remove n_montecarlo option in the inference tests and just fix it

Author: Red-Portal

test/inference/repgradelbo_distributionsad.jl

@@ -14,7 +14,6 @@ end
     @testset "$(modelname) $(objname) $(realtype) $(adbackname)" for realtype in
                                                                      [Float64, Float32],
         (modelname, modelconstr) in Dict(:Normal => normal_meanfield),
-        n_montecarlo in [10],
         (objname, objective) in Dict(
             :RepGradELBOClosedFormEntropy => RepGradELBO(n_montecarlo),
             :RepGradELBOStickingTheLanding =>
@@ -31,6 +30,7 @@ end
         T = 1000
         η = 1e-3
         opt = Optimisers.Descent(realtype(η))
+        n_montecarlo = 10
 
         # For small enough η, the error of SGD, Δλ, is bounded as
         #     Δλ ≤ ρ^T Δλ0 + O(η),

test/inference/repgradelbo_locationscale.jl

@@ -15,7 +15,6 @@ end
                                                                      [Float64, Float32],
         (modelname, modelconstr) in
         Dict(:Normal => normal_meanfield, :Normal => normal_fullrank),
-        n_montecarlo in [10],
         (objname, objective) in Dict(
             :RepGradELBOClosedFormEntropy => RepGradELBO(n_montecarlo),
             :RepGradELBOStickingTheLanding =>
@@ -32,6 +31,7 @@ end
         T = 1000
         η = 1e-3
         opt = Optimisers.Descent(realtype(η))
+        n_montecarlo = 10
 
         # For small enough η, the error of SGD, Δλ, is bounded as
         #     Δλ ≤ ρ^T Δλ0 + O(η),

test/inference/repgradelbo_locationscale_bijectors.jl

@@ -15,7 +15,6 @@ end
                                                                      [Float64, Float32],
         (modelname, modelconstr) in
         Dict(:NormalLogNormalMeanField => normallognormal_meanfield),
-        n_montecarlo in [10],
         (objname, objective) in Dict(
             :RepGradELBOClosedFormEntropy => RepGradELBO(n_montecarlo),
             :RepGradELBOStickingTheLanding =>
@@ -32,6 +31,7 @@ end
         T = 1000
         η = 1e-3
         opt = Optimisers.Descent(realtype(η))
+        n_montecarlo = 10
 
         b = Bijectors.bijector(model)
         b⁻¹ = inverse(b)

test/inference/scoregradelbo_distributionsad.jl

@@ -14,7 +14,6 @@ end
     @testset "$(modelname) $(objname) $(realtype) $(adbackname)" for realtype in
                                                                      [Float64, Float32],
         (modelname, modelconstr) in Dict(:Normal => normal_meanfield),
-        n_montecarlo in [10],
         (objname, objective) in Dict(:ScoreGradELBO => ScoreGradELBO(n_montecarlo)),
         (adbackname, adtype) in AD_scoregradelbo_distributionsad
 
@@ -27,6 +26,7 @@ end
         T = 1000
         η = 1e-4
         opt = Optimisers.Descent(realtype(η))
+        n_montecarlo = 10
 
         # For small enough η, the error of SGD, Δλ, is bounded as
         #     Δλ ≤ ρ^T Δλ0 + O(η),

test/inference/scoregradelbo_locationscale.jl

@@ -15,7 +15,6 @@ end
                                                                      [Float64, Float32],
         (modelname, modelconstr) in
         Dict(:Normal => normal_meanfield, :Normal => normal_fullrank),
-        n_montecarlo in [10],
         (objname, objective) in Dict(:ScoreGradELBO => ScoreGradELBO(n_montecarlo)),
         (adbackname, adtype) in AD_scoregradelbo_locationscale
 
@@ -28,6 +27,7 @@ end
         T = 1000
         η = 1e-4
         opt = Optimisers.Descent(realtype(η))
+        n_montecarlo = 10
 
         # For small enough η, the error of SGD, Δλ, is bounded as
         #     Δλ ≤ ρ^T Δλ0 + O(η),

test/inference/scoregradelbo_locationscale_bijectors.jl

@@ -15,7 +15,6 @@ end
                                                                      [Float64, Float32],
         (modelname, modelconstr) in
         Dict(:NormalLogNormalMeanField => normallognormal_meanfield),
-        n_montecarlo in [10],
         (objname, objective) in Dict(:ScoreGradELBO => ScoreGradELBO(n_montecarlo)),
         (adbackname, adtype) in AD_scoregradelbo_locationscale_bijectors
 
@@ -28,6 +27,7 @@ end
         T = 1000
         η = 1e-4
         opt = Optimisers.Descent(realtype(η))
+        n_montecarlo = 10
 
         b = Bijectors.bijector(model)
         b⁻¹ = inverse(b)