Skip to content

Commit bfadffb

Browse files
cscherrercscherrer
committed
Hacking on tests
1 parent 2709d8f commit bfadffb

File tree

2 files changed

+181
-203
lines changed

2 files changed

+181
-203
lines changed

‎Project.toml

Lines changed: 2 additions & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -25,9 +25,10 @@ julia = "1.3"
2525

2626
[extras]
2727
Aqua = "4c88cf16-eb10-579e-8560-4a9242c79595"
28+
JETTest = "a79fb612-4a80-4749-a9bd-c2faab13da61"
2829
LinearAlgebra = "37e2e46d-f89d-539d-b4ee-838fcccc9c8e"
2930
Statistics = "10745b16-79ce-11e8-11f9-7d13ad32a3b2"
3031
Test = "8dfed614-e22c-5e08-85e1-65c5234f0b40"
3132

3233
[targets]
33-
test = ["Test", "Aqua", "LinearAlgebra", "Statistics"]
34+
test = ["Test", "Aqua", "JETTest", "LinearAlgebra", "Statistics"]

‎test/runtests.jl

Lines changed: 179 additions & 202 deletions
Original file line numberDiff line numberDiff line change
@@ -8,208 +8,185 @@ using MeasureBase
88
using Aqua
99
Aqua.test_all(MeasureBase; ambiguities=false, unbound_args=false)
1010

11-
function draw2(μ)
12-
x = rand(μ)
13-
y = rand(μ)
14-
while x == y
15-
y = rand(μ)
16-
end
17-
return (x,y)
18-
end
19-
20-
function test_measure(μ)
21-
logdensity(μ, testvalue(μ)) isa AbstractFloat
22-
end
23-
24-
test_measures = [
25-
# Chain(x -> Normal(μ=x), Normal(μ=0.0))
26-
For(3) do j Normal(σ=j) end
27-
For(2,3) do i,j Normal(i,j) end
28-
Lebesgue(ℝ) ^ 3
29-
Lebesgue(ℝ) ^ (2,3)
30-
3 * Lebesgue(ℝ)
31-
Dirac(π)
32-
Lebesgue(ℝ)
33-
# Normal() ⊙ Cauchy()
34-
]
35-
36-
testbroken_measures = [
37-
Pushforward(as𝕀, Normal())
38-
SpikeMixture(Normal(), 2)
39-
# InverseGamma(2) # Not defined yet
40-
# MvNormal(I(3)) # Entirely broken for now
41-
CountingMeasure(Float64)
42-
Likelihood
43-
Dirac(0.0) + Lebesgue(ℝ)
44-
45-
TrivialMeasure()
46-
]
47-
48-
@testset "testvalue" begin
49-
for μ in test_measures
50-
@test test_measure(μ)
51-
end
52-
53-
for μ in testbroken_measures
54-
@test_broken test_measure(μ)
55-
end
11+
# function draw2(μ)
12+
# x = rand(μ)
13+
# y = rand(μ)
14+
# while x == y
15+
# y = rand(μ)
16+
# end
17+
# return (x,y)
18+
# end
19+
20+
# function test_measure(μ)
21+
# logdensity(μ, testvalue(μ)) isa AbstractFloat
22+
# end
23+
24+
# test_measures = [
25+
# # Chain(x -> Normal(μ=x), Normal(μ=0.0))
26+
# For(3) do j Normal(σ=j) end
27+
# For(2,3) do i,j Normal(i,j) end
28+
# Lebesgue(ℝ) ^ 3
29+
# Lebesgue(ℝ) ^ (2,3)
30+
# 3 * Lebesgue(ℝ)
31+
# Dirac(π)
32+
# Lebesgue(ℝ)
33+
# # Normal() ⊙ Cauchy()
34+
# ]
35+
36+
# testbroken_measures = [
37+
# Pushforward(as𝕀, Normal())
38+
# SpikeMixture(Normal(), 2)
39+
# # InverseGamma(2) # Not defined yet
40+
# # MvNormal(I(3)) # Entirely broken for now
41+
# CountingMeasure(Float64)
42+
# Likelihood
43+
# Dirac(0.0) + Lebesgue(ℝ)
44+
45+
# TrivialMeasure()
46+
# ]
47+
48+
# @testset "testvalue" begin
49+
# for μ in test_measures
50+
# @test test_measure(μ)
51+
# end
52+
53+
# for μ in testbroken_measures
54+
# @test_broken test_measure(μ)
55+
# end
5656

