Skip to content

Commit fc47a39

Browse files
andrewjradcliffeandrewjradcliffe
committed
Move extant tests to separate file
1 parent fd5d70c commit fc47a39

File tree

2 files changed

+126
-223
lines changed

2 files changed

+126
-223
lines changed

‎test/runtests.jl‎

Lines changed: 10 additions & 223 deletions
Original file line numberDiff line numberDiff line change
@@ -2,229 +2,16 @@ using VectorizedReduction
22
using Test
33

44
@testset "VectorizedReduction.jl" begin
5-
@testset "vvmapreduce" begin
6-
A = rand(5,5,5,5)
7-
@test vvmapreduce(abs2, +, A, dims=(1,3)) ≈ mapreduce(abs2, +, A, dims=(1,3))
8-
@test vvmapreduce(cos, *, A, dims=(2,4)) ≈ mapreduce(cos, *, A, dims=(2,4))
9-
@test vvprod(log, A, dims=1) ≈ prod(log, A, dims=1)
10-
@test vvminimum(sin, A, dims=(3,4)) ≈ minimum(sin, A, dims=(3,4))
11-
@test vvmaximum(sin, A, dims=(3,4)) ≈ maximum(sin, A, dims=(3,4))
12-
end
13-
@testset "vvmapreduce_vararg" begin
14-
A1 = rand(5,5,5,5)
15-
A2 = rand(5,5,5,5)
16-
A3 = rand(5,5,5,5)
17-
A4 = rand(1:10, 5,5,5,5)
18-
as = (A1, A2, A3)
19-
@test vvmapreduce(+, +, as, dims=(1,2,4)) ≈ mapreduce(+, +, A1, A2, A3, dims=(1,2,4))
20-
# Tests of variably typed arrays
21-
@test vvmapreduce(+, +, A1, A2, dims=(2,3,4)) ≈ mapreduce(+, +, A1, A2, dims=(2,3,4))
22-
@test vvmapreduce(+, +, A1, A4, dims=(2,3,4)) ≈ mapreduce(+, +, A1, A4, dims=(2,3,4))
23-
# interface tests
24-
r = mapreduce(*, +, A1, A2, A3)
25-
@test r ≈ vvmapreduce(*, +, zero, A1, A2, A3)
26-
@test r ≈ vvmapreduce(*, +, A1, A2, A3)
27-
@test r ≈ vvmapreduce(*, +, A1, A2, A3, dims=:)
28-
@test r ≈ vvmapreduce(*, +, A1, A2, A3, dims=:, init=0)
29-
@test r ≈ vvmapreduce(*, +, A1, A2, A3, dims=:, init=zero)
30-
@test r ≈ vvmapreduce(*, +, as)
31-
# # And for really strange stuff (e.g. posterior predictive transformations)
32-
# @benchmark vvmapreduce((x,y,z) -> ifelse(x*y+z ≥ 1, 1, 0), +, A1, A2, A3)
33-
# @benchmark vvmapreduce((x,y,z) -> ifelse(x*y+z ≥ 1, 1, 0), +, A1, A2, A3, dims=(2,3,4))
34-
# # using ifelse for just a boolean is quite slow, but the above is just for demonstration
35-
# @benchmark vvmapreduce(≥, +, A1, A2)
36-
# @benchmark vvmapreduce((x,y,z) -> ≥(x*y+z, 1), +, A1, A2, A3)
37-
# @benchmark vvmapreduce((x,y,z) -> ≥(x*y+z, 1), +, A1, A2, A3, dims=(2,3,4))
38-
# @benchmark mapreduce((x,y,z) -> ≥(x*y+z, 1), +, A1, A2, A3)
39-
# # What I mean by posterior predictive transformation? Well, one might encounter
40-
