-
Notifications
You must be signed in to change notification settings - Fork 53
/
optim.jl
151 lines (126 loc) · 6.61 KB
/
optim.jl
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
module TestOptim
using GaussianProcesses, StatsFuns, Distributions
using Test, Random
Random.seed!(1)
@testset "Optim" begin
d, n = 2, 20
X = rand(d, n) # uniform in unit square
mean = MeanLin(zeros(d))
kern = SE(log(0.5), 1.0)
@testset "Without likelihood" begin
y = X'rand(d) .+ sin.(X[1,:]*2π) .+ 0.1*randn(n)
noise = 0.0
# Just checks that it doesn't crash
# and that the final mll is better that the initial value
@testset "Basic" begin
gp = GPE(X, y, mean, deepcopy(kern), -3.0)
init_target = gp.target
optimize!(gp)
@test gp.target > init_target
end
@testset "Fixed kernel" begin
init_param = GaussianProcesses.get_params(kern)[1]
fixed = fix(deepcopy(kern), GaussianProcesses.get_param_names(kern)[1])
gp = GP(X, y, MeanZero(), fixed, -1.0)
init_target = gp.target
optimize!(gp)
@test gp.target > init_target
@test GaussianProcesses.get_params(kern)[1] == init_param
end
@testset "priors" begin
gp = GPE(X, y, MeanConst(0.), deepcopy(kern), noise)
init_params = GaussianProcesses.get_params(gp; domean=true, kern=true,
noise=true)
optimize!(gp)
ml_params = GaussianProcesses.get_params(gp; domean=true, kern=true,
noise=true)
priormeans = ml_params .- 2
set_priors!(gp.logNoise, [Normal(priormeans[1], 1)])
set_priors!(gp.mean, [Normal(priormeans[2], 1)])
set_priors!(gp.kernel, [Normal(priormeans[3], 1), Normal(priormeans[4], 1)])
optimize!(gp)
map_params = GaussianProcesses.get_params(gp; domean=true, kern=true,
noise=true)
@test (&)((map_params .< ml_params)...)
end
@testset "Keyword arguments" begin
gp = GPE(X, y, MeanLin(zeros(d)), deepcopy(kern), noise)
init_params = GaussianProcesses.get_params(gp; domean=true, kern=true,
noise=true)
# Check mean fixed
mean_params = GaussianProcesses.get_params(gp; domean=true, kern=false,
noise=false)
optimize!(gp; domean=false, kern=true, noise=true)
@test mean_params == GaussianProcesses.get_params(gp; domean=true,
kern=false, noise=false)
set_params!(gp, init_params; domean=true, kern=true, noise=true)
# Check kern fixed
kern_params = GaussianProcesses.get_params(gp; domean=false, kern=true,
noise=false)
optimize!(gp; domean=true, kern=false, noise=true)
@test kern_params == GaussianProcesses.get_params(gp; domean=false, kern=true,
noise=false)
set_params!(gp, init_params; domean=true, kern=true, noise=true)
# Check noise fixed
noise_params = GaussianProcesses.get_params(gp; domean=false, kern=false,
noise=true)
optimize!(gp; domean=true, kern=true, noise=false)
@test noise_params == GaussianProcesses.get_params(gp; domean=false, kern=false,
noise=true)
set_params!(gp, init_params; domean=true, kern=true, noise=true)
# Box
kern_params = GaussianProcesses.get_params(gp; domean=false, kern=true,
noise=false)
optimize!(gp, domean = false,
kernbounds = [kern_params .- .1, kern_params .+ .1])
new_kern_params = GaussianProcesses.get_params(gp; domean=false, kern=true,
noise=false)
@test (&)((@. kern_params - .1 <= new_kern_params <= kern_params + .1)...)
@test kern_params != new_kern_params
end
end
@testset "With likelihood" begin
f = X'rand(d) .+ sin.(X[1,:]*2π)
y = collect(rand(n) .< normcdf.(f)) # Binary data
lik = BernLik() # Bernoulli likelihood for binary data {0,1}
# Just checks that it doesn't crash
# and that the final mll is better that the initial value
@testset "Basic" begin
gp = GPA(X, y, MeanLin(zeros(d)), deepcopy(kern), BernLik())
init_target = gp.target
optimize!(gp)
@test gp.target > init_target
end
@testset "Keyword arguments" begin
gp = GPA(X, y, MeanLin(zeros(d)), deepcopy(kern), BernLik())
init_params = GaussianProcesses.get_params(gp; domean=true, kern=true, lik=true)
# Check mean fixed
mean_params = GaussianProcesses.get_params(gp.mean)
kern_params = GaussianProcesses.get_params(gp.kernel)
optimize!(gp; domean=false, kern=true, lik=true)
@test mean_params == GaussianProcesses.get_params(gp.mean)
@test kern_params != GaussianProcesses.get_params(gp.kernel)
set_params!(gp, init_params; domean=true, kern=true, lik=true)
# Check kern fixed
kern_params = GaussianProcesses.get_params(gp.kernel)
optimize!(gp; domean=true, kern=false, lik=true)
@test kern_params == GaussianProcesses.get_params(gp.kernel)
set_params!(gp, init_params; domean=true, kern=true, lik=true)
# Check lik fixed
lik_params = GaussianProcesses.get_params(gp.lik)
optimize!(gp; domean=true, kern=true, lik=false)
@test lik_params == GaussianProcesses.get_params(gp.lik)
set_params!(gp, init_params; domean=true, kern=true, lik=true)
# Box
kern_params = GaussianProcesses.get_params(gp; domean=false, kern=true,
lik=false)
optimize!(gp, domean = false,
kernbounds = [kern_params .- .01, kern_params .+ .01])
new_kern_params = GaussianProcesses.get_params(gp; domean=false, kern=true,
lik=false)
@test (&)((@. kern_params - .01 <= new_kern_params <= kern_params + .01)...)
@test kern_params != new_kern_params
end
end
end
end