-
Notifications
You must be signed in to change notification settings - Fork 2
/
Copy pathutils.py
283 lines (230 loc) · 10.9 KB
/
utils.py
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
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
import os
import torch
import dgl
import dgl.data.citation_graph as dglcitationgraph
import torch_geometric as pyg
import numpy as np
import networkx as nx
import scipy.sparse as sp
import torch_geometric.utils as pygutils
from os.path import join as opjoin
def index_to_mask(index, size):
mask = torch.zeros(size, dtype=torch.bool, device=index.device)
mask[index] = 1
return mask
def resplit(dataset, data, full_sup, num_classes, num_nodes, num_per_class):
if dataset in ['cora', 'citeseer', 'pubmed']:
if full_sup:
perm = torch.randperm(data[2].shape[0])
test_index = perm[:500]
val_index = perm[500:1500]
train_index = perm[1500:]
data[3] = index_to_mask(train_index, size=num_nodes)
data[4] = index_to_mask(val_index, size=num_nodes)
data[5] = index_to_mask(test_index, size=num_nodes)
else:
indices = []
for i in range(num_classes):
index = (data[2].long() == i).nonzero().view(-1)
index = index[torch.randperm(index.size(0))]
indices.append(index)
train_index = torch.cat([i[ : num_per_class] for i in indices], dim=0)
rest_index = torch.cat([i[num_per_class : ] for i in indices], dim=0)
rest_index = rest_index[torch.randperm(rest_index.size(0))]
data[3] = index_to_mask(train_index, size=num_nodes)
data[4] = index_to_mask(rest_index[:500], size=num_nodes)
data[5] = index_to_mask(rest_index[500:1500], size=num_nodes)
elif dataset in ['coauthorcs']:
if full_sup:
raise NotImplementedError
else:
train_index = []
val_index = []
test_index = []
for i in range(num_classes):
index = (data[2].long() == i).nonzero().view(-1)
index = index[torch.randperm(index.size(0))]
if len(index) > num_per_class + 30:
train_index.append(index[ : num_per_class])
val_index.append(index[num_per_class : num_per_class + 30])
test_index.append(index[num_per_class + 30:])
else:
continue
train_index = torch.cat(train_index)
val_index = torch.cat(val_index)
test_index = torch.cat(test_index)
data[3] = index_to_mask(train_index, size=num_nodes)
data[4] = index_to_mask(val_index, size=num_nodes)
data[5] = index_to_mask(test_index, size=num_nodes)
return data
def load_data(config):
if not config['data']['dataset'] in config['data']['all_datasets']:
raise NotImplementedError
else:
if config['data']['implement'] == 'dgl':
graph = get_dgl_dataset(dataroot=config['path']['dataroot'],
dataset=config['data']['dataset'])
graph = rewarp_graph(graph, config)
features = torch.tensor(graph.features, dtype=torch.float)
if config['data']['feature_prenorm']:
features = row_normalization(features)
labels = torch.tensor(graph.labels, dtype=torch.long)
idx_train = torch.tensor(graph.train_mask, dtype=torch.bool)
idx_val = torch.tensor(graph.val_mask, dtype=torch.bool)
idx_test = torch.tensor(graph.test_mask, dtype=torch.bool)
g_nx = graph.graph
if config['data']['add_slflp']:
g_nx.remove_edges_from(nx.selfloop_edges(g_nx))
g_nx.add_edges(zip(g_nx.nodes(), g_nx.nodes()))
graph_skeleton = dgl.DGLGraph(g_nx)
return [graph_skeleton, features, labels, idx_train, idx_val, idx_test]
elif config['data']['implement'] == 'pyg':
graph = get_pyg_dataset(dataroot=config['path']['dataroot'], dataset=config['data']['dataset'])
graph = rewarp_graph(graph, config)
data = graph.data
idx_train = data.train_mask
idx_val = data.val_mask
idx_test = data.test_mask
labels = data.y
num_nodes = data.num_nodes
features = data.x
if config['data']['feature_prenorm']:
features = row_normalization(features)
edge_index = data.edge_index
if config['data']['add_slflp']:
edge_index = pygutils.add_self_loops(edge_index)[0]
graph_skeleton = dgl.DGLGraph()
graph_skeleton.add_nodes(num_nodes)
graph_skeleton.add_edges(edge_index[0, :], edge_index[1, :])
return [graph_skeleton, features, labels, idx_train, idx_val, idx_test]
else:
raise NotImplementedError
def dummy_normalization(mx):
if isinstance(mx, np.ndarray) or isinstance(mx, sp.csr.csr_matrix):
pass
elif isinstance(mx, sp.lil.lil_matrix):
mx = np.asarray(mx.todense())
else:
raise NotImplementedError
return mx
def get_dgl_dataset(dataroot, dataset):
dglcitationgraph._normalize = dummy_normalization
dglcitationgraph._preprocess_features = dummy_normalization
if dataset == 'cora':
graph = dgl.data.CoraDataset()
elif dataset in ['citeseer', 'pubmed']:
graph = dgl.data.CitationGraphDataset(name=dataset)
elif dataset == 'coauthorcs':
np.load.__defaults__ = (None, True, True, 'ASCII')
graph = dgl.data.Coauthor(name='cs')
np.load.__defaults__ = (None, False, True, 'ASCII')
else:
raise NotImplementedError
return graph
def get_pyg_dataset(dataroot, dataset):
if dataset in ['cora', 'citeseer', 'pubmed']:
graph = pyg.datasets.Planetoid(root=opjoin(dataroot, dataset), name=dataset.capitalize())
elif dataset == 'coauthorcs':
graph = pyg.datasets.Coauthor(root=opjoin(dataroot, dataset), name='CS')
else:
raise NotImplementedError
return graph
def normalize(mx):
"""Row-normalize sparse matrix"""
rowsum = np.array(mx.sum(1))
r_inv = np.power(rowsum, -1).flatten()
r_inv[np.isinf(r_inv)] = 0.
