-
Notifications
You must be signed in to change notification settings - Fork 0
/
RCOV_synthetic_data.py
293 lines (245 loc) · 15 KB
/
RCOV_synthetic_data.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
280
281
282
283
284
285
286
287
import torch
import torch_geometric as pyg
from models import RCOVmodel
from networkx.generators.random_graphs import erdos_renyi_graph
import wandb
import argparse
import utils
import numpy as np
import train_utils
from utils import EarlyStopping
import random
if __name__ == "__main__":
global args
parser = argparse.ArgumentParser()
parser.add_argument('--num_nodes', dest='num_nodes', type=int, default=20)
parser.add_argument('--num_samples', dest='num_samples', type=int, default=1000)
parser.add_argument('--lr', dest='lr', type=float, default=0.0001)
parser.add_argument('--d', dest='d', type=int, default=16)
parser.add_argument('--hidden_channels', dest='hidden_channels', type=int, default=64)
parser.add_argument('--n_layers', dest='n_layers', type=int, default=1)
parser.add_argument('--num_test_samples', dest='num_test_samples', type=int, default=100)
parser.add_argument('--train_batch_size', dest='train_batch_size', type=int, default=8)
parser.add_argument('--num_epochs', dest='num_epochs', type=int, default=1000)
parser.add_argument('--num_train_samples', dest='num_train_samples', type=int, default=1000)
parser.add_argument('--wandb', dest='wandb', type=int, default=1)
parser.add_argument('--patience', dest='patience', type=int, default=100)
parser.add_argument('--task_type', dest='task_type', type=str, default='Node Sum')
parser.add_argument('--RCOV_threshold', dest='RCOV_threshold', type=float, default=0.15)
parser.add_argument('--wandb', dest='wandb', action='store_true')
args = parser.parse_args()
num_nodes = args.num_nodes
num_samples = args.num_samples
num_test_samples = args.num_test_samples
eps = args.eps
train_p = args.train_p
output_dim = 2
d = args.d
n_layers = args.n_layers
train_batch_size = args.train_batch_size
hidden_channels = args.hidden_channels
model_type = RCOVmodel
wd = 0.0001
RCOV_threshold = args.RCOV_threshold
n_conv_layers = args.n_layers
loss = torch.nn.CrossEntropyLoss
num_epochs = args.num_epochs
lr = args.lr
optimizer_type = torch.optim.Adam
loss_thresh = 0.0001
n_conv_layers = args.n_layers
num_epochs = args.num_epochs
num_classes = 2
margin = 0.01
teacher = utils.sample_teacher(d, self=True, top=False)
train_dist_type = erdos_renyi_graph
train_dist = lambda num_nodes: train_dist_type(n=num_nodes, p=train_p)
test_acc_original = []
test_acc_RCOV = []
test_acc_empty = []
np.random.seed(0)
seeds = np.random.randint(low=0, high=10000, size=10)
for i in range(10):
seed = seeds[i]
np.random.seed(seed)
torch.manual_seed(seed)
random.seed(seed)
train_original_pyg_graphs, train_original_nx_graphs = utils.get_dif_graphs_dataset(num_samples=num_samples,
num_nodes=num_nodes, num_features=d,
teacher=teacher, graph_dist=train_dist,
margin=margin,
self_teacher=True,
with_edge_weight=True)
utils.add_constant_feature_1(train_original_pyg_graphs)
train_original_mean, _, _, _ = utils.get_moments(train_original_nx_graphs)
if args.task_type== 'Edges':
utils.set_label_to_num_of_edges(train_original_pyg_graphs, threshold=train_original_mean * num_nodes)
val_original_pyg_graphs, _ = utils.get_dif_graphs_dataset(num_samples=num_samples,
num_nodes=num_nodes, num_features=d,
teacher=teacher, graph_dist=train_dist,
margin=margin,
self_teacher=True,
with_edge_weight=True)
utils.add_constant_feature_1(val_original_pyg_graphs)
if args.task_type== 'Edges':
utils.set_label_to_num_of_edges(val_original_pyg_graphs, threshold=train_original_mean * num_nodes)
test_original_pyg_graphs, _ = utils.get_dif_graphs_dataset(num_samples=num_samples,
num_nodes=num_nodes, num_features=d,