# # this in Bayesian model checking, which provides a convenient example.
41-
# # If one wishes to compute the Pr = ∫∫𝕀(T(yʳᵉᵖ, θ) ≥ T(y, θ))p(yʳᵉᵖ|θ)p(θ|y)dyʳᵉᵖdθ
42-
# # Let's imagine that A1 represents T(yʳᵉᵖ, θ) and A2 represents T(y, θ)
43-
# # i.e. the test variable samples computed as a functional of the Markov chain (samples of θ)
44-
# Then, Pr is computed as
45-
vvmapreduce(≥, +, A1, A2) / length(A1)
46-
# Or, if only the probability is of interest, and we do not wish to use the functionals
47-
# for any other purpose, we could compute it as:
48-
# vvmapreduce((x, y) -> ≥(f(x), f(y)), +, A1, A2)
49-
# where `f` is the functional of interest, e.g.
50-
f(x, y) = ≥(abs2(x), abs2(y))
51-
# r = mapreduce((x, y) -> ≥(abs2(x), abs2(y)), +, A1, A2)
52-
r = mapreduce(f, +, A1, A2)
53-
# @test r ≈ vvmapreduce((x, y) -> ≥(abs2(x), abs2(y)), +, A1, A2)
54-
@test r ≈ vvmapreduce(f, +, A1, A2)
55-
# R = mapreduce((x, y) -> ≥(abs2(x), abs2(y)), +, A1, A2, dims=(2,3,4))
56-
# @test R ≈ vvmapreduce((x, y) -> ≥(abs2(x), abs2(y)), +, A1, A2, dims=(2,3,4))
57-
R = mapreduce(f, +, A1, A2, dims=(2,3,4))
58-
@test R ≈ vvmapreduce(f, +, A1, A2, dims=(2,3,4))
59-
# One can also express commonly encountered reductions with ease;
60-
# these will be fused once a post-reduction operator can be specified
61-
# MSE
62-
sqdiff(x, y) = abs2(x -y)
63-
# B = vvmapreduce((x, y) -> abs2(x - y), +, A1, A2, dims=(2,4)) ./ (size(A1, 2) * size(A1, 4) )
64-
# @test B ≈ mapreduce((x, y) -> abs2(x - y), +, A1, A2, dims=(2,4)) ./ (size(A1, 2) * size(A1, 4) )
65-
B = vvmapreduce(sqdiff, +, A1, A2, dims=(2,4)) ./ (size(A1, 2) * size(A1, 4))
66-
@test B ≈ mapreduce(sqdiff, +, A1, A2, dims=(2,4)) ./ (size(A1, 2) * size(A1, 4))
67-
# Euclidean distance
68-
# B = (√).(vvmapreduce((x, y) -> abs2(x - y), +, A1, A2, dims=(2,4)))
69-
# @test B ≈ (√).(mapreduce((x, y) -> abs2(x - y), +, A1, A2, dims=(2,4)))
70-
B = (√).(vvmapreduce(sqdiff, +, A1, A2, dims=(2,4)))
71-
@test B ≈ (√).(mapreduce(sqdiff, +, A1, A2, dims=(2,4)))
72-
end
73-
@testset "vfindminmax" begin
74-
# Simple
75-
A1 = rand(5,5)
76-
A2 = rand(5,5)
77-
A3 = rand(5,5)
78-
A′ = @. A1 + A2 + A3
79-
@test findmin(A′) == vfindmin(+, A1, A2, A3)
80-
@test findmin(A′, dims=2) == vfindmin(+, A1, A2, A3, dims=2)
81-
#
82-
v1 = rand(5)
83-
v2 = rand(5)
84-
v3 = rand(5)
85-
v′ = @. v1 + v2 + v3;
86-
@test findmin(v′) == vfindmin(+, v1, v2, v3)
