Skip to content

Commit 3ac7785

Browse files
EvizeroEvizero
committed
finish SVHN2 tests
1 parent 148cdd0 commit 3ac7785

File tree

2 files changed

+201
-6
lines changed

2 files changed

+201
-6
lines changed

‎src/SVHN2/interface.jl

Lines changed: 7 additions & 6 deletions
Original file line numberDiff line numberDiff line change
@@ -75,12 +75,12 @@ for (FUN, PATH, COUNT, DESC) in (
7575

7676
function ($FUN)(::Type{T}; dir = nothing) where T
7777
path = datafile(DEPNAME, $PATH, dir)
78-
images = matopen(io->read(io, "X"), path)
78+
images = matopen(io->read(io, "X"), path)::Array{UInt8,4}
7979
bytes_to_type(T, images)
8080
end
8181

8282
function ($FUN)(::Type{T}, indices; dir = nothing) where T
83-
images = ($FUN)(T, dir = dir)
83+
images = ($FUN)(T, dir = dir)::Array{T,4}
8484
images[:,:,:,indices]
8585
end
8686
end
@@ -124,11 +124,11 @@ for (FUN, PATH, COUNT, DESC) in (
124124
function ($FUN)(; dir = nothing)
125125
path = datafile(DEPNAME, $PATH, dir)
126126
labels = matopen(io->read(io, "y"), path)
127-
Vector{Int}(vec(labels))
127+
Vector{Int}(vec(labels))::Vector{Int}
128128
end
129129

130130
function ($FUN)(indices; dir = nothing)
131-
labels = ($FUN)(dir = dir)
131+
labels = ($FUN)(dir = dir)::Vector{Int}
132132
labels[indices]
133133
end
134134
end
@@ -184,8 +184,9 @@ for (FUN, PATH, DESC) in (
184184
function ($FUN)(::Type{T}; dir = nothing) where T
185185
path = datafile(DEPNAME, $PATH, dir)
186186
vars = matread(path)
187-
images, labels = vars["X"], vars["y"]
188-
bytes_to_type(T, images), Vector{Int}(vec(labels))
187+
images = vars["X"]::Array{UInt8,4}
188+
labels = vars["y"]
189+
bytes_to_type(T, images), Vector{Int}(vec(labels))::Vector{Int}
189190
end
190191

