Addded Airspeedvelocity benchmarks

.github/workflows/TagBot.yml

@@ -6,7 +6,6 @@ on:
 jobs:
   TagBot:
     if: github.event_name == 'workflow_dispatch' || github.actor == 'JuliaTagBot'
-    # ubuntu-slim doesn't support containers, and thus won't work
     runs-on: ubuntu-latest
     steps:
       - uses: JuliaRegistries/TagBot@v1.25.11

.github/workflows/benchmark.yml

@@ -0,0 +1,27 @@
+name: Benchmark this PR
+
+on:
+  pull_request:
+    branches: [main]
+
+permissions:
+  pull-requests: write
+  contents: read
+
+jobs:
+  benchmark:
+    runs-on: ubuntu-latest
+    timeout-minutes: 30
+    strategy:
+      fail-fast: false
+      matrix:
+        version: ["1"]
+    env:
+      # Force consistent Julia depot path for self-hosted runners
+      JULIA_DEPOT_PATH: ~/.julia
+    steps:
+      - uses: MilesCranmer/AirspeedVelocity.jl@action-v1
+        with:
+          julia-version: ${{ matrix.version }}
+          script: "benchmark/benchmarks.jl"
+          job-summary: true

benchmark/benchmarks.jl

@@ -0,0 +1,35 @@
+using BenchmarkTools
+using Random, LinearAlgebra
+using NearestCorrelationMatrix
+using NearestCorrelationMatrix.Internals: symmetrize!
+
+Random.seed!(0x00c0ffee)
+
+const SUITE = BenchmarkGroup()
+
+function rand_negdef(n::Int)
+    while true
+        A = 2.0 * rand(Float64, n, n) .- 1.0
+        symmetrize!(A)
+        A[diagind(A)] .= 1.0
+        !isposdef(A) && return A
+    end
+    return zeros(Float64, 0, 0)
+end
+
+function create_benchmarkable(n, alg; evals, samples, seconds)
+    return @benchmarkable nearest_cor(A, $alg) evals = evals samples = samples seconds = seconds setup = (A = rand_negdef($n))
+end
+
+function create_benchmark_group(n::Int; evals = 5, samples = 100, seconds = 60)
+    grp = BenchmarkGroup()
+    grp["Nw"] = create_benchmarkable(n, Newton; evals, samples, seconds)
+    grp["AP"] = create_benchmarkable(n, AlternatingProjections; evals, samples, seconds)
+    grp["AA"] = create_benchmarkable(n, AcceleratedAP; evals, samples, seconds)
+    grp["Di"] = create_benchmarkable(n, DirectProjection; evals, samples, seconds)
+    return grp
+end
+
+SUITE["n=10"] = create_benchmark_group(10; samples = 30)
+SUITE["n=100"] = create_benchmark_group(100; samples = 30)
+SUITE["n=1000"] = create_benchmark_group(1000; samples = 30, seconds = 180)

src/internals/Internals.jl

@@ -5,6 +5,7 @@ using LinearAlgebra
 include("common.jl")
 include("projections.jl")
 include("checks.jl")
+include("gen_cor.jl")
 
 include("newton_internals.jl")
 

src/internals/gen_cor.jl

@@ -0,0 +1,51 @@
+export default_negdef, rand_negdef
+
+"""
+    default_negdef([T=Float64]; include_mask=false)
+
+Returns a 4×4 invalid correlation matrix and an optional mask.
+"""
+function default_negdef(::Type{T}; include_mask::Bool = false) where {T <: AbstractFloat}
+    # !WARNING! The data in this generator must never be edited to ensure consistency with
+    # future releases.
+
+    A = Float64[
+        1.0 -0.2188 -0.79 0.7773
+        -0.2188 1.0 0.2559 -0.5977
+        -0.79 0.2559 1.0 0.2266
+        0.7773 -0.5977 0.2266 1.0
+    ]
+
+    mask = [
+        1 1 0 0
+        1 1 0 0
+        0 0 1 1
+        0 0 1 1
+    ]
+
+    X = convert(AbstractMatrix{T}, A)
+    m = convert(BitMatrix, mask)
+
+    return include_mask ? (X, m) : X
+end
+
+default_negdef(; include_mask::Bool = false) = default_negdef(Float64; include_mask)
+
+"""
+    rand_negdef([T=Float64], n; max_attempts=100)
+
+Generates a random pseudo-correlation matrix that is negative definite.
+"""
+function rand_negdef(::Type{T}, n::Int; max_attempts::Int = 100) where {T <: AbstractFloat}
+    attempts = 0
+    while attempts < max_attempts
+        r = 2 * rand(T, n, n) .- one(T)
+        symmetrize!(r)
+        r[diagind(r)] .= one(T)
+        !isposdef(r) && return r
+        attempts += 1
+    end
+    return Matrix{T}(undef, 0, 0)
+end
+
+rand_negdef(n::Int; max_attempts::Int = 100) = rand_negdef(Float64, n; max_attempts)

test/Datasets.jl

@@ -5,7 +5,6 @@ using LinearAlgebra
 import MAT
 
 export
-    default_negdef,
     bccd16,
     beyu11,
     bhwi01,
@@ -48,24 +47,6 @@ function vec_to_mat(x::AbstractVector{T}; diag_val::Real = 1) where {T}
     return A
 end
 