87-
#
88-
A = rand(5,5,5,5)
89-
A′ = cos.(A)
90-
val1, ind1 = findmax(A′, dims=(2,4))
91-
val2, ind2 = vfindmax(cos, A, dims=(2,4))
92-
@test ind1 == ind2 && val1 ≈ val2
93-
#
94-
g(x) = ifelse(abs(x) ≥ 2, 100, 1)
95-
A = randn(5,5,5,5)
96-
A′ = g.(A)
97-
val1, ind1 = findmax(A′, dims=(2,4))
98-
val2, ind2 = vfindmax(g, A, dims=(2,4))
99-
@test ind1 == ind2 && val1 ≈ val2
100-
end
101-
@testset "vfindminmax_vararg" begin
102-
A1 = rand(5,5)
103-
A2 = rand(5,5)
104-
A3 = rand(5,5)
105-
A′ = @. A1 * A2 + A3;
106-
@test findmin(A′) == vfindmin((x, y, z) -> x * y + z, A1, A2, A3)
107-
val1, ind1 = findmin(A′, dims=2)
108-
val2, ind2 = vfindmin((x, y, z) -> x * y + z, A1, A2, A3, dims=2)
109-
@test ind1 == ind2 && val1 ≈ val2
110-
#
111-
B1, B2, B3 = randn(5,5,5,5), randn(5,5,5,5), randn(5,5,5,5);
112-
h(x, y, z) = ifelse(x ≥ .5, 100, 1) * y + √abs(z)
113-
B′ = h.(B1, B2, B3)
114-
val1, ind1 = findmax(B′, dims=(2,4))
115-
val2, ind2 = vfindmax(h, B1, B2, B3, dims=(2,4))
116-
@test ind1 == ind2 && val1 ≈ val2
117-
end
118-
############################################################################################
119-
@testset "from reducedim.jl tests" begin
120-
A = [1.0 5.0 6.0;
121-
5.0 2.0 4.0]
122-
for (tup, rval, rind) in [((1,), [1.0 2.0 4.0], [CartesianIndex(1,1) CartesianIndex(2,2) CartesianIndex(2,3)]),
123-
((2,), reshape([1.0,2.0], 2, 1), reshape([CartesianIndex(1,1),CartesianIndex(2,2)], 2, 1)),
124-
((1,2), fill(1.0,1,1),fill(CartesianIndex(1,1),1,1))]
125-
@test vfindmin1(A, dims=tup) == (rval, rind)
126-
@test isequal(vvminimum(A, dims=tup), rval)
127-
end
128-
129-
for (tup, rval, rind) in [((1,), [5.0 5.0 6.0], [CartesianIndex(2,1) CartesianIndex(1,2) CartesianIndex(1,3)]),
130-
((2,), reshape([6.0,5.0], 2, 1), reshape([CartesianIndex(1,3),CartesianIndex(2,1)], 2, 1)),
131-
((1,2), fill(6.0,1,1),fill(CartesianIndex(1,3),1,1))]
132-
@test vfindmax1(A, dims=tup) == (rval, rind)
133-
@test isequal(vvmaximum(A, dims=tup), rval)
134-
end
135-
end
136-
@testset "findmin/findmax transformed arguments, numeric values" begin
137-
A = [1.0 -5.0 -6.0;
138-
-5.0 2.0 4.0]
139-
TA = [((1,), [1.0 2.0 4.0], [CartesianIndex(1,1) CartesianIndex(2,2) CartesianIndex(2,3)]),
140-
((2,), reshape([1.0, 2.0], 2, 1), reshape([CartesianIndex(1,1), CartesianIndex(2,2)], 2, 1)),
141-
((1,2), fill(1.0,1,1), fill(CartesianIndex(1,1),1,1))]
142-
