narrower output types and remove locks for bootstrap + code formatting

Author: Maximilian-Stefan-Ernst

src/frontend/fit/standard_errors/bootstrap.jl

@@ -53,11 +53,11 @@ function bootstrap(
     data = prepare_data_bootstrap(data, fitted.model)
     start = solution(fitted)
     # pre-allocations
-    out = []
-    conv = []
+    out = Vector{Any}(nothing, n_boot)
+    conv = fill(false, n_boot)
     # fit to bootstrap samples
     if !parallel
-        for _ in 1:n_boot
+        for i in 1:n_boot
             sample_data = bootstrap_sample(data)
             new_model = replace_observed(
                 fitted.model;
@@ -68,8 +68,8 @@ function bootstrap(
             new_fit = fit(new_model; start_val = start, engine = engine, fit_kwargs...)
             sample = statistic(new_fit)
             c = converged(new_fit)
-            push!(out, sample)
-            push!(conv, c)
+            out[i] = sample
+            conv[i] = c
         end
     else
         n_threads = Threads.nthreads()
@@ -80,7 +80,7 @@ function bootstrap(
         end
         # fit models in parallel
         lk = ReentrantLock()
-        Threads.@threads for _ in 1:n_boot
+        Threads.@threads for i in 1:n_boot
             thread_model = take!(model_pool)
             sample_data = bootstrap_sample(data)
             new_model = replace_observed(
@@ -92,17 +92,15 @@ function bootstrap(
             new_fit = fit(new_model; start_val = start, engine = engine, fit_kwargs...)
             sample = statistic(new_fit)
             c = converged(new_fit)
-            lock(lk) do
-                push!(out, sample)
-                push!(conv, c)
-            end
+            out[i] = sample
+            conv[i] = c
             put!(model_pool, thread_model)
         end
     end
     return Dict(
-        :samples => out,
+        :samples => collect(a for a in out),
         :n_boot => n_boot,
-        :n_converged => isempty(conv) ? 0 : sum(conv),
+        :n_converged => sum(conv),
         :converged => conv,
     )
 end

src/implied/RAM/generic.jl

@@ -203,6 +203,7 @@ function update_observed(implied::RAM, observed::SemObserved; kwargs...)
             observed = observed,
             gradient_required = !isnothing(implied.∇A),
             meanstructure = MeanStruct(implied) == HasMeanStruct,
-            kwargs...)
+            kwargs...,
+        )
     end
 end

src/implied/RAM/symbolic.jl

@@ -214,13 +214,14 @@ function update_observed(implied::RAMSymbolic, observed::SemObserved; kwargs...)
         return implied
     else
         return RAMSymbolic(;
-        observed = observed,
-        vech = implied.Σ isa Vector,
-        gradient = !isnothing(implied.∇Σ),
-        hessian = !isnothing(implied.∇²Σ),
-        meanstructure = MeanStruct(implied) == HasMeanStruct,
-        approximate_hessian = isnothing(implied.∇²Σ),
-        kwargs...)
+            observed = observed,
+            vech = implied.Σ isa Vector,
+            gradient = !isnothing(implied.∇Σ),
+            hessian = !isnothing(implied.∇²Σ),
+            meanstructure = MeanStruct(implied) == HasMeanStruct,
+            approximate_hessian = isnothing(implied.∇²Σ),
+            kwargs...,
+        )
     end
 end
 

src/loss/ML/ML.jl

@@ -238,8 +238,9 @@ function update_observed(lossfun::SemML, observed::SemObserved; kwargs...)
         return lossfun
     else
         return SemML(;
-        observed = observed,
-        approximate_hessian = HessianEval(lossfun) == ApproxHessian,
-        kwargs...)
+            observed = observed,
+            approximate_hessian = HessianEval(lossfun) == ApproxHessian,
+            kwargs...,
+        )
     end
 end

src/loss/WLS/WLS.jl

@@ -178,14 +178,16 @@ function update_observed(lossfun::SemWLS, observed::SemObserved; recompute_V = t
         return SemWLS(;
             observed = observed,
             meanstructure = MeanStruct(kwargs[:implied]) == HasMeanStruct,
-            kwargs...)
+            kwargs...,
+        )
     else
         return SemWLS(;
             observed = observed,
             wls_weight_matrix = lossfun.V,
             wls_weight_matrix_mean = lossfun.V_μ,
             meanstructure = MeanStruct(kwargs[:implied]) == HasMeanStruct,
-            kwargs...)
+            kwargs...,
+        )
 
     end
 end

test/examples/helper.jl

@@ -144,7 +144,8 @@ function test_bootstrap(
         rtol_hessian = 0.2,
         compare_bs = true,
         rtol_bs = 0.1,
-        n_boot = 500)
+        n_boot = 500,
+    )
     @testset rng = Random.seed!(32432) "bootstrap" begin
         se_bs = @suppress se_bootstrap(model_fit, spec; n_boot = n_boot)
         # hessian and bootstrap se are close
@@ -157,7 +158,8 @@ function test_bootstrap(
         if compare_bs
             bs_samples = bootstrap(model_fit, spec; n_boot = n_boot)
             @test bs_samples[:n_converged] >= 0.95*n_boot
-            bs_samples = cat(bs_samples[:samples][BitVector(bs_samples[:converged])]..., dims = 2)
+            bs_samples = 
+                cat(bs_samples[:samples][BitVector(bs_samples[:converged])]..., dims = 2)
             se_bs_2 = sqrt.(var(bs_samples, corrected = false, dims = 2))
             #println(maximum(abs.(se_bs_2 - se_bs)))
             @test isapprox(se_bs_2, se_bs, rtol = rtol_bs)

test/examples/multigroup/build_models.jl

@@ -253,7 +253,12 @@ model_ls_g2 = Sem(
     loss = SemWLS,
 )
 
-model_ls_multigroup = SemEnsemble(model_ls_g1, model_ls_g2; groups = [:Pasteur, :Grant_White], optimizer = semoptimizer)
+model_ls_multigroup = SemEnsemble(
+    model_ls_g1,
+    model_ls_g2;
+    groups = [:Pasteur, :Grant_White],
+    optimizer = semoptimizer,
+)
 
 @testset "ls_gradients_multigroup" begin
     test_gradient(model_ls_multigroup, start_test; atol = 1e-9)