r_mat_inv = sp.diags(r_inv)
mx = r_mat_inv.dot(mx)
return mx
def accuracy(output, labels):
preds = output.max(1)[1].type_as(labels)
correct = preds.eq(labels).double()
correct_sum = correct.sum()
return correct_sum / len(labels), correct
def sparse_mx_to_torch_sparse_tensor(sparse_mx):
"""Convert a scipy sparse matrix to a torch sparse tensor."""
sparse_mx = sparse_mx.tocoo().astype(np.float32)
indices = torch.from_numpy(
np.vstack((sparse_mx.row, sparse_mx.col)).astype(np.int64))
values = torch.from_numpy(sparse_mx.data)
shape = torch.Size(sparse_mx.shape)
return torch.sparse.FloatTensor(indices, values, shape)
# Actually Sacred will automatically save source code files added by its add_source_file method (please see line 28 in config.py)
# But it adds a md5 hash string after the file name, and saves the source codes for every run in the same file
# So here we manually call its save_file method to save the source codes of given name in the location speicified by us
# And finally delete the source codes saved by Sacred
def save_source(run):
if run.observers:
for source_file, _ in run.experiment_info['sources']:
os.makedirs(os.path.dirname('{0}/source/{1}'.format(run.observers[0].dir, source_file)), exist_ok=True)
run.observers[0].save_file(source_file, 'source/{0}'.format(source_file))
sacred_source_path = f'{run.observers[0].basedir}/_sources'
# if os.path.exists(sacred_source_path):
# shutil.rmtree(sacred_source_path)
def adjust_learning_rate(optimizer, epoch, lr_down_epoch_list, logger):
if epoch != 0 and epoch in lr_down_epoch_list:
opt_name = list(dict(optimizer=optimizer).keys())[0]
logger.info('update learning rate of ' + opt_name)
for param_group in optimizer.param_groups:
param_group['lr'] = param_group['lr'] * 0.1
logger.info(param_group['lr'])
def check_before_pkl(data):
if type(data) == list or type(data) == tuple:
for each in data:
check_before_pkl(each)
elif type(data) == dict:
for key in data.keys():
check_before_pkl(data[key])
else:
assert not isinstance(data, torch.Tensor)
def row_normalization(features):
## normalize the feature matrix by its row sum
rowsum = features.sum(dim=1)
inv_rowsum = torch.pow(rowsum, -1)
inv_rowsum[torch.isinf(inv_rowsum)] = 0.
features = features * inv_rowsum[..., None]
return features
def rewarp_graph(graph, config):
if pyg.__version__ in ['1.4.2', '1.3.2']:
pyg_corafull_type = pyg.datasets.cora_full.CoraFull
else:
pyg_corafull_type = pyg.datasets.citation_full.CoraFull # pylint: disable=no-member
if isinstance(graph, dgl.data.gnn_benckmark.Coauthor) or \
isinstance(graph, dgl.data.gnn_benckmark.CoraFull):
graph = PseudoDGLGraph(graph)
pseudo_data = [None, None, graph.labels, None, None, None]
_, _, _, train_mask, val_mask, test_mask = resplit(dataset=config['data']['dataset'],
data=pseudo_data,
full_sup=config['data']['full_sup'],
num_classes=graph.num_classes,
num_nodes=graph.num_nodes,
num_per_class=config['data']['label_per_class'])
graph.train_mask = train_mask
graph.val_mask = val_mask
graph.test_mask = test_mask
elif isinstance(graph, pyg_corafull_type) or \
isinstance(graph, pyg.datasets.coauthor.Coauthor):
pseudo_data = [None, None, graph.data.y, None, None, None]
_, _, _, train_mask, val_mask, test_mask = resplit(dataset=config['data']['dataset'],
data=pseudo_data,
full_sup=config['data']['full_sup'],
num_classes=torch.unique(graph.data.y).shape[0],
num_nodes=graph.data.num_nodes,
num_per_class=config['data']['label_per_class'])
graph.data.train_mask = train_mask
graph.data.val_mask = val_mask
graph.data.test_mask = test_mask
else:
pass
return graph
class PseudoDGLGraph():
def __init__(self, graph):
self.graph = graph.data[0].to_networkx()
self.features = graph.data[0].ndata['feat']
self.labels = graph.data[0].ndata['label']
self.num_classes = torch.unique(self.labels).shape[0]
self.num_nodes = graph.data[0].number_of_nodes()
self.train_mask = None
self.val_mask = None
self.test_mask = None