-"""
-    default_negdef(T)
-
-Gets a 4×4 invalid correlation matrix for testing.
-"""
-function default_negdef(::Type{T}) where {T}
-    r = [
-        1.0 -0.2188 -0.79 0.7773
-        -0.2188 1.0 0.2559 -0.5977
-        -0.79 0.2559 1.0 0.2266
-        0.7773 -0.5977 0.2266 1.0
-    ]
-
-    return convert(Matrix{T}, r)
-end
-
-default_negdef() = default_negdef(Float64)
-
 """
 BCCD16 is a 3250×3250 invalid correlation matrix constructed from data for banks in 27 EU member states.
 
@@ -146,13 +127,13 @@ Source:
 """
 function fing97()
     A = Float64[
-        1.0  0.18 -0.13 -0.26  0.19 -0.25 -0.12
-        0.18  1.0  0.22 -0.14  0.31  0.16  0.09
-        -0.13  0.22  1.0  0.06 -0.08  0.04  0.04
-        -0.26 -0.14  0.06  1.0  0.85  0.85  0.85
-        0.19  0.31 -0.08  0.85  1.0  0.85  0.85
-        -0.25  0.16  0.04  0.85  0.85  1.0  0.85
-        -0.12  0.09  0.04  0.85  0.85  0.85  1.0
+        1.0 0.18 -0.13 -0.26 0.19 -0.25 -0.12
+        0.18 1.0 0.22 -0.14 0.31 0.16 0.09
+        -0.13 0.22 1.0 0.06 -0.08 0.04 0.04
+        -0.26 -0.14 0.06 1.0 0.85 0.85 0.85
+        0.19 0.31 -0.08 0.85 1.0 0.85 0.85
+        -0.25 0.16 0.04 0.85 0.85 1.0 0.85
+        -0.12 0.09 0.04 0.85 0.85 0.85 1.0
     ]
 
     mask = [trues(3, 3) falses(3, 4); falses(4, 3) I(4)]
@@ -182,12 +163,12 @@ St. Petersburg, Russia, 2013. Revised June 2014.
 """
 function mmb13()
     A = Float64[
-        0.010712 0.000654  0.002391  0.010059 -0.008321  0.001738
-        0.000654 0.000004  0.002917  0.00065  0.002263  0.002913
-        0.002391 0.002917  0.013225 -0.000525  0.010834  0.010309
-        0.010059 0.00065 -0.000525  0.009409 -0.010584 -0.001175
-        -0.008321 0.002263  0.010834 -0.010584  0.019155  0.008571
-        0.001738 0.002913  0.010309 -0.001175  0.008571  0.007396
+        0.010712 0.000654 0.002391 0.010059 -0.008321 0.001738
+        0.000654 0.000004 0.002917 0.00065 0.002263 0.002913
+        0.002391 0.002917 0.013225 -0.000525 0.010834 0.010309
+        0.010059 0.00065 -0.000525 0.009409 -0.010584 -0.001175
+        -0.008321 0.002263 0.010834 -0.010584 0.019155 0.008571
+        0.001738 0.002913 0.010309 -0.001175 0.008571 0.007396
     ]
 
     d = sqrt.(diag(A))
@@ -206,10 +187,10 @@ stress testing for value-at-risk. Journal of Risk, 5(4):75-89, 2003
 """
 function tec03()
     return Float64[
-        1   -0.55 -0.15 -0.1
-        -0.55 1     0.9  0.9
-        -0.15 0.9  1     0.9
-        -0.1 0.9  0.9  1
+        1 -0.55 -0.15 -0.1
+        -0.55 1 0.9 0.9
+        -0.15 0.9 1 0.9
+        -0.1 0.9 0.9 1
     ]
 end
 

test/test_api.jl

@@ -1,21 +1,19 @@
 using Test
 using InteractiveUtils
 using NearestCorrelationMatrix
-
-include("Datasets.jl")
-using .Datasets
+using NearestCorrelationMatrix.Internals: default_negdef
 
 include("CustomTestMacros.jl")
 using .CustomTestMacros
 
 internal_algtypes = setdiff(subtypes(NCMAlgorithm), (JuMPAlgorithm,))
 
-r0 = default_negdef(Float64)
+A = default_negdef()
 