teacher=teacher, graph_dist=train_dist,
margin=margin,
self_teacher=True,
with_edge_weight=True)
utils.add_constant_feature_1(test_original_pyg_graphs)
if args.task_type== 'Edges':
utils.set_label_to_num_of_edges(test_original_pyg_graphs, threshold=train_original_mean * num_nodes)
train_RCOV_pyg_graphs, train_RCOV_nx_graphs = utils.batch_adges_reduce_ratio_add_random_edges_to_pyg_graphs(
pyg_graphs=train_original_pyg_graphs, reduce_goal=RCOV_threshold)
train_RCOV_mean, train_RCOV_std, train_RCOV_skew, train_RCOV_kurt = utils.get_moments(train_RCOV_nx_graphs)
train_RCOV_std_mean_ratio = train_RCOV_std / train_RCOV_mean
print('train_RCOV_std_mean_ratio: ', train_RCOV_std_mean_ratio)
val_RCOV_pyg_graphs, val_RCOV_nx_graphs = utils.batch_adges_reduce_ratio_add_random_edges_to_pyg_graphs(
pyg_graphs=val_original_pyg_graphs, reduce_goal=RCOV_threshold)
val_RCOV_mean, val_RCOV_std, val_RCOV_skew, val_RCOV_kurt = utils.get_moments(val_RCOV_nx_graphs)
val_RCOV_std_mean_ratio = val_RCOV_std / val_RCOV_mean
print('val_RCOV_std_mean_ratio: ', val_RCOV_std_mean_ratio)
test_RCOV_pyg_graphs, test_RCOV_nx_graphs = utils.batch_adges_reduce_ratio_add_random_edges_to_pyg_graphs(
pyg_graphs=test_original_pyg_graphs, reduce_goal=RCOV_threshold)
test_RCOV_mean, test_RCOV_std, test_RCOV_skew, test_RCOV_kurt = utils.get_moments(test_RCOV_nx_graphs)
test_RCOV_std_mean_ratio = test_RCOV_std / test_RCOV_mean
print('test_RCOV_std_mean_ratio: ', test_RCOV_std_mean_ratio)
n_features = train_original_pyg_graphs[0].x.shape[1]
train_original_loader = pyg.loader.DataLoader(train_original_pyg_graphs, batch_size=train_batch_size)
train_RCOV_loader = pyg.loader.DataLoader(train_original_pyg_graphs, batch_size=train_batch_size)
val_original_loader = pyg.loader.DataLoader(val_original_pyg_graphs, batch_size=num_test_samples)
val_RCOV_loader = pyg.loader.DataLoader(val_RCOV_pyg_graphs, batch_size=num_test_samples)
if torch.cuda.is_available():
device = torch.device("cuda")
else:
device = torch.device("cpu")
config = {
'data_dir_number': args.data_dir_number,
'd': d,
'num_nodes': num_nodes,
'eps': eps,
'lr': lr,
'loss': loss.__name__,
'train_batch_size': train_batch_size,
'hidden_channels': hidden_channels,
'n_conv_layers': n_conv_layers,
'num_epochs': num_epochs,
'optimizer': optimizer_type.__name__,
'train_dist': train_dist_type.__name__,
'model': model_type.__name__,
'file': 'adding_edges_to_real_data.py',
'num_final_feauturs': n_features,
'device': device.type,
'loss_thresh': loss_thresh,
'wd': wd,
'iter': i,
'task_type' : args.task_type,
}
if args.wandb:
for name, val in config.items():
print(f'{name}: {val}')
run = wandb.init(project='your_project', reinit=True, entity='you_entity',
settings=wandb.Settings(start_method='fork'),
config=config)
# train on empty graphs
model_empty = model_type(in_channels=n_features, out_channels=num_classes, num_layers=n_layers,
hidden_channels=hidden_channels)
optimizer_empty = optimizer_type(params=model_empty.parameters(), lr=lr, weight_decay=wd)
loss_fn_empty = loss()
model_empty.to(device)
early_stop = EarlyStopping(patience=args.patience)
empty_graphs_train = utils.get_empty_graphs_from_datasets(train_original_pyg_graphs)
empty_graphs_val = utils.get_empty_graphs_from_datasets(val_original_pyg_graphs)
empty_train_loader = pyg.loader.DataLoader(empty_graphs_train, batch_size=train_batch_size, shuffle=True)
empty_val_loader = pyg.loader.DataLoader(empty_graphs_val, batch_size=train_batch_size, shuffle=True)
for epoch in range(num_epochs):
train_empty_loss, train_empty_acc = train_utils.train_epoch(model_empty, dloader=empty_train_loader,
loss_fn=loss_fn_empty,
optimizer=optimizer_empty,
classify=True, device=device)
val_empty_loss, val_empty_acc = train_utils.test_epoch(model_empty, dloader=empty_val_loader, loss_fn=loss_fn_empty,