57-
@testset "testvalue(::Chain)" begin
58-
mc = Chain(x -> Normal(μ=x), Normal(μ=0.0))
59-
r = testvalue(mc)
60-
@test logdensity(mc, Iterators.take(r, 10)) isa AbstractFloat
61-
end
62-
end
63-
64-
65-
@testset "Kernel" begin
66-
κ = MeasureBase.kernel(MeasureBase.Dirac, identity)
67-
@test rand(κ(1.1)) == 1.1
68-
end
69-
70-
@testset "SpikeMixture" begin
71-
@test rand(SpikeMixture(Dirac(0), 0.5)) == 0
72-
@test rand(SpikeMixture(Dirac(1), 1.0)) == 1
73-
w = 1/3
74-
m = SpikeMixture(Normal(), w)
75-
bm = basemeasure(m)
76-
@test (bm.s*bm.w)*bm.m == 1.0*basemeasure(Normal())
77-
@test density(m, 1.0)*(bm.s*bm.w) == w*density(Normal(),1.0)
78-
@test density(m, 0)*(bm.s*(1-bm.w)) ≈ (1-w)
79-
end
80-
81-
@testset "Dirac" begin
82-
@test rand(Dirac(0.2)) == 0.2
83-
@test logdensity(Dirac(0.3), 0.3) == 0.0
84-
@test logdensity(Dirac(0.3), 0.4) == -Inf
85-
end
86-
87-
@testset "For" begin
88-
FORDISTS = [
89-
For(1:10) do j Normal(μ=j) end
90-
For(4,3) do μ,σ Normal(μ,σ) end
91-
For(1:4, 1:4) do μ,σ Normal(μ,σ) end
92-
For(eachrow(rand(4,2))) do x Normal(x[1], x[2]) end
93-
For(rand(4), rand(4)) do μ,σ Normal(μ,σ) end
94-
]
95-
96-
for d in FORDISTS
97-
@test logdensity(d, rand(d)) isa Float64
98-
end
99-
end
100-
101-
import MeasureBase.:⋅
102-
function ⋅(μ::Normal, kernel)
103-
m = kernel(μ)
104-
Normal(μ = m.μ.μ, σ = sqrt(m.μ.σ^2 + m.σ^2))
105-
end
106-
struct AffineMap{S,T}
107-
B::S
108-
β::T
109-
end
110-
(a::AffineMap)(x) = a.B*x + a.β
111-
(a::AffineMap)(p::Normal) = Normal(μ = a.B*mean(p) + a.β, σ = sqrt(a.B*p.σ^2*a.B'))
112-
113-
@testset "DynamicFor" begin
114-
mc = Chain(Normal(μ=0.0)) do x Normal(μ=x) end
115-
r = rand(mc)
57+
# @testset "testvalue(::Chain)" begin
58+
# mc = Chain(x -> Normal(μ=x), Normal(μ=0.0))
59+
# r = testvalue(mc)
60+
# @test logdensity(mc, Iterators.take(r, 10)) isa AbstractFloat
61+
# end
62+
# end
63+
64+
65+
# @testset "Kernel" begin
66+
# κ = MeasureBase.kernel(MeasureBase.Dirac, identity)
67+
# @test rand(κ(1.1)) == 1.1
68+
# end
69+
70+
# @testset "SpikeMixture" begin
71+
# @test rand(SpikeMixture(Dirac(0), 0.5)) == 0
72+
# @test rand(SpikeMixture(Dirac(1), 1.0)) == 1
73+
# w = 1/3
74+
# m = SpikeMixture(Normal(), w)
75+
# bm = basemeasure(m)
76+
# @test (bm.s*bm.w)*bm.m == 1.0*basemeasure(Normal())
77+
# @test density(m, 1.0)*(bm.s*bm.w) == w*density(Normal(),1.0)
78+
# @test density(m, 0)*(bm.s*(1-bm.w)) ≈ (1-w)
79+
# end
80+
81+
# @testset "Dirac" begin
82+
# @test rand(Dirac(0.2)) == 0.2
83+
# @test logdensity(Dirac(0.3), 0.3) == 0.0
84+
# @test logdensity(Dirac(0.3), 0.4) == -Inf
85+
# end
86+
87+
# @testset "For" begin
88+
# FORDISTS = [
89+
# For(1:10) do j Normal(μ=j) end
90+
# For(4,3) do μ,σ Normal(μ,σ) end
91+
# For(1:4, 1:4) do μ,σ Normal(μ,σ) end
92+
# For(eachrow(rand(4,2))) do x Normal(x[1], x[2]) end
93+
# For(rand(4), rand(4)) do μ,σ Normal(μ,σ) end
94+
# ]
95+
96+
# for d in FORDISTS
97+
# @test logdensity(d, rand(d)) isa Float64
98+
# end
99+
# end
100+
101+
# import MeasureBase.:⋅
102+
# function ⋅(μ::Normal, kernel)
103+
# m = kernel(μ)
104+
# Normal(μ = m.μ.μ, σ = sqrt(m.μ.σ^2 + m.σ^2))
105+
# end
106+
# struct AffineMap{S,T}
107+
# B::S
108+
# β::T
109+
# end
110+
# (a::AffineMap)(x) = a.B*x + a.β
111+
# (a::AffineMap)(p::Normal) = Normal(μ = a.B*mean(p) + a.β, σ = sqrt(a.B*p.σ^2*a.B'))
112+
113+
# @testset "DynamicFor" begin
114+
# mc = Chain(Normal(μ=0.0)) do x Normal(μ=x) end
115+
# r = rand(mc)
116116

