-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathmain_PubMed_coarsen.py
237 lines (209 loc) · 11.7 KB
/
main_PubMed_coarsen.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
#!/usr/bin/env python3
# -*- coding: utf-8 -*-
"""
Created on Tue Jul 25 14:23:46 2023
@author: chris
"""
import datasets
import utility
import torch
import numpy as np
import models
import train_SIGN
import train_GCN
import writer
import argparse
import GraphCoarsening.utils as gc
parser = argparse.ArgumentParser()
parser.add_argument('--model_type', nargs='?', const='QSIGN', type=str, default='QSIGN')
parser.add_argument('--coarsen_ratio', nargs='?', const=0.75, type=int, default=0.75) #0.75 corresponds to ~5000 nodes
parser.add_argument('--max_hop', nargs='?', const=3, type=int, default=3)
parser.add_argument('--layers', nargs='?', const=3, type=int, default=3)
parser.add_argument('--hidden_channels', nargs='?', const=1536, type=int, default=1536)
parser.add_argument('--dropout', nargs='?', const=0.5, type=float, default=0.5)
parser.add_argument('--batch_norm', nargs='?', const=False, type=bool, default=False)
parser.add_argument('--lr', nargs='?', const=0.0005, type=float, default=0.0005)
parser.add_argument('--num_batch', nargs='?', const=10, type=int, default=10)
parser.add_argument('--num_epoch', nargs='?', const=400, type=int, default=400)
parser.add_argument('--multilabel', nargs='?', const=False, type=bool, default=False)
parser.add_argument('--do_eval', nargs='?', const=True, type=bool, default=True)
parser.add_argument('--residual', nargs='?', const=False, type=bool, default=False)
parser.add_argument('--print_result', nargs='?', const=True, type=bool, default=True)
args = parser.parse_args()
print(args)
model_type = args.model_type
coarsening_ratio = args.coarsen_ratio
max_hop = args.max_hop
layers = args.layers
hidden_channels = args.hidden_channels
dropout = args.dropout
batch_norm = args.batch_norm
lr = args.lr
num_batch = args.num_batch
num_epoch = args.num_epoch
multilabel = args.multilabel
do_evaluation = args.do_eval
residual = args.residual
print_result = args.print_result
dataset_name = 'PubMed'
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
#%% Prepare Dataset
# get dataset
(x, y, edge_index, train_mask, val_mask, test_mask) = datasets.get_Planetoid(name="PubMed")
num_features = x.shape[1]
num_classes = max(y) + 1
# construct networkx graph
G = utility.construct_graph(x, y, edge_index, train_mask, val_mask,
test_mask)
#%% split graph to get uncontracted train, val, and test graphs
(x_train, y_train, edge_train, train_mask, x_val, y_val, edge_val, val_mask, x_test,
y_test, edge_test, test_mask) = utility.split_graph(G, multilabel = multilabel)
# get initial label distribution
Y_dist_before = utility.get_label_distribution_tensor(y_train[train_mask], multilabel)
#%% coarsening train graph (all training nodes)
edge_train = edge_train.T
edge_train = torch.tensor(edge_train)
edge_train_flip = torch.flip(edge_train,[0]) # re-adds flipped edges that were removed by networkx
edge_train = torch.cat((edge_train, edge_train_flip), 1)
x_train = torch.tensor(x_train)
y_train = torch.tensor(y_train)
train_mask = torch.tensor(train_mask)
candidate, C_list, Gc_list = gc.coarsening(edge_train, coarsening_ratio, coarsening_method='variation_neighborhoods')
(x_train, y_train, train_mask, _, _, edge_train) = gc.load_data(train_mask, ~train_mask,
x_train, y_train,
num_classes, candidate,
C_list, Gc_list)
# converting training tensors back to networkx to remove duplicate edges
edge_train = edge_train.numpy()
edge_train = edge_train.T
G = utility.construct_graph(x_train, y_train, edge_train, train_mask, ~train_mask,
~train_mask)
#%% convert train networkx graph to tensor
x_train, y_train, edge_train, train_mask, _, _ = utility.convert_graph_to_tensor(G, multilabel)
# get final label distribution
Y_dist_after = utility.get_label_distribution_tensor(y_train[train_mask], multilabel)
# get label distribution errors
Y_dist_error = utility.compute_label_distribution_error(Y_dist_before, Y_dist_after)
print(f"Contraction Label Distribution Avg Error: {np.mean(Y_dist_error)}")
#%% Normalize Adjacency Matrices
adj_train = utility.construct_normalized_adj(edge_train, x_train.shape[0])
adj_val = utility.construct_normalized_adj(edge_val, x_val.shape[0])
adj_test = utility.construct_normalized_adj(edge_test, x_test.shape[0])
#%% Convert feature and labels to torch tensor
x_train = torch.tensor(x_train, dtype=torch.float32)
y_train = torch.tensor(y_train)
train_mask = torch.tensor(train_mask)
x_val = torch.tensor(x_val, dtype=torch.float32)
y_val = torch.tensor(y_val)
val_mask = torch.tensor(val_mask)
x_test = torch.tensor(x_test, dtype=torch.float32)
y_test = torch.tensor(y_test)
test_mask = torch.tensor(test_mask)
#%% Preparing Model
if (model_type == 'SIGN' or model_type == 'QSIGN'):
# precomputing SIGN aggregation
print("Pre-aggregating embeddings ...")
