add some single benchmarks

bm_results.jl

@@ -0,0 +1,99 @@
+bm_nonsymbolic_small = @benchmark sem_fit(model_ml_small)
+
+BenchmarkTools.Trial: 7037 samples with 1 evaluation.
+ Range (min … max):  536.094 μs …  13.142 ms  ┊ GC (min … max): 0.00% … 84.11%
+ Time  (median):     623.008 μs               ┊ GC (median):    0.00%
+ Time  (mean ± σ):   703.678 μs ± 577.872 μs  ┊ GC (mean ± σ):  4.18% ±  4.91%
+
+     ▃█▇▅▃▁
+  ▃▄███████▇▆▆▄▄▄▃▃▃▃▃▃▂▂▂▂▂▂▂▂▂▂▁▂▂▁▁▂▁▁▂▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁ ▂
+  536 μs           Histogram: frequency by time         1.24 ms <
+
+ Memory estimate: 439.34 KiB, allocs estimate: 696.
+
+bm_nonsymbolic_big = @benchmark sem_fit(model_ml_big)
+
+BenchmarkTools.Trial: 10 samples with 1 evaluation.
+ Range (min … max):  377.030 ms … 742.732 ms  ┊ GC (min … max): 0.00% … 0.62%
+ Time  (median):     516.519 ms               ┊ GC (median):    0.00%
+ Time  (mean ± σ):   538.650 ms ± 122.110 ms  ┊ GC (mean ± σ):  0.36% ± 0.49%
+
+  ▁ ▁           █ ▁            ▁         ▁   ▁  ▁             ▁
+  █▁█▁▁▁▁▁▁▁▁▁▁▁█▁█▁▁▁▁▁▁▁▁▁▁▁▁█▁▁▁▁▁▁▁▁▁█▁▁▁█▁▁█▁▁▁▁▁▁▁▁▁▁▁▁▁█ ▁
+  377 ms           Histogram: frequency by time          743 ms <
+
+ Memory estimate: 56.59 MiB, allocs estimate: 956.
+
+# MKL
+
+using MKL
+
+bm_nonsymbolic_small_mkl = @benchmark sem_fit(model_ml_small)
+
+BenchmarkTools.Trial: 4948 samples with 1 evaluation.
+ Range (min … max):  712.489 μs …  12.225 ms  ┊ GC (min … max): 0.00% … 82.20%
+ Time  (median):     899.797 μs               ┊ GC (median):    0.00%
+ Time  (mean ± σ):     1.003 ms ± 549.112 μs  ┊ GC (mean ± σ):  2.60% ±  4.65%
+
+        ▂▁█▄▁
+  ▁▂▂▃▃▄██████▆▅▅▅▄▃▃▃▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▁▂▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁ ▂
+  712 μs           Histogram: frequency by time         1.78 ms <
+
+ Memory estimate: 439.34 KiB, allocs estimate: 696.
+
+bm_nonsymbolic_big_mkl = @benchmark sem_fit(model_ml_big)
+
+BenchmarkTools.Trial: 21 samples with 1 evaluation.
+ Range (min … max):  149.162 ms … 401.825 ms  ┊ GC (min … max): 0.00% … 0.00%
+ Time  (median):     243.611 ms               ┊ GC (median):    0.00%
+ Time  (mean ± σ):   249.148 ms ±  53.616 ms  ┊ GC (mean ± σ):  0.54% ± 0.63%
+
+  ▁   ▁      ▁    ▁█▁█▁ ▁ ▁ ▁▁█▁    ▁▁     ▁                  ▁
+  █▁▁▁█▁▁▁▁▁▁█▁▁▁▁█████▁█▁█▁████▁▁▁▁██▁▁▁▁▁█▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁█ ▁
+  149 ms           Histogram: frequency by time          402 ms <
+
+ Memory estimate: 56.59 MiB, allocs estimate: 956.
+
+# BFGS
+
+algo = BFGS(;linesearch = BackTracking(order=3), alphaguess = InitialHagerZhang())
+
+bm_algo_small = @benchmark sem_fit(model_ml_small)
+
+BenchmarkTools.Trial: 3270 samples with 1 evaluation.
+ Range (min … max):  937.452 μs … 21.758 ms  ┊ GC (min … max): 0.00% … 0.00%
+ Time  (median):       1.255 ms              ┊ GC (median):    0.00%
+ Time  (mean ± σ):     1.510 ms ±  1.043 ms  ┊ GC (mean ± σ):  2.31% ± 4.83%
+
+   ▆█▄▂
+  ▆█████▇▆▆▅▄▄▅▄▄▄▄▄▄▄▄▄▄▄▃▃▃▃▂▂▂▂▂▂▂▂▂▁▂▂▂▂▁▁▂▁▁▁▂▂▁▁▁▁▂▂▁▁▁▂ ▃
+  937 μs          Histogram: frequency by time          4.1 ms <
+
+ Memory estimate: 439.34 KiB, allocs estimate: 696.
+
+bm_algo_big = @benchmark sem_fit(model_ml_big)
+
+BenchmarkTools.Trial: 14 samples with 1 evaluation.
+ Range (min … max):  274.281 ms … 578.298 ms  ┊ GC (min … max): 0.00% … 0.58%
+ Time  (median):     340.173 ms               ┊ GC (median):    0.47%
+ Time  (mean ± σ):   358.096 ms ±  82.599 ms  ┊ GC (mean ± σ):  0.54% ± 0.57%
+
+  ▁ ▁█▁    ▁ ▁  ▁  ▁▁  ▁         ▁   ▁                        ▁
+  █▁███▁▁▁▁█▁█▁▁█▁▁██▁▁█▁▁▁▁▁▁▁▁▁█▁▁▁█▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁█ ▁
+  274 ms           Histogram: frequency by time          578 ms <
+
+ Memory estimate: 56.59 MiB, allocs estimate: 956.
+
+
+# profiling
+
+ProfileView.@profview profile_test(model_ml_big, 10)
+
+- small model: matrix inversion takes more time
+
+big model takes more time
+- SemML: 153
+
+ML.jl, 221, uniformscaling bad performance for small model
+
+- factorize upper triangular