TA2 = [((1,), [1.0 4.0 16.0], [CartesianIndex(1,1) CartesianIndex(2,2) CartesianIndex(2,3)]),
143-
((2,), reshape([1.0, 4.0], 2, 1), reshape([CartesianIndex(1,1), CartesianIndex(2,2)], 2, 1)),
144-
((1,2), fill(1.0,1,1), fill(CartesianIndex(1,1),1,1))]
145-
TAc = [((1,), [0.28366218546322625 -0.4161468365471424 -0.6536436208636119], [CartesianIndex(2,1) CartesianIndex(2,2) CartesianIndex(2,3)]),
146-
((2,), reshape([0.28366218546322625, -0.6536436208636119], 2, 1), reshape([CartesianIndex(1,2), CartesianIndex(2,3)], 2, 1)),
147-
((1,2), fill(-0.6536436208636119,1,1), fill(CartesianIndex(2,3),1,1))]
148-
for (f, At) in ((abs, TA), (abs2, TA2), (cos, TAc))
149-
A′ = map(f, A)
150-
for (tup, rval, rind) in At
151-
(rval′, rind′) = vfindmin1(f, A, dims=tup)
152-
@test all(rval′ .≈ rval)
153-
@test rind′ == rind
154-
(rval′′, rind′′) = vfindmin1(A′, dims=tup)
155-
@test all(rval′ .≈ rval′′)
156-
@test rind′ == rind′′
157-
end
158-
end
159-
160-
TA = [((1,), [5.0 5.0 6.0], [CartesianIndex(2,1) CartesianIndex(1,2) CartesianIndex(1,3)]),
161-
((2,), reshape([6.0,5.0], 2, 1), reshape([CartesianIndex(1,3), CartesianIndex(2,1)], 2, 1)),
162-
((1,2), fill(6.0,1,1),fill(CartesianIndex(1,3),1,1))]
163-
TA2 = [((1,), [25.0 25.0 36.0], [CartesianIndex(2,1) CartesianIndex(1,2) CartesianIndex(1,3)]),
164-
((2,), reshape([36.0, 25.0], 2, 1), reshape([CartesianIndex(1,3), CartesianIndex(2,1)], 2, 1)),
165-
((1,2), fill(36.0,1,1), fill(CartesianIndex(1,3),1,1))]
166-
TAc = [((1,), [0.5403023058681398 0.28366218546322625 0.960170286650366], [CartesianIndex(1,1) CartesianIndex(1,2) CartesianIndex(1,3)]),
167-
((2,), reshape([0.960170286650366, 0.28366218546322625], 2, 1), reshape([CartesianIndex(1,3), CartesianIndex(2,1)], 2, 1)),
168-
((1,2), fill(0.960170286650366,1,1), fill(CartesianIndex(1,3),1,1))]
169-
for (f, At) in ((abs, TA), (abs2, TA2), (cos, TAc))
170-
A′ = map(f, A)
171-
for (tup, rval, rind) in At
172-
(rval′, rind′) = vfindmax1(f, A, dims=tup)
173-
@test all(rval′ .≈ rval)
174-
@test rind′ == rind
175-
(rval′′, rind′′) = vfindmax1(A′, dims=tup)
176-
@test all(rval′ .≈ rval′′)
177-
@test rind′ == rind′′
178-
end
179-
end
180-
end
181-
182-
@testset "NaN in findmin/findmax/minimum/maximum" begin
183-
A = [1.0 NaN 6.0;
184-
NaN 2.0 4.0]
185-
A′ = [-1.0 NaN -6.0;
186-
NaN -2.0 4.0]
187-
for (tup, rval, rind) in [((1,), [1.0 2.0 4.0], [CartesianIndex(1,1) CartesianIndex(2,2) CartesianIndex(2,3)]),
188-
