Skip to content

Commit 938a2fc

Browse files
lrnvlrnv
committed
let's see if chairmarks works correctly
1 parent 166544d commit 938a2fc

File tree

2 files changed

+111
-204
lines changed

2 files changed

+111
-204
lines changed

‎benchmark/Project.toml‎

Lines changed: 1 addition & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -1,5 +1,6 @@
11
[deps]
22
BenchmarkTools = "6e4b80f9-dd63-53aa-95a3-0cdb28fa8baf"
3+
Chairmarks = "0ca39b1e-fe0b-4e98-acfc-b1656634c4de"
34
Copulas = "ae264745-0b69-425e-9d9d-cf662c5eec93"
45
Distributions = "31c24e10-a181-5473-b8eb-7969acd0382f"
56
PkgBenchmark = "32113eaa-f34f-5b0d-bd6c-c81e245fc73d"

‎benchmark/benchmarks.jl‎

Lines changed: 110 additions & 204 deletions
Original file line numberDiff line numberDiff line change
@@ -1,216 +1,122 @@
1+
using Chairmarks
12
using BenchmarkTools
23
using Copulas
34
using StatsBase
45
using Distributions
56
using StableRNGs
6-
const rng = StableRNG(123)
77

8-
# PkgBenchmark entrypoint; must define SUITE
9-
const SUITE = BenchmarkGroup()
10-
11-
12-
const EXAMPLES = unique([
13-
# AMH: keep one non-2D representative
14-
AMHCopula(3, 0.6),
15-
16-
# Archimax: cover each generator family and each tail type at least once (no full cartesian product)
17-
ArchimaxCopula(2, Copulas.BB1Generator(1.3, 1.4), Copulas.GalambosTail(0.7)),
18-
ArchimaxCopula(2, Copulas.ClaytonGenerator(1.5), Copulas.HuslerReissTail(0.6)),
19-
ArchimaxCopula(2, Copulas.FrankGenerator(6.0), Copulas.LogTail(1.5)),
20-
ArchimaxCopula(2, Copulas.GumbelGenerator(2.0), Copulas.AsymGalambosTail(0.35, 0.65, 0.3)),
21-
ArchimaxCopula(2, Copulas.JoeGenerator(2.5), Copulas.GalambosTail(2.5)),
22-
23-
# Archimedean: one empirical generator (non-2D)
24-
ArchimedeanCopula(3, EmpiricalGenerator(randn(rng, 3, 200))),
25-
26-
# Asymmetric families: one representative each
27-
AsymGalambosCopula(2, 5.0, 0.8, 0.3),
28-
AsymLogCopula(2, 1.5, 0.5, 0.2),
29-
AsymMixedCopula(2, 0.12, 0.13),
30-
31-
# BB families: one per family
32-
BB10Copula(2, 3.0, 0.8),
33-
BB1Copula(2, 1.2, 1.5),
34-
BB2Copula(2, 1.5, 1.8),
35-
BB3Copula(2, 2.5, 0.5),
36-
BB6Copula(2, 1.5, 1.4),
37-
BB7Copula(2, 1.5, 0.4),
38-
BB8Copula(2, 1.5, 0.6),
39-
BB9Copula(2, 2.0, 1.5),
40-
41-
# BC2: symmetric and asymmetric
42-
BC2Copula(2, 0.5, 0.5),
43-
BC2Copula(2, 0.7, 0.3),
44-
45-
# Bernstein copulas: a couple of bases
46-
BernsteinCopula(ClaytonCopula(3, 3.3); m=5),
47-
BernsteinCopula(randn(rng, 3,100); m=5, pseudo_values=false),
48-
49-
# Beta & Checkerboard: prefer non-2D where available
50-
BetaCopula(randn(rng, 3,100)),
51-
CheckerboardCopula(randn(rng, 3,100); pseudo_values=false),
52-
53-
# Clayton: keep one negative and one positive; plus one non-2D
54-
ClaytonCopula(2, -0.7),
55-
ClaytonCopula(4, 3.0),
56-
57-