117-
# Check that `r` is now deterministic
118-
@test logdensity(mc, take(r, 100)) == logdensity(mc, take(r, 100))
117+
# # Check that `r` is now deterministic
118+
# @test logdensity(mc, take(r, 100)) == logdensity(mc, take(r, 100))
119119

120-
d2 = For(r) do x Normal(μ=x) end
121-
122-
@test_broken let r2 = rand(d2)
123-
logdensity(d2, take(r2, 100)) == logdensity(d2, take(r2, 100))
124-
end
125-
end
126-
127-
@testset "Univariate chain" begin
128-
ξ0 = 1.
129-
x = 1.2
130-
P0 = 1.0
131-
132-
Φ = 0.8
133-
β = 0.1
134-
Q = 0.2
135-
136-
μ = Normal(μ=ξ0, σ=sqrt(P0))
137-
kernel = MeasureBase.kernel(Normal; μ=AffineMap(Φ, β), σ=Const(Q))
138-
139-
@test (μ ⋅ kernel).μ == Normal(μ = 0.9, σ = 0.824621).μ
140-
141-
chain = Chain(kernel, μ)
142-
143-
144-
dyniterate(iter::TimeLift, ::Nothing) = dyniterate(iter, 0=>nothing)
145-
tr1 = trace(TimeLift(chain), nothing, u -> u[1] > 15)
146-
tr2 = trace(TimeLift(rand(Random.GLOBAL_RNG, chain)), nothing, u -> u[1] > 15)
147-
collect(Iterators.take(chain, 10))
148-
collect(Iterators.take(rand(Random.GLOBAL_RNG, chain), 10))
149-
end
150-
151-
@testset "Likelihood" begin
152-
dps = [
153-
(Normal() , 2.0 )
154-
# (Pushforward(as((μ=asℝ,)), Normal()^1), (μ=2.0,))
155-
]
156-
157-
ℓs = [
158-
Likelihood(Normal{(:μ,)}, 3.0)
159-
Likelihood(kernel(Normal, x -> (μ=x, σ=2.0)), 3.0)
160-
]
161-
162-
for (d,p) in dps
163-
for ℓ in ℓs
164-
@test logdensity(d ⊙ ℓ, p) == logdensity(d, p) + logdensity(ℓ, p)
165-
end
166-
end
167-
end
168-
169-
170-
@testset "ProductMeasure" begin
171-
d = For(1:10) do j Poisson(exp(j)) end
172-
x = Vector{Int16}(undef, 10)
173-
@test rand!(d,x) isa Vector
174-
@test rand(d) isa Vector
175-
176-
@testset "Indexed by Generator" begin
177-
d = For((j^2 for j in 1:10)) do i Poisson(i) end
178-
x = Vector{Int16}(undef, 10)
179-
@test rand!(d,x) isa Vector
180-
@test_broken rand(d) isa Base.Generator
181-
end
182-
183-
@testset "Indexed by multiple Ints" begin
184-
d = For(2,3) do μ,σ Normal(μ,σ) end
185-
x = Matrix{Float16}(undef, 2, 3)
186-
@test rand!(d, x) isa Matrix
187-
@test_broken rand(d) isa Matrix{Float16}
188-
end
189-
end
190-
191-
@testset "Show methods" begin
192-
@testset "PowerMeasure" begin
193-
@test repr(Lebesgue(ℝ) ^ 5) == "Lebesgue(ℝ) ^ 5"
194-
@test repr(Lebesgue(ℝ) ^ (3, 2)) == "Lebesgue(ℝ) ^ (3, 2)"
195-
end
196-
end
197-
198-
@testset "Density measures and Radon-Nikodym" begin
199-
x = randn()
200-
let d = ∫(𝒹(Cauchy(), Normal()), Normal())
