Skip to content

Commit f37c40d

Browse files
CarloLucibelloCarloLucibello
authored
fix tests (#266)
* fix laplacian test * fix tests * to64 test
1 parent 0e6c370 commit f37c40d

File tree

6 files changed

+276
-268
lines changed

6 files changed

+276
-268
lines changed

‎.gitignore

Lines changed: 1 addition & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -9,3 +9,4 @@ Manifest.toml
99
.vscode
1010
LocalPreferences.toml
1111
.DS_Store
12+
/test.jl

‎src/GNNGraphs/query.jl

Lines changed: 3 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -284,6 +284,9 @@ function normalized_adjacency(g::GNNGraph, T::DataType = Float32;
284284
A = A + I
285285
end
286286
degs = vec(sum(A; dims = 2))
287+
ChainRulesCore.ignore_derivatives() do
288+
@assert all(!iszero, degs) "Graph contains isolated nodes, cannot compute `normalized_adjacency`."
289+
end
287290
inv_sqrtD = Diagonal(inv.(sqrt.(degs)))
288291
return inv_sqrtD * A * inv_sqrtD
289292
end

‎test/GNNGraphs/query.jl

Lines changed: 163 additions & 165 deletions
Original file line numberDiff line numberDiff line change
@@ -1,193 +1,191 @@
1-
@testset "Query" begin
2-
@testset "is_bidirected" begin
3-
g = rand_graph(10, 20, bidirected = true, graph_type = GRAPH_T)
4-
@test is_bidirected(g)
1+
@testset "is_bidirected" begin
2+
g = rand_graph(10, 20, bidirected = true, graph_type = GRAPH_T)
3+
@test is_bidirected(g)
54

6-
g = rand_graph(10, 20, bidirected = false, graph_type = GRAPH_T)
7-
@test !is_bidirected(g)
5+
g = rand_graph(10, 20, bidirected = false, graph_type = GRAPH_T)
6+
@test !is_bidirected(g)
7+
end
8+
9+
@testset "has_multi_edges" begin if GRAPH_T == :coo
10+
s = [1, 1, 2, 3]
11+
t = [2, 2, 2, 4]
12+
g = GNNGraph(s, t, graph_type = GRAPH_T)
13+
@test has_multi_edges(g)
14+
15+
s = [1, 2, 2, 3]
16+
t = [2, 1, 2, 4]
17+
g = GNNGraph(s, t, graph_type = GRAPH_T)
18+
@test !has_multi_edges(g)
19+
end end
20+
21+
@testset "edges" begin
22+
g = rand_graph(4, 10, graph_type = GRAPH_T)
23+
@test edgetype(g) <: Graphs.Edge
24+
for e in edges(g)
25+
@test e isa Graphs.Edge
826
end
27+
end
28+
29+
@testset "has_isolated_nodes" begin
30+
s = [1, 2, 3]
31+
t = [2, 3, 2]
32+
g = GNNGraph(s, t, graph_type = GRAPH_T)
33+
@test has_isolated_nodes(g) == false
34+
@test has_isolated_nodes(g, dir = :in) == true
35+
end
36+
37+
@testset "has_self_loops" begin
38+
s = [1, 1, 2, 3]
39+
t = [2, 2, 2, 4]
40+
g = GNNGraph(s, t, graph_type = GRAPH_T)
41+
@test has_self_loops(g)
942

10-
@testset "has_multi_edges" begin if GRAPH_T == :coo
43+
s = [1, 1, 2, 3]
44+
t = [2, 2, 3, 4]
45+
g = GNNGraph(s, t, graph_type = GRAPH_T)
46+
@test !has_self_loops(g)
47+
end
48+
49+
@testset "degree" begin
50+
@testset "unweighted" begin
1151
s = [1, 1, 2, 3]
1252
t = [2, 2, 2, 4]
1353
g = GNNGraph(s, t, graph_type = GRAPH_T)
14-
@test has_multi_edges(g)
1554