# Subset copula
58-
Copulas.SubsetCopula(RafteryCopula(3, 0.5), (2,1)),
59-
60-
# Cuadras-Auge: two extremes
61-
CuadrasAugeCopula(2, 0.0),
62-
CuadrasAugeCopula(2, 0.8),
63-
64-
# Empirical copulas
65-
EmpiricalCopula(randn(2,20), pseudo_values=false),
66-
EmpiricalEVCopula(randn(rng, 2,20); method=:pickands, pseudo_values=false),
67-
68-
# FGM: one 2D max and one 3D vector case
69-
FGMCopula(2, 1),
70-
FGMCopula(3, [0.1,0.2,0.3,0.4]),
71-
72-
# Frank: negative and positive, with a non-2D example
73-
FrankCopula(2, -5),
74-
FrankCopula(4, 30),
75-
76-
# Galambos: small and large parameter
77-
GalambosCopula(2, 4.3),
78-
GalambosCopula(2, 120),
79-
80-
# Gaussian
81-
GaussianCopula([1 0.7; 0.7 1]),
82-
83-
# Gumbel-Barnett
84-
GumbelBarnettCopula(2, 1.0),
85-
GumbelBarnettCopula(3, 0.35),
86-
87-
# Gumbel: 2D and non-2D
88-
GumbelCopula(2, 1.2),
89-
GumbelCopula(4, 7.0),
90-
91-
# Hüsler–Reiss
92-
HuslerReissCopula(2, 3.5),
93-
94-
# Independent: prefer non-2D
95-
IndependentCopula(4),
96-
97-
# Inverse Gaussian: 2D and non-2D
98-
InvGaussianCopula(2, 1.0),
99-
InvGaussianCopula(4, 0.05),
100-
101-
# Joe: include ∞ special case and one non-2D
102-
JoeCopula(2, Inf),
103-
JoeCopula(3, 7),
104-
105-
# Log copula
106-
LogCopula(2, 1.5),
107-
108-
# Marshall–Olkin variants
109-
MCopula(4),
110-
MixedCopula(2, 0.5),
111-
MOCopula(2, 0.5, 0.5, 0.5),
112-
113-
# Plackett
114-
PlackettCopula(2.0),
115-
116-
# Raftery: prefer non-2D
117-
RafteryCopula(3, 0.5),
118-
119-
# t-Copula: prefer non-2D
120-
TCopula(4, [1 0.5; 0.5 1]),
121-
122-
# t-EV: one representative
123-
tEVCopula(2, 4.0, 0.5),
124-
125-
# W copula
126-
WCopula(2),
127-
]);
128-
129-
let rng = StableRNG(123)
130-
top = SUITE
131-
# helper to get stable display names
132-
_show_name(C) = sprint(show, C)
133-
# helper to ensure nested groups exist and return them
134-
function ensure_group!(grp::BenchmarkGroup, key)
135-
if haskey(grp, key)
136-
return grp[key]
137-
else
138-
sub = BenchmarkGroup()
139-
grp[key] = sub
140-
return sub
141-
end
8+
function exercise(rng, C)
9+
d = length(C)
10+
u = rand(rng, C, 5)
11+
c = cdf(C, u)
12+
p = pdf(C, u)
13+
r = rosenblatt(C, u)
14+
ir = inverse_rosenblatt(C, r)
15+
return nothing
16+
end
17+
function metrics(C)
18+
Copulas.τ(C)
19+
Copulas.ρ(C)
20+
Copulas.β(C)
21+
Copulas.γ(C)
22+
Copulas.ι(C)
23+
return nothing
24+
end
25+
function fitting(rng, C)
26+
d = length(C)
27+
CT = typeof(C)
28+
u = rand(rng, C, 10)
29+
for m in Copulas._available_fitting_methods(CT, d)
30+
fit(CT, u, m)
31+
fit(CopulaModel, CT, u, m)
14232
end
33+
return nothing
34+
end
35+
function conditioning(rng, C)
36+
d = length(C)
37+
u = rand(rng, C)
38+
D = condition(C, 2:d, u[2:d,1])
39+
cdf(D, rand(rng))
40+
quantile(D, rand(rng))
41+
return nothing
42+
end
14343