 # variants of NCMProblem
 @test_isdefined NCMProblem
-@test_isimplemented NCMProblem(r0)
-prob = NCMProblem(r0)
+@test_isimplemented NCMProblem(A)
+prob = NCMProblem(A)
 
 # variations of init
 @test_isdefined init

test/test_convergence.jl

@@ -2,20 +2,18 @@ using Test
 using InteractiveUtils
 using LinearAlgebra
 using NearestCorrelationMatrix
+using NearestCorrelationMatrix.Internals: default_negdef
 import NearestCorrelationMatrix as NCM
 
-include("Datasets.jl")
-using .Datasets
-
 include("CustomTestMacros.jl")
 using .CustomTestMacros
 
 internal_algtypes = setdiff(subtypes(NCMAlgorithm), (JuMPAlgorithm,))
 
 function test_simple(algtype)
     return @testset "$(NCM.alg_name(algtype))" begin
-        r0 = default_negdef(Float64)
-        prob = NCMProblem(r0)
+        A = default_negdef()
+        prob = NCMProblem(A)
         alg = autotune(algtype, prob)
         cache = init(prob, alg)
         sol = solve!(cache)
@@ -26,14 +24,14 @@ function test_simple(algtype)
         @test sol.X isa Symmetric
 
         # Handle Symmetric type matrices
-        r0 = default_negdef(Float64)
-        prob = NCMProblem(Symmetric(r0))
+        A = default_negdef()
+        prob = NCMProblem(Symmetric(A))
         alg = autotune(algtype, prob)
         @test_nothrow solve(prob, alg)
 
         # Handle Float16 input matrices
-        r0 = default_negdef(Float16)
-        prob = NCMProblem(r0)
+        A = default_negdef(Float16)
+        prob = NCMProblem(A)
         alg = autotune(algtype, prob)
         if NCM.supports_float16(alg)
             @test_nothrow solve(prob, alg)

test/test_jump.jl

@@ -1,15 +1,13 @@
 using Test
 using NearestCorrelationMatrix
+using NearestCorrelationMatrix.Internals: default_negdef
 using JuMP, COSMO
 
-include("Datasets.jl")
-using .Datasets
-
 include("CustomTestMacros.jl")
 using .CustomTestMacros
 
-r0 = default_negdef(Float64)
-prob = NCMProblem(r0)
+A = default_negdef()
+prob = NCMProblem(A)
 
 @test_isdefined JuMPAlgorithm
 @test_isimplemented JuMPAlgorithm(COSMO.Optimizer)

test/test_masking.jl

@@ -2,11 +2,9 @@ using Test
 using LinearAlgebra
 using InteractiveUtils
 using NearestCorrelationMatrix
+using NearestCorrelationMatrix.Internals: default_negdef
 import NearestCorrelationMatrix as NCM
 
-include("Datasets.jl")
-using .Datasets
-
 include("CustomTestMacros.jl")
 using .CustomTestMacros
 
@@ -15,8 +13,7 @@ supported_types = (Float64, Float32, Float16)
 
 for algtype in masking_algs, T in supported_types
     @testset "$(NCM.alg_name(algtype)) - $T" begin
-        A, m = usgs13()
-        A = convert(Matrix{T}, A)
+        A, m = default_negdef(T; include_mask = true)
         X = copy(A)
         prob = NCMProblem(X; mask = m)
         alg = autotune(algtype, prob)

test/test_robustness.jl

@@ -1,20 +1,17 @@
-function rand_negdef(::Type{T}, n) where {T}
-    while true
-        r = 2 * rand(T, n, n) .- one(T)
-        symmetrize!(r)
-        r[diagind(r)] .= one(T)
-
-        !isposdef(r) && return r
-    end
-    return
-end
+using Test
+using LinearAlgebra
+using NearestCorrelationMatrix
+using NearestCorrelationMatrix.Internals: rand_negdef
+using JuMP, COSMO
 
+include("CustomTestMacros.jl")
+using .CustomTestMacros
 
 function test_robust_reps(algtype::Type, nreps, size, T, test_pd; kwargs...)
     return @testset "$(size)×$(size)" begin
         for _ in 1:nreps
-            r0 = rand_negdef(T, size)
-            prob = NCMProblem(r0)
+            A = rand_negdef(T, size)
+            prob = NCMProblem(A)
             alg = autotune(algtype, prob)
             cache = init(prob, alg; kwargs...)
             sol = solve!(cache)
@@ -31,8 +28,8 @@ end
 function test_robust_reps(alg::NCMAlgorithm, nreps, size, T, test_pd; kwargs...)
     return @testset "$(size)×$(size)" begin
         for _ in 1:nreps
-            r0 = rand_negdef(T, size)
-            prob = NCMProblem(r0)
+            A = rand_negdef(T, size)
+            prob = NCMProblem(A)
             cache = init(prob, alg; kwargs...)
             sol = solve!(cache)