-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathtrainner_nc.py
141 lines (123 loc) · 6.09 KB
/
trainner_nc.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
import time
import torch
import numpy as np
from tqdm import tqdm
import random
def setup_seed(seed):
torch.manual_seed(seed)
torch.cuda.manual_seed_all(seed)
np.random.seed(seed)
random.seed(seed)
torch.backends.cudnn.deterministic = True
class NCTrainer(object):
def __init__(self, args, model, data, writer=None, **kwargs):
self.args = args
self.data = data
self.model = model
self.writer = writer
self.len = len(data['edge_index'])
self.len_train = self.len - args.testlength - args.vallength
self.len_val = args.vallength
self.len_test = args.testlength
x = data['x'].to(args.device)
self.x = [x for _ in range(self.len)] if len(x.shape) <= 2 else x
setup_seed(args.seed)
print('total length: {}, test length: {}'.format(
self.len, args.testlength))
def run(self):
args = self.args
max_acc = 0
max_test_acc = 0
max_train_acc = 0
min_epoch = args.min_epoch
max_patience = args.patience
self.optimizer = torch.optim.Adam(self.model.parameters(), lr=args.lr, weight_decay=args.weight_decay)
t_total0 = time.time()
with tqdm(range(1, args.max_epoch + 1)) as bar:
for epoch in bar:
t0 = time.time()
average_epoch_loss, average_train_acc, average_val_acc, average_test_acc, test_acc_list = self.train(epoch, self.data)
# update the best results.
if average_val_acc > max_acc:
max_acc = average_val_acc
max_test_acc = average_test_acc
max_train_acc = average_train_acc
best_test_acc_list = test_acc_list
patience = 0
best_epoch = epoch
else:
patience += 1
if epoch > min_epoch and patience > max_patience:
break
if epoch == 1 or epoch % self.args.log_interval == 0:
print(
"Epoch:{}, Loss: {:.4f}, Time: {:.3f}".format(epoch, average_epoch_loss, time.time() - t0))
print(
f"Current: Epoch:{epoch}, Train AUC :{average_train_acc:.4f}, Val AUC: {average_val_acc:.4f}, Test AUC: {average_test_acc:.4f}"
)
print(
f"Best_metric: Epoch:{best_epoch}, Train AUC:{max_train_acc:.4f}, Val AUC: {max_acc:.4f}, Test AUC: {max_test_acc:.4f}"
)
print(
f"Every Test: Epoch:{best_epoch}, Test15:{best_test_acc_list[0]:.4f}, Test16: {best_test_acc_list[1]:.4f}, Test17: {best_test_acc_list[2]:.4f}")
epoch_time = (time.time() - t_total0) / (epoch - 1)
return epoch, average_train_acc, average_val_acc, average_test_acc, best_epoch, max_train_acc, max_acc, max_test_acc, epoch_time, best_test_acc_list
def train(self, epoch, data):
self.model.train()
optimizer = self.optimizer
causal_embedding_list, conf_embedding_list, env_kl_loss = \
self.model(data['edge_index'], self.x[:self.len_train], self.len_train)
conf_embedding_list = [emb.detach() for emb in conf_embedding_list]
criterion = torch.nn.CrossEntropyLoss()
causal_loss = torch.tensor([]).to(self.args.device)
env_loss = torch.tensor([]).to(self.args.device)
for t in range(self.len_train):
causal_embedding = causal_embedding_list[:, t, :].squeeze() #[N, F]
causal_pred = self.cal_pred(causal_embedding, self.model.cs_decoder, data['node_masks'][t].to(self.args.device))
causal_loss = torch.cat([causal_loss, criterion(causal_pred, data['y'][data['node_masks'][t]].squeeze().to(self.args.device)).unsqueeze(0)])
# causal intervention
for times in range(self.args.intervention_times):
n_s = np.random.randint(self.len_train)
conf_z = conf_embedding_list[n_s]
conf_pred = self.cal_pred(conf_z, self.model.ss_decoder, data['node_masks'][t].to(self.args.device))
s1 = np.random.randint(len(conf_pred))
conf_pred_s = torch.sigmoid(conf_pred[s1]).detach()
conf = conf_pred_s * causal_pred
env_loss = torch.cat([env_loss, criterion(conf, data['y'][data['node_masks'][t]].squeeze().to(self.args.device)).unsqueeze(0)])
env_mean = env_loss.mean()
env_var = torch.var(env_loss * self.args.intervention_times *self.len_train)
penalty = env_mean + env_var
la = self.args.weight1
if epoch < self.args.warm_epoch:
la = 0
loss = torch.mean(causal_loss) + la * penalty + self.args.weight2 *torch.mean(env_kl_loss)
optimizer.zero_grad()
loss.backward()
optimizer.step()
average_epoch_loss = loss.item()
# get acc
self.model.eval()
train_acc_list, val_acc_list, test_acc_list = [], [], []
causal_embedding_list, _, _ = self.model(data['edge_index'], self.x, self.len)
for t in range(self.len):
z = causal_embedding_list[:, t, :].squeeze() #[N, F]
acc = self.predict(z, self.model.cs_decoder, data['node_masks'][t], data['y'])
if t < self.len_train:
train_acc_list.append(acc)
elif t < self.len_train + self.len_val:
val_acc_list.append(acc)
else:
test_acc_list.append(acc)
return average_epoch_loss, np.mean(train_acc_list), np.mean(val_acc_list), np.mean(test_acc_list), test_acc_list
def cal_pred(self, z, decoder, node_masks):
pred = decoder(z)[node_masks]
return pred
def predict(self, z, decoder, node_mask, y):
pred = decoder(z)[node_mask]
pred = pred.argmax(dim=-1).squeeze()
y = y[node_mask].squeeze()
y, pred = y.detach().cpu().numpy(), pred.detach().cpu().numpy()
acc = (pred == y).sum().item() / y.shape[0]
acc = float(acc)
# acc = accuracy_score(y, pred)
return acc