-
Notifications
You must be signed in to change notification settings - Fork 4
/
Copy pathMainResponseWAE4Early.py
executable file
·127 lines (118 loc) · 5.57 KB
/
MainResponseWAE4Early.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
# 相比MainPathVoting的不同之处：random使用全部路径文本；wae使用response path，即除去了源文本
# 二者使用不同的id编码，即每一条路径对应random path ids和wae response path ids
from models.PathBased import ResponseWAE, ResponseCatWAE, ResponseWAECat
from torch import optim
import datetime
from evaluate import *
from get_args import _args, print_args
from data_io import *
import sys
from torch.optim.lr_scheduler import StepLR
def main():
print_args(_args)
# 固定随机数种子
setup_seed(_args.seed)
print("Step1:processing data")
# x_train_random, y_train_random, x_test_random, y_test_random = \
# load_path_data(train_path, test_path, label_path,
# "/data1/text",
# path_random_npz)
#
x_train_random, y_train_random, x_test_random, y_test_random = \
load_path_data(train_path, test_path, label_path,
path_random_id2paths_dict_path,
path_random_npz)
x_train_response, y_train_response, x_test_response, y_test_response = \
load_path_data(train_path, test_path, label_path,
response_id2paths_dict_path,
response_wae_npz)
# early detection代码，待测试
# x_train_random, y_train_random, x_test_random, y_test_random = \
# load_path_data_for_early_detection(train_path, test_path, label_path,
# early_path_random_id2paths_dict_path,
# path_node_ids_dict_path,
# random_tweet2token_ids_dict_path,
# False,
# False)
#
# x_train_response, y_train_response, x_test_response, y_test_response = \
# load_path_data_for_early_detection(train_path, test_path, label_path,
# early_response_id2paths_dict_path,
# path_node_ids_dict_path,
# response_tweet2token_ids_dict_path,
# True,
# False)
print('Step2:build model')
# model = PathVoting(_args.vocab_dim, _args.random_dim, _args.vae_dim, _args.class_num)
# model = ResponseWAE(_args.random_vocab_dim, _args.response_vocab_dim,
# wae_best_encoder_path, _args.random_dim,
# _args.vae_dim, _args.class_num)
model = ResponseWAE(_args.random_vocab_dim, _args.response_vocab_dim,
wae_best_encoder_path, _args.random_dim,
_args.vae_dim, _args.class_num)
model.to(device)
# 3. looping SGD
print('Step3:start training')
if _args.optim == 'adagrad':
optimizer = optim.Adagrad(model.parameters(), lr=_args.lr)
elif _args.optim == 'adam':
optimizer = optim.Adam(model.parameters(), lr=_args.lr)
else:
print('optim %s not correct' % _args.optim)
return
# scheduler = StepLR(optimizer, 10, 0.5, -1)
losses_5, losses = [], []
num_examples_seen = 0
indexs = list(range(len(y_train_random)))
highest_acc = 0
best_result = []
for epoch in range(1, _args.epoch + 1):
# print('epoch：', epoch, '学习率：', scheduler.get_lr())
# 每次训练打乱顺序
random.shuffle(indexs)
# 训练模型
for cnt, i in enumerate(indexs):
pred_y, loss = model.forward(torch.Tensor(x_train_random[i]).cuda(device).long(),
torch.Tensor(x_train_response[i]).cuda(device).long(),
torch.Tensor(y_train_random[i]).cuda(device),
)
optimizer.zero_grad()
loss.backward()
optimizer.step()
losses.append(loss.data.cpu().tolist())
num_examples_seen += 1
# if (cnt + 1) % 500 == 0:
# break
# scheduler.step()
# cal loss & evaluate
with torch.no_grad():
if epoch % 1 == 0:
losses_5.append((num_examples_seen, np.mean(losses)))
time = datetime.datetime.now().strftime('%Y-%m-%d %H:%M:%S')
if _args.verbose:
print("%s: Loss after num_examples_seen=%d epoch=%d: %f" %
(time, num_examples_seen, epoch, np.mean(losses)))
sys.stdout.flush()
prediction = []
# 因为每棵树都不同，所以测试的训练和测试的batch都为1；后续有待改进
for j in range(len(y_test_random)):
prediction.append(
model.predict_up(torch.Tensor(x_test_random[j]).cuda(device).long(),
torch.Tensor(x_test_response[j]).cuda(device).long())
.cpu().data.numpy().tolist())
res = evaluation_4class(prediction, y_test_random)
# highest_acc = max(highest_acc, res[1])
if res[1] > highest_acc:
best_result = res
highest_acc = res[1]
if _args.verbose:
print(res)
print()
sys.stdout.flush()
sys.stdout.flush()
losses = []
print('最高acc：', highest_acc)
print("最优性能：", best_result)
print('#' * 80)
if __name__ == '__main__':
main()