cfa/single_benchmarks.jl

@@ -0,0 +1,115 @@
+using DataFrames, StructuralEquationModels, Symbolics, 
+    LinearAlgebra, SparseArrays, Optim, LineSearches,
+    BenchmarkTools, CSV, Statistics
+
+cd("cfa")
+
+include("functions.jl")
+
+config = DataFrame(CSV.File("config.csv"))
+config2 = copy(config)
+config.backend .= "Optim.jl"
+config2.backend .= "NLopt.jl"
+
+config = [config; config2]
+
+config = filter(row -> (row.backend == "Optim.jl") & (row.meanstructure == 0), config)
+
+data_vec = read_files("data", get_data_paths(config))
+par_vec = read_files("parest", get_data_paths(config))
+# start_vec = read_files("start", get_data_paths(config))
+
+##############################################
+
+models = gen_models(config, data_vec)
+
+# fits = get_fits(models)
+
+# correct = compare_estimates(fits, par_vec, config)
+
+# config.correct = correct
+
+############################################
+# some benchmarks
+############################################
+
+using BenchmarkTools
+
+semdiff = SemDiffOptim(
+    LBFGS(
+        ;linesearch = BackTracking(order=3), 
+        alphaguess = InitialHagerZhang()
+        ),
+    Optim.Options(
+        ;f_tol = 1e-10,
+        x_tol = 1.5e-8)
+    )
+
+ram_matrices_small = gen_CFA_RAM(3, 5)
+
+ram_matrices_big = gen_CFA_RAM(5, 40)
+
+model_ml_small = Sem(
+    ram_matrices = ram_matrices_small,
+    data = Matrix(data_vec[2]),
+    imply = RAM,
+    diff = semdiff 
+)
+
+model_ml_big = Sem(
+    ram_matrices = ram_matrices_big,
+    data = Matrix(data_vec[6]),
+    imply = RAM,
+    diff = semdiff 
+)
+
+############################################
+# some benchmarks
+############################################
+
+bm_nonsymbolic_small = @benchmark sem_fit(model_ml_small)
+
+bm_nonsymbolic_big = @benchmark sem_fit(model_ml_big)
+
+# MKL
+
+using MKL
+
+bm_nonsymbolic_small_mkl = @benchmark sem_fit(model_ml_small)
+
+bm_nonsymbolic_big_mkl = @benchmark sem_fit(model_ml_big)
+
+# BFGS
+
+algo = BFGS(;linesearch = BackTracking(order=3), alphaguess = InitialHagerZhang())
+
+model_ml_small = Sem(
+    ram_matrices = ram_matrices_small,
+    data = Matrix(data_vec[2]),
+    imply = RAM,
+    diff = semdiff,
+    algorithm = algo
+)
+
+model_ml_big = Sem(
+    ram_matrices = ram_matrices_big,
+    data = Matrix(data_vec[6]),
+    imply = RAM,
+    diff = semdiff,
+    algorithm = algo
+)
+
+bm_algo_small = @benchmark sem_fit(model_ml_small)
+
+bm_algo_big = @benchmark sem_fit(model_ml_big)
+
+
+using ProfileView
+
+function profile_test(model, n)
+    for _ = 1:n
+        sem_fit(model)
+    end
+end
+
+ProfileView.@profview profile_test(model_ml_big, 10)