((2,), reshape([1.0, 2.0], 2, 1), reshape([CartesianIndex(1,1),CartesianIndex(2,2)], 2, 1)),
189-
((1,2), fill(1.0,1,1),fill(CartesianIndex(1,1),1,1))]
190-
@test isequal(vfindmin1(A, dims=tup), (rval, rind))
191-
@test isequal(vfindmin1(abs, A′, dims=tup), (rval, rind))
192-
@test isequal(vvminimum(A, dims=tup), rval)
193-
@test isequal(vvminimum(abs, A′, dims=tup), rval)
194-
end
195-
196-
for (tup, rval, rind) in [((1,), [1.0 2.0 6.0], [CartesianIndex(1,1) CartesianIndex(2,2) CartesianIndex(1,3)]),
197-
((2,), reshape([6.0, 4.0], 2, 1), reshape([CartesianIndex(1,3),CartesianIndex(2,3)], 2, 1)),
198-
((1,2), fill(6.0,1,1),fill(CartesianIndex(1,3),1,1))]
199-
@test isequal(vfindmax1(A, dims=tup), (rval, rind))
200-
@test isequal(vfindmax1(abs, A′, dims=tup), (rval, rind))
201-
@test isequal(vvmaximum(A, dims=tup), rval)
202-
@test isequal(vvmaximum(abs, A′, dims=tup), rval)
203-
end
204-
end
205-
206-
@testset "+/-Inf in findmin/findmax/minimum/maximum" begin
207-
A = [Inf -Inf Inf -Inf;
208-
Inf Inf -Inf -Inf]
209-
A′ = [1 0 1 0;
210-
1 1 0 0]
211-
retinf(x::T) where {T} = ifelse(x == one(T), Inf, -Inf)
212-
for (tup, rval, rind) in [((1,), [Inf -Inf -Inf -Inf], [CartesianIndex(1,1) CartesianIndex(1,2) CartesianIndex(2,3) CartesianIndex(1,4)]),
213-
((2,), reshape([-Inf -Inf], 2, 1), reshape([CartesianIndex(1,2),CartesianIndex(2,3)], 2, 1)),
214-
((1,2), fill(-Inf,1,1),fill(CartesianIndex(1,2),1,1))]
215-
@test isequal(vfindmin1(A, dims=tup), (rval, rind))
216-
@test isequal(vfindmin1(retinf, A′, dims=tup), (rval, rind))
217-
@test isequal(vvminimum(A, dims=tup), rval)
218-
@test isequal(vvminimum(retinf, A′, dims=tup), rval)
219-
end
220-
221-
for (tup, rval, rind) in [((1,), [Inf Inf Inf -Inf], [CartesianIndex(1,1) CartesianIndex(2,2) CartesianIndex(1,3) CartesianIndex(1,4)]),
222-
((2,), reshape([Inf Inf], 2, 1), reshape([CartesianIndex(1,1),CartesianIndex(2,1)], 2, 1)),
223-
((1,2), fill(Inf,1,1),fill(CartesianIndex(1,1),1,1))]
224-
@test isequal(vfindmax1(A, dims=tup), (rval, rind))
225-
@test isequal(vfindmax1(retinf, A′, dims=tup), (rval, rind))
226-
@test isequal(vvmaximum(A, dims=tup), rval)
227-
@test isequal(vvmaximum(retinf, A′, dims=tup), rval)
5+
const tests = [
6+
"vrspecials.jl"
7+
"reduce.jl",
8+
"reducedim.jl",
9+
"treduce.jl",
10+
"treducedim.jl"
11+
]
12+
for t in tests
13+
@testset "Test $t" begin
14+
include(t)
22815
end
22916
end
23017
end

