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
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
|
import pytest
import torch
from torch_geometric.data import Data
from torch_geometric.loader import ClusterData, ClusterLoader
from torch_geometric.testing import onlyFullTest, onlyOnline, withMETIS
from torch_geometric.utils import sort_edge_index
@withMETIS
def test_cluster_gcn():
adj = torch.tensor([
[1, 1, 1, 0, 1, 0],
[1, 1, 0, 1, 0, 1],
[1, 0, 1, 0, 1, 0],
[0, 1, 0, 1, 0, 1],
[1, 0, 1, 0, 1, 0],
[0, 1, 0, 1, 0, 1],
])
x = torch.tensor([
[0.0, 0.0],
[1.0, 1.0],
[2.0, 2.0],
[3.0, 3.0],
[4.0, 4.0],
[5.0, 5.0],
])
edge_index = adj.nonzero(as_tuple=False).t()
edge_attr = torch.arange(edge_index.size(1))
n_id = torch.arange(6)
data = Data(x=x, n_id=n_id, edge_index=edge_index, edge_attr=edge_attr)
data.num_nodes = 6
cluster_data = ClusterData(data, num_parts=2, log=False)
partition = cluster_data._partition(
edge_index, cluster=torch.tensor([0, 1, 0, 1, 0, 1]))
assert partition.partptr.tolist() == [0, 3, 6]
assert partition.node_perm.tolist() == [0, 2, 4, 1, 3, 5]
assert partition.edge_perm.tolist() == [
0, 2, 3, 1, 8, 9, 10, 14, 15, 16, 4, 5, 6, 7, 11, 12, 13, 17, 18, 19
]
assert cluster_data.partition.partptr.tolist() == [0, 3, 6]
assert torch.equal(
cluster_data.partition.node_perm.sort()[0],
torch.arange(data.num_nodes),
)
assert torch.equal(
cluster_data.partition.edge_perm.sort()[0],
torch.arange(data.num_edges),
)
out = cluster_data[0]
expected = data.subgraph(out.n_id)
out.validate()
assert out.num_nodes == 3
assert out.n_id.size() == (3, )
assert torch.equal(out.x, expected.x)
tmp = sort_edge_index(expected.edge_index, expected.edge_attr)
assert torch.equal(out.edge_index, tmp[0])
assert torch.equal(out.edge_attr, tmp[1])
out = cluster_data[1]
out.validate()
assert out.num_nodes == 3
assert out.n_id.size() == (3, )
expected = data.subgraph(out.n_id)
assert torch.equal(out.x, expected.x)
tmp = sort_edge_index(expected.edge_index, expected.edge_attr)
assert torch.equal(out.edge_index, tmp[0])
assert torch.equal(out.edge_attr, tmp[1])
loader = ClusterLoader(cluster_data, batch_size=1)
iterator = iter(loader)
out = next(iterator)
out.validate()
assert out.num_nodes == 3
assert out.n_id.size() == (3, )
expected = data.subgraph(out.n_id)
assert torch.equal(out.x, expected.x)
tmp = sort_edge_index(expected.edge_index, expected.edge_attr)
assert torch.equal(out.edge_index, tmp[0])
assert torch.equal(out.edge_attr, tmp[1])
out = next(iterator)
out.validate()
assert out.num_nodes == 3
assert out.n_id.size() == (3, )
expected = data.subgraph(out.n_id)
assert torch.equal(out.x, expected.x)
tmp = sort_edge_index(expected.edge_index, expected.edge_attr)
assert torch.equal(out.edge_index, tmp[0])
assert torch.equal(out.edge_attr, tmp[1])
loader = ClusterLoader(cluster_data, batch_size=2, shuffle=False)
out = next(iter(loader))
out.validate()
assert out.num_nodes == 6
assert out.n_id.size() == (6, )
expected = data.subgraph(out.n_id)
assert torch.equal(out.x, expected.x)
tmp = sort_edge_index(expected.edge_index, expected.edge_attr)
assert torch.equal(out.edge_index, tmp[0])
assert torch.equal(out.edge_attr, tmp[1])
@withMETIS
def test_keep_inter_cluster_edges():
adj = torch.tensor([
[1, 1, 1, 0, 1, 0],
[1, 1, 0, 1, 0, 1],
[1, 0, 1, 0, 1, 0],
[0, 1, 0, 1, 0, 1],
[1, 0, 1, 0, 1, 0],
[0, 1, 0, 1, 0, 1],
])
x = torch.tensor([
[0.0, 0.0],
[1.0, 1.0],
[2.0, 2.0],
[3.0, 3.0],
[4.0, 4.0],
[5.0, 5.0],
])
edge_index = adj.nonzero(as_tuple=False).t()
edge_attr = torch.arange(edge_index.size(1))
data = Data(x=x, edge_index=edge_index, edge_attr=edge_attr)
data.num_nodes = 6
cluster_data = ClusterData(data, num_parts=2, log=False,
keep_inter_cluster_edges=True)
data = cluster_data[0]
assert data.edge_index[0].min() == 0
assert data.edge_index[0].max() == 2
assert data.edge_index[1].min() == 0
assert data.edge_index[1].max() > 2
assert data.edge_index.size(1) == data.edge_attr.size(0)
data = cluster_data[1]
assert data.edge_index[0].min() == 0
assert data.edge_index[0].max() == 2
assert data.edge_index[1].min() == 0
assert data.edge_index[1].max() > 2
assert data.edge_index.size(1) == data.edge_attr.size(0)
@withMETIS
@onlyOnline
@onlyFullTest
@pytest.mark.parametrize('sparse_format', ['csr', 'csc'])
def test_cluster_gcn_correctness(get_dataset, sparse_format):
dataset = get_dataset('Cora')
data = dataset[0].clone()
data.n_id = torch.arange(data.num_nodes)
cluster_data = ClusterData(
data,
num_parts=10,
log=False,
sparse_format=sparse_format,
)
loader = ClusterLoader(cluster_data, batch_size=3, shuffle=False)
for batch1 in loader:
batch1.validate()
batch2 = data.subgraph(batch1.n_id)
assert batch1.num_nodes == batch2.num_nodes
assert batch1.num_edges == batch2.num_edges
assert torch.equal(batch1.x, batch2.x)
assert torch.equal(
batch1.edge_index,
sort_edge_index(
batch2.edge_index,
sort_by_row=sparse_format == 'csr',
),
)
if __name__ == '__main__':
import argparse
from ogb.nodeproppred import PygNodePropPredDataset
from tqdm import tqdm
parser = argparse.ArgumentParser()
parser.add_argument('--num_workers', type=int, default=0)
args = parser.parse_args()
data = PygNodePropPredDataset('ogbn-products', root='/tmp/ogb')[0]
loader = ClusterLoader(
ClusterData(data, num_parts=15_000, save_dir='/tmp/ogb/ogbn_products'),
batch_size=32,
shuffle=True,
num_workers=args.num_workers,
)
for batch in tqdm(loader):
pass
|