191192
function ($FUN)(::Type{T}, indices; dir = nothing) where T

‎test/tst_svhn2.jl

Lines changed: 194 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -57,6 +57,200 @@ else
5757
data_dir = withenv("DATADEPS_ALWAY_ACCEPT"=>"true") do
5858
datadep"SVHN2"
5959
end
60+
61+
@testset "Images" begin
62+
X_train = @inferred SVHN2.traintensor()
63+
X_test = @inferred SVHN2.testtensor()
64+
X_extra = @inferred SVHN2.extratensor()
65+
@test size(X_train, 4) == 73_257
66+
@test size(X_test, 4) == 26_032
67+
@test size(X_extra, 4) == 531_131
68+
69+
# Sanity check that the first trainimage is not the
70+
# first testimage nor extra image
71+
@test X_train[:,:,:,1] != X_test[:,:,:,1]
72+
@test X_train[:,:,:,1] != X_extra[:,:,:,1]
73+
@test X_test[:,:,:,1] != X_extra[:,:,:,1]
74+
# Make sure other integer types work as indicies
75+
@test SVHN2.testtensor(0xBAE) == SVHN2.testtensor(2990)
76+
77+
@test reinterpret(UInt8, X_train)[11:13, 12:14, 1, 1] == [
78+
0x5a 0x5c 0x5b
79+
0x5c 0x5b 0x5d
80+
0x5d 0x57 0x59
81+
]
82+
@test reinterpret(UInt8, X_test)[11:13, 12:14, 1, 1] == [
83+
0x28 0x2f 0x33
84+
0x2e 0x38 0x3b
85+
0x2d 0x37 0x3b
86+
]
87+
@test reinterpret(UInt8, X_extra)[11:13, 12:14, 1, 1] == [
88+
0x51 0x51 0x50
89+
0x53 0x4e 0x4c
90+
0x52 0x4c 0x49
91+
]
92+
93+
# These tests check if the functions return internaly
94+
# consistent results for different parameters (e.g. index
95+
# as int or as vector). That means no matter how you
96+
# specify an index, you will always get the same result
97+
# for a specific index.
98+
for (image_fun, T, nimages) in (
99+
(SVHN2.testtensor, UInt8, 26_032),
100+
(SVHN2.testtensor, Int, 26_032),
101+
(SVHN2.testtensor, Float64, 26_032),
102+
(SVHN2.testtensor, Float32, 26_032),
103+
(SVHN2.testtensor, N0f8, 26_032),
104+
)
105+
@testset "$image_fun with T=$T" begin
106+
# whole image set
107+
A = @inferred image_fun(T)
108+
@test typeof(A) <: Array{T,4}
109+
@test size(A) == (32,32,3,nimages)
110+
111+
@test_throws BoundsError image_fun(T,-1)
112+
@test_throws BoundsError image_fun(T,0)
113+
@test_throws BoundsError image_fun(T,nimages+1)
114+
115+
@testset "load single images" begin
116+
# Sample a few random images to compare
117+
for i = rand(1:nimages, 3)
118+
A_i = @inferred image_fun(T,i)
119+
@test typeof(A_i) <: Array{T,3}
120+
@test size(A_i) == (32,32,3)
121+
@test A_i == A[:,:,:,i]
122+
end
123+
end
124+
125+
@testset "load multiple images" begin
126+
A_5_10 = @inferred image_fun(T,5:10)
127+
@test typeof(A_5_10) <: Array{T,4}
128+
@test size(A_5_10) == (32,32,3,6)
129+
for i = 1:6
130+
@test A_5_10[:,:,:,i] == A[:,:,:,i+4]
131+
end
132+
133+
# also test edge cases `1`, `nimages`
134+
indices = [10,3,9,1,nimages]
135+
A_vec = image_fun(T,indices)
136+
A_vec_f = image_fun(T,Vector{Int32}(indices))
137+
@test typeof(A_vec) <: Array{T,4}
138+
@test typeof(A_vec_f) <: Array{T,4}
139+
@test size(A_vec) == (32,32,3,5)
140+
@test size(A_vec_f) == (32,32,3,5)
141+
for i in 1:5
142+
@test A_vec[:,:,:,i] == A[:,:,:,indices[i]]
143+
@test A_vec[:,:,:,i] == A_vec_f[:,:,:,i]
144+
end
145+
end
146+
end
147+
end
148+
end
149+
150+
@testset "Labels" begin
151+
# Sanity check that the first trainlabel is not also
152+
# the first testlabel
153+
@test SVHN2.trainlabels(1) != SVHN2.testlabels(1)
154+
@test SVHN2.trainlabels(1) != SVHN2.extralabels(1)
155+
@test SVHN2.testlabels(1) != SVHN2.extralabels(1)
156+
157+
# Check a few hand picked examples. I looked at both the
158+
# pictures and the native output to make sure these
159+
# values are correspond to the image at the same index.
160+
@test SVHN2.trainlabels(1) === 1
161+
@test SVHN2.trainlabels(2) === 9
162+
@test SVHN2.trainlabels(1337) === 2
163+
@test SVHN2.trainlabels(0xCAF) === 3
164+
@test SVHN2.trainlabels(73_257) === 9
165+
@test SVHN2.testlabels(1) === 5
166+
@test SVHN2.testlabels(4) === 10
167+
@test SVHN2.testlabels(0xDAD) === 4
168+
@test SVHN2.testlabels(26_032) === 7
169+
@test SVHN2.extralabels(1) === 4
170+
@test SVHN2.extralabels(3) === 8
171+
@test SVHN2.extralabels(531_131) === 4
172+
173+
# These tests check if the functions return internaly
174+
# consistent results for different parameters (e.g. index
175+
# as int or as vector). That means no matter how you
176+
# specify an index, you will always get the same result
177+
# for a specific index.
178+
# -- However, technically these tests do not check if
179+
# these are the actual SVHN labels of that index!
180+
for (label_fun, nlabels) in
181+
((SVHN2.trainlabels, 73_257),
182+
(SVHN2.testlabels, 26_032),
183+
(SVHN2.extralabels, 531_131))
184+
@testset "$label_fun" begin
185+
# whole label set
186+
A = @inferred label_fun()
187+
@test typeof(A) <: Vector{Int64}
188+
@test size(A) == (nlabels,)
189+
190+
@testset "load single label" begin
191+
# Sample a few random labels to compare
192+
for i = rand(1:nlabels, 10)
193+
A_i = @inferred label_fun(i)
194+
@test typeof(A_i) <: Int64
195+
@test A_i == A[i]
196+
end
197+
end
198+
199+
@testset "load multiple labels" begin
200+
A_5_10 = @inferred label_fun(5:10)
201+
@test typeof(A_5_10) <: Vector{Int64}
202+
@test size(A_5_10) == (6,)
203+
for i = 1:6
204+
@test A_5_10[i] == A[i+4]
205+
end
206+
207+
# also test edge cases `1`, `nlabels`
208+
indices = [10,3,9,1,nlabels]
209+
A_vec = @inferred label_fun(indices)
210+
A_vec_f = @inferred label_fun(Vector{Int32}(indices))
211+
@test typeof(A_vec) <: Vector{Int64}
212+
@test typeof(A_vec_f) <: Vector{Int64}
213+
@test size(A_vec) == (5,)
214+
@test size(A_vec_f) == (5,)
215+
for i in 1:5
216+
@test A_vec[i] == A[indices[i]]
217+
@test A_vec[i] == A_vec_f[i]
218+
end
219+
end
220+
end
221+
end
222+
end
223+
224+
# Check against the already tested tensor and labels functions
225+
@testset "Data" begin
226+
for (data_fun, feature_fun, label_fun, nobs) in
227+
((SVHN2.testdata, SVHN2.testtensor, SVHN2.testlabels, 26_032),)
228+
@testset "check $data_fun against $feature_fun and $label_fun" begin
229+
data, labels = @inferred data_fun()
230+
@test data == @inferred feature_fun()
231+
@test labels == @inferred label_fun()
232+
233+
for i = rand(1:nobs, 10)
234+
d_i, l_i = @inferred data_fun(i)
235+
@test d_i == @inferred feature_fun(i)
236+
@test l_i == @inferred label_fun(i)
237+
end
238+
239+
data, labels = @inferred data_fun(5:10)
240+
@test data == @inferred feature_fun(5:10)
241+
@test labels == @inferred label_fun(5:10)
242+
243+
data, labels = @inferred data_fun(Int, 5:10)
244+
@test data == @inferred feature_fun(Int, 5:10)
245+
@test labels == @inferred label_fun(5:10)
246+
247+
indices = [10,3,9,1,nobs]
248+
data, labels = @inferred data_fun(indices)
249+
@test data == @inferred feature_fun(indices)
250+
@test labels == @inferred label_fun(indices)
251+
end
252+
end
253+
end
60254
end
61255

62256
end

0 commit comments

Comments
 (0)