‎test/vrspecials.jl‎

Lines changed: 116 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,116 @@
1+
# Tests of some unique syntax, and some shared syntax
2+
3+
@testset "vvmapreduce" begin
4+
A = rand(5,5,5,5)
5+
@test vvmapreduce(abs2, +, A, dims=(1,3)) ≈ mapreduce(abs2, +, A, dims=(1,3))
6+
@test vvmapreduce(cos, *, A, dims=(2,4)) ≈ mapreduce(cos, *, A, dims=(2,4))
7+
@test vvprod(log, A, dims=1) ≈ prod(log, A, dims=1)
8+
@test vvminimum(sin, A, dims=(3,4)) ≈ minimum(sin, A, dims=(3,4))
9+
@test vvmaximum(sin, A, dims=(3,4)) ≈ maximum(sin, A, dims=(3,4))
10+
end
11+
@testset "vvmapreduce_vararg" begin
12+
A1 = rand(5,5,5,5)
13+
A2 = rand(5,5,5,5)
14+
A3 = rand(5,5,5,5)
15+
A4 = rand(1:10, 5,5,5,5)
16+
as = (A1, A2, A3)
17+
@test vvmapreduce(+, +, as, dims=(1,2,4)) ≈ mapreduce(+, +, A1, A2, A3, dims=(1,2,4))
18+
# Tests of variably typed arrays
19+
@test vvmapreduce(+, +, A1, A2, dims=(2,3,4)) ≈ mapreduce(+, +, A1, A2, dims=(2,3,4))
20+
@test vvmapreduce(+, +, A1, A4, dims=(2,3,4)) ≈ mapreduce(+, +, A1, A4, dims=(2,3,4))
21+
# interface tests
22+
r = mapreduce(*, +, A1, A2, A3)
23+
@test r ≈ vvmapreduce(*, +, zero, A1, A2, A3)
24+
@test r ≈ vvmapreduce(*, +, A1, A2, A3)
25+
@test r ≈ vvmapreduce(*, +, A1, A2, A3, dims=:)
26+
@test r ≈ vvmapreduce(*, +, A1, A2, A3, dims=:, init=0)
27+
@test r ≈ vvmapreduce(*, +, A1, A2, A3, dims=:, init=zero)
28+
@test r ≈ vvmapreduce(*, +, as)
29+
# # And for really strange stuff (e.g. posterior predictive transformations)
30+
# @benchmark vvmapreduce((x,y,z) -> ifelse(x*y+z ≥ 1, 1, 0), +, A1, A2, A3)
31+
# @benchmark vvmapreduce((x,y,z) -> ifelse(x*y+z ≥ 1, 1, 0), +, A1, A2, A3, dims=(2,3,4))
32+
# # using ifelse for just a boolean is quite slow, but the above is just for demonstration
33+
# @benchmark vvmapreduce(≥, +, A1, A2)
34+
# @benchmark vvmapreduce((x,y,z) -> ≥(x*y+z, 1), +, A1, A2, A3)
35+
# @benchmark vvmapreduce((x,y,z) -> ≥(x*y+z, 1), +, A1, A2, A3, dims=(2,3,4))
36+
# @benchmark mapreduce((x,y,z) -> ≥(x*y+z, 1), +, A1, A2, A3)
37+
# # What I mean by posterior predictive transformation? Well, one might encounter
38+
# # this in Bayesian model checking, which provides a convenient example.
39+
# # If one wishes to compute the Pr = ∫∫𝕀(T(yʳᵉᵖ, θ) ≥ T(y, θ))p(yʳᵉᵖ|θ)p(θ|y)dyʳᵉᵖdθ
40+
# # Let's imagine that A1 represents T(yʳᵉᵖ, θ) and A2 represents T(y, θ)
41+
# # i.e. the test variable samples computed as a functional of the Markov chain (samples of θ)
42+
# Then, Pr is computed as
43+
# vvmapreduce(≥, +, A1, A2) / length(A1)
44+
@test mapreduce(≥, +, A1, A2) ≈ vvmapreduce(≥, +, A1, A2)
45+
# Or, if only the probability is of interest, and we do not wish to use the functionals
46+
# for any other purpose, we could compute it as:
47+
# vvmapreduce((x, y) -> ≥(f(x), f(y)), +, A1, A2)
48+
# where `f` is the functional of interest, e.g.
49+