16-
s = [1, 2, 2, 3]
17-
t = [2, 1, 2, 4]
18-
g = GNNGraph(s, t, graph_type = GRAPH_T)
19-
@test !has_multi_edges(g)
20-
end end
21-
22-
@testset "edges" begin
23-
g = rand_graph(4, 10, graph_type = GRAPH_T)
24-
@test edgetype(g) <: Graphs.Edge
25-
for e in edges(g)
26-
@test e isa Graphs.Edge
55+
@test degree(g) == degree(g; dir = :out) == [2, 1, 1, 0] # default is outdegree
56+
@test degree(g; dir = :in) == [0, 3, 0, 1]
57+
@test degree(g; dir = :both) == [2, 4, 1, 1]
58+
@test eltype(degree(g, Float32)) == Float32
59+
60+
if TEST_GPU
61+
g_gpu = g |> gpu
62+
d = degree(g)
63+
d_gpu = degree(g_gpu)
64+
@test d_gpu isa CuVector{Int}
65+
@test Array(d_gpu) == d
2766
end
2867
end
2968

30-
@testset "has_isolated_nodes" begin
31-
s = [1, 2, 3]
32-
t = [2, 3, 2]
33-
g = GNNGraph(s, t, graph_type = GRAPH_T)
34-
@test has_isolated_nodes(g) == false
35-
@test has_isolated_nodes(g, dir = :in) == true
36-
end
37-
38-
@testset "has_self_loops" begin
69+
@testset "weighted" begin
70+
# weighted degree
3971
s = [1, 1, 2, 3]
4072
t = [2, 2, 2, 4]
41-
g = GNNGraph(s, t, graph_type = GRAPH_T)
42-
@test has_self_loops(g)
43-
44-
s = [1, 1, 2, 3]
45-
t = [2, 2, 3, 4]
46-
g = GNNGraph(s, t, graph_type = GRAPH_T)
47-
@test !has_self_loops(g)
48-
end
73+
eweight = [0.1, 2.1, 1.2, 1]
74+
g = GNNGraph((s, t, eweight), graph_type = GRAPH_T)
75+
@test degree(g) == [2.2, 1.2, 1.0, 0.0]
76+
@test degree(g, edge_weight = nothing) == degree(g)
77+
d = degree(g, edge_weight = false)
78+
if GRAPH_T == :coo
79+
@test d == [2, 1, 1, 0]
80+
else
81+
# Adjacency matrix representation cannot disambiguate multiple edges
82+
# and edge weights
83+
@test d == [1, 1, 1, 0]
84+
end
85+
@test eltype(d) <: Integer
86+
if GRAPH_T == :coo
87+
# TODO use the @test option broken = (GRAPH_T != :coo) on julia >= 1.7
88+
@test degree(g, edge_weight = 2 * eweight) == [4.4, 2.4, 2.0, 0.0]
89+
else
90+
@test_broken degree(g, edge_weight = 2 * eweight) == [4.4, 2.4, 2.0, 0.0]
91+
end
4992

50-
@testset "degree" begin
51-
@testset "unweighted" begin
52-
s = [1, 1, 2, 3]
53-
t = [2, 2, 2, 4]
54-
g = GNNGraph(s, t, graph_type = GRAPH_T)
55-
56-
@test degree(g) == degree(g; dir = :out) == [2, 1, 1, 0] # default is outdegree
57-
@test degree(g; dir = :in) == [0, 3, 0, 1]
58-
@test degree(g; dir = :both) == [2, 4, 1, 1]
59-
@test eltype(degree(g, Float32)) == Float32
60-
61-
if TEST_GPU
62-
g_gpu = g |> gpu
63-
d = degree(g)
64-
d_gpu = degree(g_gpu)
65-
@test d_gpu isa CuVector{Int}
66-
@test Array(d_gpu) == d
67-
end
93+
if TEST_GPU
94+
g_gpu = g |> gpu
95+
d = degree(g)
96+
d_gpu = degree(g_gpu)
97+
@test d_gpu isa CuVector{Float32}
98+
@test Array(d_gpu) ≈ d
6899
end
100+
@testset "gradient" begin
101+
gw = gradient(eweight) do w
102+
g = GNNGraph((s, t, w), graph_type = GRAPH_T)
103+
sum(degree(g, edge_weight = false))
104+
end[1]
69105

