-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathexport.py
148 lines (144 loc) · 6.22 KB
/
export.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
import paddle
import os
import time
import pickle
import argparse
from model import TiSASRec
from tqdm import tqdm
from utils import *
def str2bool(s):
if s not in {'false', 'true'}:
raise ValueError('Not a valid boolean string')
return s == 'true'
parser = argparse.ArgumentParser()
parser.add_argument('--dataset', required=True)
parser.add_argument('--train_dir', required=True)
parser.add_argument('--batch_size', default=128, type=int)
parser.add_argument('--lr', default=0.001, type=float)
parser.add_argument('--maxlen', default=50, type=int)
parser.add_argument('--hidden_units', default=50, type=int)
parser.add_argument('--num_blocks', default=2, type=int)
parser.add_argument('--num_epochs', default=201, type=int)
parser.add_argument('--num_heads', default=1, type=int)
parser.add_argument('--dropout_rate', default=0.2, type=float)
parser.add_argument('--l2_emb', default=5e-05, type=float)
parser.add_argument('--device', default='cpu', type=str)
parser.add_argument('--inference_only', default=False, type=str2bool)
parser.add_argument('--state_dict_path', default=None, type=str)
parser.add_argument('--time_span', default=256, type=int)
args = parser.parse_args()
if not os.path.isdir(args.dataset + '_' + args.train_dir):
os.makedirs(args.dataset + '_' + args.train_dir)
with open(os.path.join(args.dataset + '_' + args.train_dir, '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()
dataset = data_partition(args.dataset)
[user_train, user_valid, user_test, usernum, itemnum, timenum] = 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)))
f = open(os.path.join(args.dataset + '_' + args.train_dir, 'log.txt'), 'w')
try:
relation_matrix = pickle.load(open(
'data/relation_matrix_%s_%d_%d.pickle' % (args.dataset, args.maxlen,
args.time_span), 'rb'))
except:
relation_matrix = Relation(user_train, usernum, args.maxlen, args.time_span
)
pickle.dump(relation_matrix, open(
'data/relation_matrix_%s_%d_%d.pickle' % (args.dataset, args.maxlen,
args.time_span), 'wb'))
sampler = WarpSampler(user_train, usernum, itemnum, relation_matrix,
batch_size=args.batch_size, maxlen=args.maxlen, n_workers=3)
model = TiSASRec(usernum, itemnum, itemnum, args).to(args.device)
for name, param in model.named_parameters():
try:
init_XavierUniform = paddle.nn.initializer.XavierUniform()
init_XavierUniform(param.data)
except:
pass
model.train()
epoch_start_idx = 1
if args.state_dict_path is not None:
try:
model.set_state_dict(state_dict=paddle.load(path=args.state_dict_path))
tail = args.state_dict_path[args.state_dict_path.find('epoch=') + 6:]
epoch_start_idx = int(tail[:tail.find('.')]) + 1
except:
print('failed loading state_dicts, pls check file path: ', end='')
print(args.state_dict_path)
if args.inference_only:
model.eval()
t_test = evaluate(model, dataset, args)
print('test (NDCG@10: %.4f, HR@10: %.4f)' % (t_test[0], t_test[1]))
bce_criterion = paddle.nn.BCEWithLogitsLoss()
adam_optimizer = paddle.optimizer.Adam(parameters=model.parameters(),
learning_rate=args.lr, beta1=(0.9, 0.98)[0], beta2=(0.9, 0.98)[1],
weight_decay=0.0)
T = 0.0
t0 = time.time()
for epoch in range(epoch_start_idx, args.num_epochs + 1):
if args.inference_only:
break
for step in range(num_batch):
u, seq, time_seq, time_matrix, pos, neg = sampler.next_batch()
u, seq, pos, neg = np.array(u), np.array(seq), np.array(pos), np.array(
neg)
time_seq, time_matrix = np.array(time_seq), np.array(time_matrix)
pos_logits, neg_logits = model(seq, time_matrix, pos, neg)
try:
input_spec = list(paddle.static.InputSpec.from_tensor(paddle.to_tensor(t)) for t in (seq, time_matrix, pos, neg))
paddle.jit.save(model, input_spec=input_spec, path="./model")
print('[JIT] paddle.jit.save successed.')
exit(0)
except Exception as e:
print('[JIT] paddle.jit.save failed.')
raise e
pos_labels, neg_labels = paddle.ones(shape=pos_logits.shape
), paddle.zeros(shape=neg_logits.shape)
adam_optimizer.clear_grad()
indices = np.where(pos != 0)
loss = bce_criterion(pos_logits[indices], pos_labels[indices])
loss += bce_criterion(neg_logits[indices], neg_labels[indices])
for param in model.item_emb.parameters():
loss += args.l2_emb * paddle.linalg.norm(x=param)
for param in model.abs_pos_K_emb.parameters():
loss += args.l2_emb * paddle.linalg.norm(x=param)
for param in model.abs_pos_V_emb.parameters():
loss += args.l2_emb * paddle.linalg.norm(x=param)
for param in model.time_matrix_K_emb.parameters():
loss += args.l2_emb * paddle.linalg.norm(x=param)
for param in model.time_matrix_V_emb.parameters():
loss += args.l2_emb * paddle.linalg.norm(x=param)
loss.backward()
adam_optimizer.step()
print('loss in epoch {} iteration {}: {}'.format(epoch, step, loss.
item()))
if epoch % 20 == 0:
model.eval()
t1 = time.time() - t0
T += t1
print('Evaluating', end='')
t_test = evaluate(model, dataset, args)
t_valid = evaluate_valid(model, dataset, args)
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]))
f.write(str(t_valid) + ' ' + str(t_test) + '\n')
f.flush()
t0 = time.time()
model.train()
if epoch == args.num_epochs:
folder = args.dataset + '_' + args.train_dir
fname = (
'TiSASRec.epoch={}.lr={}.layer={}.head={}.hidden={}.maxlen={}.pth')
fname = fname.format(args.num_epochs, args.lr, args.num_blocks,
args.num_heads, args.hidden_units, args.maxlen)
paddle.save(obj=model.state_dict(), path=os.path.join(folder, fname))
f.close()
sampler.close()
print('Done')