201-
@test logdensity(d, x) ≈ logdensity(Cauchy(), x)
202-
end
203-
204-
let f = 𝒹(∫(x -> x^2, Normal()), Normal())
205-
@test f(x) ≈ x^2
206-
end
207-
208-
let d = ∫exp(log𝒹(Cauchy(), Normal()), Normal())
209-
@test logdensity(d, x) ≈ logdensity(Cauchy(), x)
210-
end
211-
212-
let f = log𝒹(∫exp(x -> x^2, Normal()), Normal())
213-
@test f(x) ≈ x^2
214-
end
215-
end
120+
# d2 = For(r) do x Normal(μ=x) end
121+
122+
# @test_broken let r2 = rand(d2)
123+
# logdensity(d2, take(r2, 100)) == logdensity(d2, take(r2, 100))
124+
# end
125+
# end
126+
127+
128+
# @testset "Likelihood" begin
129+
# dps = [
130+
# (Normal() , 2.0 )
131+
# # (Pushforward(as((μ=asℝ,)), Normal()^1), (μ=2.0,))
132+
# ]
133+
134+
# ℓs = [
135+
# Likelihood(Normal{(:μ,)}, 3.0)
136+
# Likelihood(kernel(Normal, x -> (μ=x, σ=2.0)), 3.0)
137+
# ]
138+
139+
# for (d,p) in dps
140+
# for ℓ in ℓs
141+
# @test logdensity(d ⊙ ℓ, p) == logdensity(d, p) + logdensity(ℓ, p)
142+
# end
143+
# end
144+
# end
145+
146+
147+
# @testset "ProductMeasure" begin
148+
# d = For(1:10) do j Poisson(exp(j)) end
149+
# x = Vector{Int16}(undef, 10)
150+
# @test rand!(d,x) isa Vector
151+
# @test rand(d) isa Vector
152+
153+
# @testset "Indexed by Generator" begin
154+
# d = For((j^2 for j in 1:10)) do i Poisson(i) end
155+
# x = Vector{Int16}(undef, 10)
156+
# @test rand!(d,x) isa Vector
157+
# @test_broken rand(d) isa Base.Generator
158+
# end
159+
160+
# @testset "Indexed by multiple Ints" begin
161+
# d = For(2,3) do μ,σ Normal(μ,σ) end
162+
# x = Matrix{Float16}(undef, 2, 3)
163+
# @test rand!(d, x) isa Matrix
164+
# @test_broken rand(d) isa Matrix{Float16}
165+
# end
166+
# end
167+
168+
# @testset "Show methods" begin
169+
# @testset "PowerMeasure" begin
170+
# @test repr(Lebesgue(ℝ) ^ 5) == "Lebesgue(ℝ) ^ 5"
171+
# @test repr(Lebesgue(ℝ) ^ (3, 2)) == "Lebesgue(ℝ) ^ (3, 2)"
172+
# end
173+
# end
174+
175+
# @testset "Density measures and Radon-Nikodym" begin
176+
# x = randn()
177+
# let d = ∫(𝒹(Cauchy(), Normal()), Normal())
178+
# @test logdensity(d, x) ≈ logdensity(Cauchy(), x)
179+
# end
180+
181+
# let f = 𝒹(∫(x -> x^2, Normal()), Normal())
182+
# @test f(x) ≈ x^2
183+
# end
184+
185+
# let d = ∫exp(log𝒹(Cauchy(), Normal()), Normal())
186+
# @test logdensity(d, x) ≈ logdensity(Cauchy(), x)
187+
# end
188+
189+
# let f = log𝒹(∫exp(x -> x^2, Normal()), Normal())
190+
# @test f(x) ≈ x^2
191+
# end
192+
# end

0 commit comments

Comments
 (0)