144-
for C in EXAMPLES # only the first two for the moment to see.
145-
CT, d = typeof(C), length(C)
146-
name_full = _show_name(C)
147-
148-
# Pre-generate inputs for cdf/pdf/rosenblatt and inverse
149-
u = rand(rng, C, 5)
150-
v = rand(rng, d, 5) # uniform
151-
152-
# Core ops grouped first by op, then by copula name
153-
ensure_group!(top, "rng")[name_full] = @benchmarkable rand($rng, $C, 64)
154-
ensure_group!(top, "cdf")[name_full] = @benchmarkable cdf($C, $u)
155-
# ensure_group!(top, "pdf")[name_full] = @benchmarkable pdf($C, $u)
156-
# ensure_group!(top, "rbt")[name_full] = @benchmarkable rosenblatt($C, $u)
157-
# ensure_group!(top, "irbt")[name_full] = @benchmarkable inverse_rosenblatt($C, $v)
158-
159-
# # Metrics: scalar and matrix-style
160-
# mgrp = ensure_group!(top, "metrics")
161-
# ensure_group!(mgrp, "τ")[name_full] = @benchmarkable Copulas.τ($C)
162-
# ensure_group!(mgrp, "ρ")[name_full] = @benchmarkable Copulas.ρ($C)
163-
# ensure_group!(mgrp, "β")[name_full] = @benchmarkable Copulas.β($C)
164-
# ensure_group!(mgrp, "γ")[name_full] = @benchmarkable Copulas.γ($C)
165-
# ensure_group!(mgrp, "ι")[name_full] = @benchmarkable Copulas.ι($C)
166-
# ensure_group!(mgrp, "corτ")[name_full] = @benchmarkable StatsBase.corkendall($C)
167-
# ensure_group!(mgrp, "corρ")[name_full] = @benchmarkable StatsBase.corspearman($C)
168-
# ensure_group!(mgrp, "corβ")[name_full] = @benchmarkable Copulas.corblomqvist($C)
169-
# ensure_group!(mgrp, "corγ")[name_full] = @benchmarkable Copulas.corgini($C)
170-
# ensure_group!(mgrp, "corι")[name_full] = @benchmarkable Copulas.corentropy($C)
171-
172-
# # Fitting: iterate available methods for this copula type
173-
# fitgrp = ensure_group!(top, "fit")
174-
# for m in Copulas._available_fitting_methods(CT, d)
175-
# ensure_group!(fitgrp, string(m))[name_full] = @benchmarkable fit($CT, $u, $m)
176-
# end
177-
178-
# # Conditioning: distortions (any d>=2) and bivariate conditional copulas (if d>3)
179-
# condgrp = ensure_group!(top, "cond")
180-
# D1 = condition(C, 2:d, u[2:d,1]) # condition the copula on the first sampled point.
181-
# t = rand(rng, 32)
182-
# ensure_group!(condgrp, "dst/cdf")[name_full] = @benchmarkable Distributions.cdf($D1, $t)
183-
# ensure_group!(condgrp, "dst/quantile")[name_full] = @benchmarkable Distributions.quantile($D1, $t)
44+
function run_benches()
45+
rng = StableRNG(123)
46+
47+
EXAMPLES = unique([
48+
AMHCopula(3, 0.2),
49+
ArchimaxCopula(2, Copulas.BB1Generator(1.3, 1.4), Copulas.GalambosTail(0.7)),
50+
ArchimaxCopula(2, Copulas.ClaytonGenerator(1.5), Copulas.HuslerReissTail(0.6)),
51+
ArchimaxCopula(2, Copulas.FrankGenerator(6.0), Copulas.LogTail(1.5)),
52+
ArchimaxCopula(2, Copulas.GumbelGenerator(2.0), Copulas.AsymGalambosTail(0.35, 0.65, 0.3)),
53+
ArchimaxCopula(2, Copulas.JoeGenerator(2.5), Copulas.GalambosTail(2.5)),
54+
ArchimedeanCopula(3, EmpiricalGenerator(randn(rng, 3, 200))),
55+
AsymGalambosCopula(2, 5.0, 0.8, 0.3),
56+
AsymLogCopula(2, 1.5, 0.5, 0.2),
57+
AsymMixedCopula(2, 0.12, 0.13),
58+
BB10Copula(2, 3.0, 0.8),
59+
BB1Copula(2, 1.2, 1.5),
60+
BB2Copula(2, 1.5, 1.8),
61+
BB3Copula(2, 2.5, 0.5),
62+
BB6Copula(2, 1.5, 1.4),
63+
BB7Copula(2, 1.5, 0.4),
64+
BB8Copula(2, 1.5, 0.6),
65+
BB9Copula(2, 2.0, 1.5),
66+
BC2Copula(2, 0.5, 0.5),
67+
BC2Copula(2, 0.7, 0.3),
68+
BernsteinCopula(ClaytonCopula(3, 3.3); m=5),
69+
BernsteinCopula(randn(rng, 3,100); m=5, pseudo_values=false),
70+
BetaCopula(randn(rng, 3,100)),
71+
CheckerboardCopula(randn(rng, 3,100); pseudo_values=false),
72+
ClaytonCopula(2, -0.7),
73+
ClaytonCopula(4, 3.0),
74+
Copulas.SubsetCopula(RafteryCopula(3, 0.5), (2,1)),
75+
CuadrasAugeCopula(2, 0.0),
76+
CuadrasAugeCopula(2, 0.8),
77+
EmpiricalCopula(randn(2,20), pseudo_values=false),
78+
EmpiricalEVCopula(randn(rng, 2,20); method=:pickands, pseudo_values=false),
79+
FGMCopula(2, 1),
80+
FGMCopula(3, [0.1,0.2,0.3,0.4]),
81+
FrankCopula(2, -5),
82+
FrankCopula(4, 5),
83+
GalambosCopula(2, 4.3),
84+
GalambosCopula(2, 120),
85+
GaussianCopula([1 0.7; 0.7 1]),
86+
GumbelBarnettCopula(2, 1.0),
87+
GumbelBarnettCopula(3, 0.35),
88+
GumbelCopula(2, 1.2),
89+
GumbelCopula(4, 7.0),
90+
HuslerReissCopula(2, 3.5),
91+
IndependentCopula(4),
92+
InvGaussianCopula(2, 1.0),
93+
InvGaussianCopula(4, 0.05),
94+
JoeCopula(2, Inf),
95+
JoeCopula(3, 7),
96+
LogCopula(2, 1.5),
97+
MCopula(4),
98+
MixedCopula(2, 0.5),
99+
MOCopula(2, 0.5, 0.5, 0.5),
100+
PlackettCopula(2.0),
101+
RafteryCopula(3, 0.5),
102+
TCopula(4, [1 0.5; 0.5 1]),
103+
tEVCopula(2, 4.0, 0.5),
104+
WCopula(2),
105+
]);
106+
107+
G = BenchmarkGroup()
184108

