat′)

test/inference/repgradelbo_distributionsad.jl

@@ -14,7 +14,7 @@ using Test
             :RepGradELBOClosedFormEntropy  => RepGradELBO(n_montecarlo),
             :RepGradELBOStickingTheLanding => RepGradELBO(n_montecarlo, entropy = StickingTheLandingEntropy()),
         ),
-        (adbackname, adbackend) ∈ Dict(
+        (adbackname, adtype) ∈ Dict(
             :ForwarDiff  => AutoForwardDiff(),
             #:ReverseDiff => AutoReverseDiff(),
             :Zygote      => AutoZygote(), 
@@ -39,7 +39,7 @@ using Test
                 rng, model, objective, q0, T;
                 optimizer     = Optimisers.Adam(realtype(η)),
                 show_progress = PROGRESS,
-                adbackend     = adbackend,
+                adtype        = adtype,
             )
 
             μ  = mean(q)
@@ -57,7 +57,7 @@ using Test
                 rng, model, objective, q0, T;
                 optimizer     = Optimisers.Adam(realtype(η)),
                 show_progress = PROGRESS,
-                adbackend     = adbackend,
+                adtype        = adtype,
             )
             μ  = mean(q)
             L  = sqrt(cov(q))
@@ -67,7 +67,7 @@ using Test
                 rng_repl, model, objective, q0, T;
                 optimizer     = Optimisers.Adam(realtype(η)),
                 show_progress = PROGRESS,
-                adbackend     = adbackend,
+                adtype        = adtype,
             )
             μ_repl = mean(q)
             L_repl = sqrt(cov(q))

test/inference/repgradelbo_locationscale.jl

@@ -15,7 +15,7 @@ using Test
             :RepGradELBOClosedFormEntropy  => RepGradELBO(n_montecarlo),
             :RepGradELBOStickingTheLanding => RepGradELBO(n_montecarlo, entropy = StickingTheLandingEntropy()),
         ),
-        (adbackname, adbackend) in Dict(
+        (adbackname, adtype) in Dict(
             :ForwarDiff  => AutoForwardDiff(),
             :ReverseDiff => AutoReverseDiff(),
             :Zygote      => AutoZygote(), 
@@ -43,7 +43,7 @@ using Test
                 rng, model, objective, q0, T;
                 optimizer     = Optimisers.Adam(realtype(η)),
                 show_progress = PROGRESS,
-                adbackend     = adbackend,
+                adtype        = adtype,
             )
 
             μ  = q.location
@@ -61,7 +61,7 @@ using Test
                 rng, model, objective, q0, T;
                 optimizer     = Optimisers.Adam(realtype(η)),
                 show_progress = PROGRESS,
-                adbackend     = adbackend,
+                adtype        = adtype,
             )
             μ  = q.location
             L  = q.scale
@@ -71,7 +71,7 @@ using Test
                 rng_repl, model, objective, q0, T;
                 optimizer     = Optimisers.Adam(realtype(η)),
                 show_progress = PROGRESS,
-                adbackend     = adbackend,
+                adtype        = adtype,
             )
             μ_repl = q.location
             L_repl = q.scale

test/inference/repgradelbo_locationscale_bijectors.jl

@@ -14,7 +14,7 @@ using Test
             :RepGradELBOClosedFormEntropy  => RepGradELBO(n_montecarlo),
             :RepGradELBOStickingTheLanding => RepGradELBO(n_montecarlo, entropy = StickingTheLandingEntropy()),
         ),
-        (adbackname, adbackend) in Dict(
+        (adbackname, adtype) in Dict(
             :ForwarDiff  => AutoForwardDiff(),
             :ReverseDiff => AutoReverseDiff(),
             #:Zygote      => AutoZygote(), 
@@ -48,7 +48,7 @@ using Test
                 rng, model, objective, q0_z, T;
                 optimizer     = Optimisers.Adam(realtype(η)),
                 show_progress = PROGRESS,
-                adbackend     = adbackend,
+                adtype        = adtype,
             )
 
             μ  = q.dist.location
@@ -66,7 +66,7 @@ using Test
                 rng, model, objective, q0_z, T;
                 optimizer     = Optimisers.Adam(realtype(η)),
                 show_progress = PROGRESS,
-                adbackend     = adbackend,
+                adtype        = adtype,
             )
             μ  = q.dist.location
             L  = q.dist.scale
@@ -76,7 +76,7 @@ using Test
                 rng_repl, model, objective, q0_z, T;
                 optimizer     = Optimisers.Adam(realtype(η)),
                 show_progress = PROGRESS,
-                adbackend     = adbackend,
+                adtype        = adtype,
             )
             μ_repl = q.dist.location
             L_repl = q.dist.scale

test/interface/optimize.jl

@@ -14,15 +14,15 @@ using Test
     q0  = TuringDiagMvNormal(zeros(Float64, n_dims), ones(Float64, n_dims))
     obj = RepGradELBO(10)
 
-    adbackend = AutoForwardDiff()
+    adtype    = AutoForwardDiff()
     optimizer = Optimisers.Adam(1e-2)
 
     rng  = StableRNG(seed)
     q_ref, stats_ref, _ = optimize(
         rng, model, obj, q0, T;
         optimizer,
         show_progress = false,
-        adbackend,
+        adtype,
     )
     λ_ref, _ = Optimisers.destructure(q_ref)
 
@@ -31,15 +31,15 @@ using Test
             model, obj, q0, T;
             optimizer,
             show_progress = false,
-            adbackend,
+            adtype,
         )
 
         λ₀, re  = Optimisers.destructure(q0)
         optimize(
             model, obj, re, λ₀, T;
             optimizer,
             show_progress = false,
-            adbackend,
+            adtype,
         )
     end
 
@@ -51,7 +51,7 @@ using Test
             rng, model, obj, re, λ₀, T;
             optimizer,
             show_progress = false,
-            adbackend,
+            adtype,
         )
         @test λ     == λ_ref
         @test stats == stats_ref
@@ -67,7 +67,7 @@ using Test
         _, stats, _ = optimize(
             rng, model, obj, q0, T;
             show_progress = false,
-            adbackend,
+            adtype,
             callback
         )
         @test [stat.test_value for stat ∈ stats] == test_values
@@ -83,15 +83,15 @@ using Test
             rng, model, obj, q0, T_first;
             optimizer,
             show_progress = false,
-            adbackend
+            adtype
         )
 
         q, stats, _ = optimize(
             rng, model, obj, q_first, T_last;
             optimizer,
             show_progress = false,
             state_init    = state,
-            adbackend
+            adtype
         )
         @test q == q_ref
     end