70-
@testset "weighted" begin
71-
# weighted degree
72-
s = [1, 1, 2, 3]
73-
t = [2, 2, 2, 4]
74-
eweight = [0.1, 2.1, 1.2, 1]
75-
g = GNNGraph((s, t, eweight), graph_type = GRAPH_T)
76-
@test degree(g) == [2.2, 1.2, 1.0, 0.0]
77-
@test degree(g, edge_weight = nothing) == degree(g)
78-
d = degree(g, edge_weight = false)
79-
if GRAPH_T == :coo
80-
@test d == [2, 1, 1, 0]
81-
else
82-
# Adjacency matrix representation cannot disambiguate multiple edges
83-
# and edge weights
84-
@test d == [1, 1, 1, 0]
85-
end
86-
@test eltype(d) <: Integer
87-
if GRAPH_T == :coo
88-
# TODO use the @test option broken = (GRAPH_T != :coo) on julia >= 1.7
89-
@test degree(g, edge_weight = 2 * eweight) == [4.4, 2.4, 2.0, 0.0]
90-
else
91-
@test_broken degree(g, edge_weight = 2 * eweight) == [4.4, 2.4, 2.0, 0.0]
92-
end
106+
@test gw === nothing
93107

94-
if TEST_GPU
95-
g_gpu = g |> gpu
96-
d = degree(g)
97-
d_gpu = degree(g_gpu)
98-
@test d_gpu isa CuVector{Float32}
99-
@test Array(d_gpu) ≈ d
100-
end
101-
@testset "gradient" begin
102-
gw = gradient(eweight) do w
103-
g = GNNGraph((s, t, w), graph_type = GRAPH_T)
104-
sum(degree(g, edge_weight = false))
105-
end[1]
106-
107-
@test gw === nothing
108-
109-
gw = gradient(eweight) do w
110-
g = GNNGraph((s, t, w), graph_type = GRAPH_T)
111-
sum(degree(g, edge_weight = true))
112-
end[1]
113-
114-
if GRAPH_T == :sparse
115-
@test_broken gw isa Vector{Float64}
116-
@test gw isa AbstractVector{Float64}
117-
else
118-
@test gw isa Vector{Float64}
119-
end
108+
gw = gradient(eweight) do w
109+
g = GNNGraph((s, t, w), graph_type = GRAPH_T)
110+
sum(degree(g, edge_weight = true))
111+
end[1]
112+
113+
if GRAPH_T == :sparse
114+
@test_broken gw isa Vector{Float64}
115+
@test gw isa AbstractVector{Float64}
116+
else
117+
@test gw isa Vector{Float64}
120118
end
121119
end
122120
end
121+
end
123122

124-
@testset "laplacian_matrix" begin
125-
g = rand_graph(10, 30, graph_type = GRAPH_T)
126-
A = adjacency_matrix(g)
127-
D = Diagonal(vec(sum(A, dims = 2)))
128-
L = laplacian_matrix(g)
129-
@test eltype(L) == eltype(g)
130-
@test L ≈ D - A
131-
end
132-
133-
@testset "laplacian_lambda_max" begin
134-
s = [1, 2, 3, 4, 5, 1, 2, 3, 4, 5]
135-
t = [2, 3, 4, 5, 1, 5, 1, 2, 3, 4]
136-
g = GNNGraph(s, t)
137-
@test laplacian_lambda_max(g) ≈ Float32(1.809017)
138-
data1 = [g for i in 1:5]
139-
gall1 = Flux.batch(data1)
140-
@test laplacian_lambda_max(gall1) ≈ [Float32(1.809017) for i in 1:5]
141-
data2 = [rand_graph(10, 20) for i in 1:3]
142-
gall2 = Flux.batch(data2)
143-
@test length(laplacian_lambda_max(gall2)) == 3
144-
end
145-
146-
@testset "adjacency_matrix" begin
147-
a = sprand(5, 5, 0.5)
148-
abin = map(x -> x > 0 ? 1 : 0, a)
123+
@testset "laplacian_matrix" begin
124+
g = rand_graph(10, 30, graph_type = GRAPH_T)
125+
A = adjacency_matrix(g)
126+
D = Diagonal(vec(sum(A, dims = 2)))
127+
L = laplacian_matrix(g)
128+
@test eltype(L) == eltype(g)
129+
@test L ≈ D - A
130+
end
149131