185-
# if d > 3
186-
# CC = condition(C, 3:d, u[3:d,1])
187-
# Ucc = rand(rng, CC, 64)
188-
# Vcc = rand(rng, 2, 64)
189-
# ensure_group!(condgrp, "cc/rng")[name_full] = @benchmarkable rand($rng, $CC, 64)
190-
# ensure_group!(condgrp, "cc/cdf")[name_full] = @benchmarkable cdf($CC, $Ucc)
191-
# ensure_group!(condgrp, "cc/pdf")[name_full] = @benchmarkable pdf($CC, $Ucc)
192-
# ensure_group!(condgrp, "cc/rbt")[name_full] = @benchmarkable rosenblatt($CC, $Ucc)
193-
# ensure_group!(condgrp, "cc/irbt")[name_full] = @benchmarkable inverse_rosenblatt($CC, $Vcc)
194-
# end
109+
for C in EXAMPLES # only the first two for the moment to see.
110+
tp = typeof(C).name.wrapper
111+
nm = sprint(show, C)
112+
@show nm
113+
G["exercise"][tp][nm] = @be exercise(rng, C)
114+
# G["metrics"][nm] = @be metrics(C)
115+
# G["fitting"][nm] = @be fitting(rng, C)
116+
# G["conditioning"][tp][nm] = @be conditioning(rng, C)
195117
end
118+
return(G)
196119
end
197120

198-
199-
# Separate SklarDist benchmarks on a single representative model (Clayton d=5)
200-
# let rng = StableRNG(321)
201-
# top = SUITE["sklar"] = BenchmarkGroup()
202-
# C = ClaytonCopula(5, 2.0)
203-
# m = (Distributions.Normal(), Distributions.LogNormal(), Distributions.Gamma(2,2), Distributions.Beta(2,5), Distributions.Uniform())
204-
# S = SklarDist(C, m)
205-
# X = rand(rng, S, 128)
206-
# U = pseudos(X)
207-
# top["rand/128"] = @benchmarkable rand($rng, $S, 128)
208-
# top["cdf/128"] = @benchmarkable cdf($S, $X)
209-
# top["pdf/128"] = @benchmarkable pdf($S, $X)
210-
# top["rosenblatt/128"] = @benchmarkable rosenblatt($S, $X)
211-
# V = rand(rng, 5, 128)
212-
# top["inverse_rosenblatt/128"] = @benchmarkable inverse_rosenblatt($S, $V)
213-
# # Fitting pathways (keep light)
214-
# top["fit/ifm"] = @benchmarkable Distributions.fit(CopulaModel, SklarDist{ClaytonCopula, typeof(m)}, $X; sklar_method=:ifm, copula_method=:itau, summaries=false)
215-
# top["fit/ecdf"] = @benchmarkable Distributions.fit(CopulaModel, SklarDist{ClaytonCopula, typeof(m)}, $X; sklar_method=:ecdf, copula_method=:itau, summaries=false)
216-
# end
121+
# PkgBenchmark entrypoint; must define SUITE
122+
const SUITE = run_benches();

0 commit comments

Comments
 (0)