f(x, y) = ≥(abs2(x), abs2(y))
50+
# r = mapreduce((x, y) -> ≥(abs2(x), abs2(y)), +, A1, A2)
51+
r = mapreduce(f, +, A1, A2)
52+
# @test r ≈ vvmapreduce((x, y) -> ≥(abs2(x), abs2(y)), +, A1, A2)
53+
@test r ≈ vvmapreduce(f, +, A1, A2)
54+
# R = mapreduce((x, y) -> ≥(abs2(x), abs2(y)), +, A1, A2, dims=(2,3,4))
55+
# @test R ≈ vvmapreduce((x, y) -> ≥(abs2(x), abs2(y)), +, A1, A2, dims=(2,3,4))
56+
R = mapreduce(f, +, A1, A2, dims=(2,3,4))
57+
@test R ≈ vvmapreduce(f, +, A1, A2, dims=(2,3,4))
58+
# One can also express commonly encountered reductions with ease;
59+
# these will be fused once a post-reduction operator can be specified
60+
# MSE
61+
sqdiff(x, y) = abs2(x -y)
62+
# B = vvmapreduce((x, y) -> abs2(x - y), +, A1, A2, dims=(2,4)) ./ (size(A1, 2) * size(A1, 4) )
63+
# @test B ≈ mapreduce((x, y) -> abs2(x - y), +, A1, A2, dims=(2,4)) ./ (size(A1, 2) * size(A1, 4) )
64+
B = vvmapreduce(sqdiff, +, A1, A2, dims=(2,4)) ./ (size(A1, 2) * size(A1, 4))
65+
@test B ≈ mapreduce(sqdiff, +, A1, A2, dims=(2,4)) ./ (size(A1, 2) * size(A1, 4))
66+
# Euclidean distance
67+
# B = (√).(vvmapreduce((x, y) -> abs2(x - y), +, A1, A2, dims=(2,4)))
68+
# @test B ≈ (√).(mapreduce((x, y) -> abs2(x - y), +, A1, A2, dims=(2,4)))
69+
B = (√).(vvmapreduce(sqdiff, +, A1, A2, dims=(2,4)))
70+
@test B ≈ (√).(mapreduce(sqdiff, +, A1, A2, dims=(2,4)))
71+
end
72+
@testset "vfindminmax" begin
73+
# Simple
74+
A1 = rand(5,5)
75+
A2 = rand(5,5)
76+
A3 = rand(5,5)
77+
A′ = @. A1 + A2 + A3
78+
@test findmin(A′) == vfindmin(+, A1, A2, A3)
79+
@test findmin(A′, dims=2) == vfindmin(+, A1, A2, A3, dims=2)
80+
#
81+
v1 = rand(5)
82+
v2 = rand(5)
83+
v3 = rand(5)
84+
v′ = @. v1 + v2 + v3;
85+
@test findmin(v′) == vfindmin(+, v1, v2, v3)
86+
#
87+
A = rand(5,5,5,5)
88+
A′ = cos.(A)
89+
val1, ind1 = findmax(A′, dims=(2,4))
90+
val2, ind2 = vfindmax(cos, A, dims=(2,4))
91+
@test ind1 == ind2 && val1 ≈ val2
92+
#
93+
g(x) = ifelse(abs(x) ≥ 2, 100, 1)
94+
A = randn(5,5,5,5)
95+
A′ = g.(A)
96+
val1, ind1 = findmax(A′, dims=(2,4))
97+
val2, ind2 = vfindmax(g, A, dims=(2,4))
98+
@test ind1 == ind2 && val1 ≈ val2
99+
end
100+
@testset "vfindminmax_vararg" begin
101+
A1 = rand(5,5)
102+
A2 = rand(5,5)
103+
A3 = rand(5,5)
104+
A′ = @. A1 * A2 + A3;
105+
@test findmin(A′) == vfindmin((x, y, z) -> x * y + z, A1, A2, A3)
106+
val1, ind1 = findmin(A′, dims=2)
107+
val2, ind2 = vfindmin((x, y, z) -> x * y + z, A1, A2, A3, dims=2)
108+
@test ind1 == ind2 && val1 ≈ val2
109+
#
110+
B1, B2, B3 = randn(5,5,5,5), randn(5,5,5,5), randn(5,5,5,5);
111+
h(x, y, z) = ifelse(x ≥ .5, 100, 1) * y + √abs(z)
112+
B′ = h.(B1, B2, B3)
113+
val1, ind1 = findmax(B′, dims=(2,4))
114+
val2, ind2 = vfindmax(h, B1, B2, B3, dims=(2,4))
115+
@test ind1 == ind2 && val1 ≈ val2
116+
end

0 commit comments

Comments
 (0)