x_train_aggregated = models.precompute_SIGN_aggregation(x_train, adj_train,
max_hop)
x_val_aggregated = models.precompute_SIGN_aggregation(x_val, adj_val,
max_hop)
x_test_aggregated = models.precompute_SIGN_aggregation(x_test, adj_test,
max_hop)
print(f"Pre-aggregated embedding size: {x_train_aggregated.shape[1]}")
# in SIGN, after aggregation phase, data points are decoupled, we only need
# to pass labeled data
x_train_aggregated = x_train_aggregated[train_mask]
y_train = y_train[train_mask]
x_val_aggregated = x_val_aggregated[val_mask]
y_val = y_val[val_mask]
x_test_aggregated = x_test_aggregated[test_mask]
y_test = y_test[test_mask]
if model_type == 'QSIGN':
model = models.QMLP(hidden_channels=hidden_channels, num_layers= layers,
in_channels= x_train_aggregated.shape[1],
out_channels=num_classes, batch_norm=batch_norm,
dropout=dropout)
elif model_type == 'SIGN':
model = models.MLP(hidden_channels=hidden_channels, num_layers= layers,
in_channels= x_train_aggregated.shape[1],
out_channels=num_classes, batch_norm=batch_norm,
dropout=dropout)
elif (model_type == 'GCN'):
model = models.GCN(hidden_channels=hidden_channels, num_layers= layers,
in_channels= x_train.shape[1], out_channels=num_classes,
batch_norm=batch_norm, dropout=dropout, residual=residual)
print(model)
#%% Training
if model_type == 'SIGN' or model_type == 'QSIGN':
if do_evaluation:
(max_val_acc, max_val_f1, max_val_sens, max_val_spec, max_val_test_acc,
max_val_test_f1, max_val_test_sens, max_val_test_spec, session_memory,
train_memory, train_time_avg) = train_SIGN.train(model, device,
x_train=x_train_aggregated,
y_train=y_train,
x_val=x_val_aggregated,
y_val=y_val,
x_test=x_test_aggregated,
y_test=y_test,
multilabel=multilabel,
lr=lr, num_batch=num_batch,
num_epoch=num_epoch)
else:
(max_val_acc, max_val_f1, max_val_sens, max_val_spec, max_val_test_acc,
max_val_test_f1, max_val_test_sens, max_val_test_spec, session_memory,
train_memory, train_time_avg) = train_SIGN.train(model, device,
x_train=x_train_aggregated,
y_train=y_train,
x_val=None,
y_val=None,
x_test=None,
y_test=None,
multilabel=multilabel,
lr=lr, num_batch=num_batch,
num_epoch=num_epoch)
elif model_type == 'GCN':
if do_evaluation:
(max_val_acc, max_val_f1, max_val_sens, max_val_spec, max_val_test_acc,
max_val_test_f1, max_val_test_sens, max_val_test_spec, session_memory,
train_memory, train_time_avg) = train_GCN.train(model, device,
x_train=x_train,
y_train=y_train,
adj_train=adj_train,
train_mask=train_mask,
x_val=x_val,
y_val=y_val,
adj_val=adj_val,
val_mask=val_mask,
x_test=x_test,
y_test=y_test,
adj_test=adj_test,
test_mask=test_mask,
multilabel=multilabel,
lr=lr, num_epoch=num_epoch)
else:
(max_val_acc, max_val_f1, max_val_sens, max_val_spec, max_val_test_acc,
max_val_test_f1, max_val_test_sens, max_val_test_spec, session_memory,
train_memory, train_time_avg) = train_GCN.train(model, device,
x_train=x_train,
y_train=y_train,
adj_train=adj_train,
train_mask=train_mask,
x_val=None,
y_val=None,
adj_val=None,
val_mask=None,
x_test=None,
y_test=None,
adj_test=None,
test_mask=None,
multilabel=multilabel,
lr=lr, num_epoch=num_epoch)
#%% Printing Result
if print_result:
writer.write_result(dataset_name, model_type, "Coarsening", x_train.shape[0],
max_val_acc, max_val_f1, max_val_sens, max_val_spec,
max_val_test_acc, max_val_test_f1, max_val_test_sens,
max_val_test_spec, session_memory, train_memory,
train_time_avg)
writer.write_label_dist_error(dataset_name, "Coarsening", x_train.shape[0],
0, Y_dist_error)