classify=True,
device=device)
early_stop(val_empty_loss)
if early_stop.early_stop:
print(f'early stop at epoch: {epoch}')
break
if args.wandb:
wandb.log({'train_empty_loss': train_empty_loss,
'train_empty_acc': train_empty_acc,
'val_empty_loss': val_empty_loss,
'val_empty_acc': val_empty_acc})
model_original = model_type(in_channels=n_features, out_channels=num_classes, num_layers=n_layers,
hidden_channels=hidden_channels)
optimizer_original = optimizer_type(params=model_original.parameters(), lr=lr, weight_decay=wd)
loss_fn_original = loss()
model_original.to(device)
early_stop_original = EarlyStopping(patience=args.patience)
for epoch in range(num_epochs):
train_original_loss, train_original_acc = train_utils.train_epoch(model_original, dloader=train_original_loader, loss_fn=loss_fn_original,
optimizer=optimizer_original,
classify=True, device=device)
val_original_loss, val_original_acc = train_utils.test_epoch(model_original, dloader=val_original_loader,
loss_fn=loss_fn_original, classify=True, device=device)
early_stop_original(val_original_loss)
if early_stop_original.early_stop:
print(f'early stop at epoch: {epoch}')
break
if args.wandb:
wandb.log({'train_original_loss': train_original_loss,
'train_original_acc': train_original_acc,
'val_original_loss': val_original_loss,
'val_original_acc': val_original_acc})
model_RCOV = model_type(in_channels=n_features, out_channels=num_classes, num_layers=n_layers,
hidden_channels=hidden_channels)
optimizer_RCOV = optimizer_type(params=model_RCOV.parameters(), lr=lr, weight_decay=wd)
loss_fn_RCOV = loss()
model_RCOV.to(device)
early_stop = EarlyStopping(patience=args.patience)
for epoch in range(num_epochs):
train_RCOV_loss, train_RCOV_acc = train_utils.train_epoch(model_RCOV, dloader=train_RCOV_loader, loss_fn=loss_fn_RCOV,
optimizer=optimizer_RCOV,
classify=True, device=device)
val_RCOV_loss, val_RCOV_acc = train_utils.test_epoch(model_RCOV, dloader=val_RCOV_loader, loss_fn=loss_fn_RCOV,
classify=True,
device=device)
early_stop(val_RCOV_loss)
if early_stop_original.early_stop:
print(f'early stop at epoch: {epoch}')
break
if args.wandb:
wandb.log({'train_RCOV_loss': train_RCOV_loss,
'train_RCOV_acc': train_RCOV_acc,
'val_RCOV_loss': val_RCOV_loss,
'val_RCOV_acc': val_RCOV_acc})
#test
test_original_loader = pyg.loader.DataLoader(test_original_pyg_graphs, batch_size=num_test_samples, shuffle=True)
test_RCOV_loader = pyg.loader.DataLoader(test_RCOV_pyg_graphs, batch_size=num_test_samples, shuffle=True)
empty_graphs_test = utils.get_empty_graphs_from_datasets(test_original_pyg_graphs)
empty_test_loader = pyg.loader.DataLoader(empty_graphs_test, batch_size=num_test_samples, shuffle=True)
test_original_loss, test_original_acc = train_utils.test_epoch(model_original, dloader=test_original_loader
, loss_fn=loss_fn_original, classify=True, device=device)
test_RCOV_loss, test_RCOV_acc = train_utils.test_epoch(model_RCOV, dloader=test_RCOV_loader
, loss_fn=loss_fn_RCOV, classify=True, device=device)
test_empty_loss, test_empty_acc = train_utils.test_epoch(model_empty, dloader=empty_test_loader
, loss_fn=loss_fn_empty, classify=True, device=device)
test_acc_original.append(test_original_acc)
test_acc_RCOV.append(test_RCOV_acc)
test_acc_empty.append(test_empty_acc)
if args.wandb:
wandb.log({'test_original_loss': test_original_loss,
'test_original_acc': test_original_acc,
'test_RCOV_loss': test_RCOV_loss,
'test_RCOV_acc': test_RCOV_acc,
'test_empty_loss': test_empty_loss,
'test_empty_acc': test_empty_acc})
if args.wandb:
wandb.log({'test_acc_original': np.mean(test_acc_original),
'test_acc_RCOV': np.mean(test_acc_RCOV),
'test_acc_empty': np.mean(test_acc_empty),
'test_acc_original_std': np.std(test_acc_original),
'test_acc_RCOV_std': np.std(test_acc_RCOV),
'test_acc_empty_std': np.std(test_acc_empty)})