-
Notifications
You must be signed in to change notification settings - Fork 1
/
Copy pathASSR_main.py
151 lines (131 loc) · 6.75 KB
/
ASSR_main.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
import os
import time
import argparse
from ASSR.utils import *
from model.ASSR.ASSR_model import ASSR
import torch
def str2bool(s):
if s not in {'false', 'true'}:
raise ValueError('Not a valid boolean string')
return s == 'true'
if __name__ == '__main__':
parser = argparse.ArgumentParser()
parser.add_argument('--result', default='../../experiments/ASSR/')
parser.add_argument('--dataset', default='bili_newdata', type=str)
parser.add_argument('--lr', default=0.001, type=float)
parser.add_argument('--maxlen', default=50, type=int)
parser.add_argument('--hidden_units', default=20, type=int)
parser.add_argument('--num_epochs', default=500, type=int)
parser.add_argument('--batch_size', default=128, type=int)
parser.add_argument('--num_blocks', default=2, type=int)
parser.add_argument('--num_heads', default=5, type=int)
parser.add_argument('--dropout_rate', default=0.2, type=float)
parser.add_argument('--device', default='cuda', type=str)
parser.add_argument('--l2_emb', default=0.0, type=float)
parser.add_argument('--train_dir', default='default', type=str)
parser.add_argument('--inference_only', default=False, type=str2bool)
parser.add_argument('--state_dict_path', default=None, type=str)
args = parser.parse_args()
file_name = 'ASSR' + '.dataset={}.lr={}.maxlen={}.hidden={}.epoch={}.batch={}.layer={}.head={}.dropout={}'
file_name = file_name.format(args.dataset, args.lr,args.maxlen, args.hidden_units, args.num_epochs,args.batch_size,
args.num_blocks, args.num_heads, args.dropout_rate)
if not os.path.isdir(args.result + '_' + args.train_dir):
os.makedirs(args.result + '_' + args.train_dir)
with open(os.path.join(args.result + '_' + args.train_dir, file_name+'_args.txt'), 'w') as f:
f.write('\n'.join([str(k) + ',' + str(v) for k, v in sorted(vars(args).items(), key=lambda x: x[0])]))
f.close()
# 完成数据集切分
dataset = ASSR_data_partition(args.dataset)
[user_train, user_valid, user_test,
con_train, con_valid, con_test,
pos_train, pos_valid, pos_test,
neg_train, neg_valid, neg_test,
time_train, time_valid, time_test,
usernum, itemnum] = dataset
num_batch = len(user_train) // args.batch_size
cc = 0.0
for u in user_train:
cc += len(user_train[u])
print('average sequence length: %.2f' % (cc / len(user_train)))
save_path = os.path.join(args.result, file_name + '_log.txt')
f = open(save_path, 'w')
seed = np.random.randint(2e9)
sampler = WarpSampler([user_train, con_train, pos_train, neg_train, time_train],
usernum, itemnum, batch_size=args.batch_size, maxlen=args.maxlen, n_workers=1, seed=seed)
# 调用模型
model = ASSR(usernum, itemnum, args).to(args.device)
for name, param in model.named_parameters(): # 参数初始化
try:
torch.nn.init.xavier_uniform_(param.data) # 使得初始参数服从均匀分布
except:
pass
model.train() # enable model training
epoch_start_idx = 1
bce_criterion = torch.nn.BCEWithLogitsLoss() # torch.nn.BCELoss()：二进制交叉熵损失函数
adam_optimizer = torch.optim.Adam(model.parameters(), lr=args.lr, betas=(0.9, 0.98)) # 亚当优化器
T = 0.0
t0 = time.time()
a = []
# 存放训练结果
best_vaile = {0: [0, 0, 0, 0]}
best_test = {0: [0, 0, 0, 0]}
all_data = []
# 进入迭代训练
for epoch in range(epoch_start_idx, args.num_epochs + 1):
for step in range(num_batch):
# 获取训练序列
u, seq, pos, neg, con_seq, pos_seq, neg_seq, time_seq = sampler.next_batch()
u, seq, pos, neg, con_seq, pos_seq, neg_seq, time_seq =\
np.array(u), np.array(seq), np.array(pos), np.array(neg), np.array(con_seq), np.array(pos_seq), np.array(neg_seq), np.array(time_seq)
pos_logits, neg_logits = model(u, seq, con_seq, pos_seq, neg_seq, time_seq, pos, neg)
pos_labels, neg_labels = torch.ones(pos_logits.shape, device=args.device), torch.zeros(neg_logits.shape,
device=args.device)
adam_optimizer.zero_grad() # 先将梯度归零
indices = np.where(pos != 0) # 得到正样不为0的索引
loss = bce_criterion(pos_logits[indices], pos_labels[indices]) # 正loss
loss += bce_criterion(neg_logits[indices], neg_labels[indices]) # 加上负采样loss
for param in model.item_emb.parameters():
loss += args.l2_emb * torch.norm(param)
loss.backward()
adam_optimizer.step()
print("loss in epoch {} iteration {}: {}".format(epoch, step,
loss.item())) # expected 0.4~0.6 after init few epochs
if epoch % 20 == 0:
data_list = []
model.eval()
t1 = time.time() - t0
T += t1
print('Evaluating', end='')
t_test = ASSR_evaluate_test(model, dataset, args) # 评估测试集
t_valid = ASSR_evaluate_valid(model, dataset, args) # 评估验证集
print('epoch:%d, time: %f(s), valid (NDCG@10: %.4f, HR@10: %.4f), test (NDCG@10: %.4f, HR@10: %.4f)'
% (epoch, T, t_valid[0], t_valid[1], t_test[0], t_test[1]))
if t_valid[1] > best_vaile[0][1]:
best_vaile[0] = t_valid
if t_test[1] > best_test[0][1]:
best_test[0] = t_test
# NDCG10\HIT10\NDCG20\HIT20
data_list.append(t_test[0])
data_list.append(t_test[1])
data_list.append(t_test[2])
data_list.append(t_test[3])
all_data.append(data_list)
# 写入文件
# valid:NDCG10,HR10;test:NDCG10,HR10
# f.write(t_test + '\n')
f.write(str(t_test[0])+',')
f.write(str(t_test[1])+',')
f.write(str(t_test[2])+',')
f.write(str(t_test[3])+'\n')
f.flush()
t0 = time.time()
model.train()
# 训练完成
if epoch == args.num_epochs:
folder = args.result + '_' + args.train_dir
torch.save(model.state_dict(), os.path.join(folder, file_name + '.pth'))
f.close()
sampler.close()
print('best valid:', best_vaile[0])
print('best test:', best_test[0])
print("Done")