150-
g = GNNGraph(a, graph_type = GRAPH_T)
151-
A = adjacency_matrix(g, Float32)
152-
@test A ≈ a
153-
@test eltype(A) == Float32
132+
@testset "laplacian_lambda_max" begin
133+
s = [1, 2, 3, 4, 5, 1, 2, 3, 4, 5]
134+
t = [2, 3, 4, 5, 1, 5, 1, 2, 3, 4]
135+
g = GNNGraph(s, t)
136+
@test laplacian_lambda_max(g) ≈ Float32(1.809017)
137+
data1 = [g for i in 1:5]
138+
gall1 = Flux.batch(data1)
139+
@test laplacian_lambda_max(gall1) ≈ [Float32(1.809017) for i in 1:5]
140+
data2 = [rand_graph(10, 20) for i in 1:3]
141+
gall2 = Flux.batch(data2)
142+
@test length(laplacian_lambda_max(gall2, add_self_loops=true)) == 3
143+
end
154144

155-
Abin = adjacency_matrix(g, Float32, weighted = false)
156-
@test Abin ≈ abin
157-
@test eltype(Abin) == Float32
145+
@testset "adjacency_matrix" begin
146+
a = sprand(5, 5, 0.5)
147+
abin = map(x -> x > 0 ? 1 : 0, a)
158148

159-
@testset "gradient" begin
160-
s = [1, 2, 3]
161-
t = [2, 3, 1]
162-
w = [0.1, 0.1, 0.2]
163-
gw = gradient(w) do w
164-
g = GNNGraph(s, t, w, graph_type = GRAPH_T)
165-
A = adjacency_matrix(g, weighted = false)
166-
sum(A)
167-
end[1]
168-
@test gw === nothing
149+
g = GNNGraph(a, graph_type = GRAPH_T)
150+
A = adjacency_matrix(g, Float32)
151+
@test A ≈ a
152+
@test eltype(A) == Float32
169153

170-
gw = gradient(w) do w
171-
g = GNNGraph(s, t, w, graph_type = GRAPH_T)
172-
A = adjacency_matrix(g, weighted = true)
173-
sum(A)
174-
end[1]
154+
Abin = adjacency_matrix(g, Float32, weighted = false)
155+
@test Abin ≈ abin
156+
@test eltype(Abin) == Float32
175157

176-
@test gw == [1, 1, 1]
177-
end
158+
@testset "gradient" begin
159+
s = [1, 2, 3]
160+
t = [2, 3, 1]
161+
w = [0.1, 0.1, 0.2]
162+
gw = gradient(w) do w
163+
g = GNNGraph(s, t, w, graph_type = GRAPH_T)
164+
A = adjacency_matrix(g, weighted = false)
165+
sum(A)
166+
end[1]
167+
@test gw === nothing
168+
169+
gw = gradient(w) do w
170+
g = GNNGraph(s, t, w, graph_type = GRAPH_T)
171+
A = adjacency_matrix(g, weighted = true)
172+
sum(A)
173+
end[1]
174+
175+
@test gw == [1, 1, 1]
176+
end
178177

179-
@testset "khop_adj" begin
180-
s = [1, 2, 3]
181-
t = [2, 3, 1]
182-
w = [0.1, 0.1, 0.2]
183-
g = GNNGraph(s, t, w)
184-
@test khop_adj(g, 2) == adjacency_matrix(g) * adjacency_matrix(g)
185-
@test khop_adj(g, 2, Int8; weighted = false) == sparse([0 0 1; 1 0 0; 0 1 0])
186-
@test khop_adj(g, 2, Int8; dir = in, weighted = false) ==
187-
sparse([0 0 1; 1 0 0; 0 1 0]')
188-
@test khop_adj(g, 1) == adjacency_matrix(g)
189-
@test eltype(khop_adj(g, 4)) == Float64
190-
@test eltype(khop_adj(g, 10, Float32)) == Float32
191-
end
178+
@testset "khop_adj" begin
179+
s = [1, 2, 3]
180+
t = [2, 3, 1]
181+
w = [0.1, 0.1, 0.2]
182+
g = GNNGraph(s, t, w)
183+
@test khop_adj(g, 2) == adjacency_matrix(g) * adjacency_matrix(g)
184+
@test khop_adj(g, 2, Int8; weighted = false) == sparse([0 0 1; 1 0 0; 0 1 0])
185+
@test khop_adj(g, 2, Int8; dir = in, weighted = false) ==
186+
sparse([0 0 1; 1 0 0; 0 1 0]')
187+
@test khop_adj(g, 1) == adjacency_matrix(g)
188+
@test eltype(khop_adj(g, 4)) == Float64
189+
@test eltype(khop_adj(g, 10, Float32)) == Float32
192190
end
193191
end

0 commit comments

Comments
 (0)