forked from HSLCY/VCWE
/
trainer.py
168 lines (145 loc) · 7.33 KB
/
trainer.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
import torch
import torch.optim as optim
from torch.utils.data import DataLoader
from tqdm import tqdm
import numpy as np
import argparse
from data_reader import DataReader, Word2vecDataset
from model import VCWEModel
from optimization import VCWEAdam
class Word2VecTrainer:
def __init__(self, input_file, vocabulary_file, img_data_file, char2ix_file, output_dir, maxwordlength, emb_dimension, line_batch_size, sample_batch_size,
neg_num, window_size, discard, epochs, initial_lr, seed):
torch.manual_seed(seed)
self.img_data = np.load(img_data_file)
self.data = DataReader(input_file, vocabulary_file, char2ix_file, maxwordlength, discard, seed)
dataset = Word2vecDataset(self.data, window_size, sample_batch_size, neg_num)
self.dataloader = DataLoader(dataset, batch_size=line_batch_size,
shuffle=True, num_workers=0, collate_fn=dataset.collate)
self.output_dir = output_dir
self.emb_size = len(self.data.word2id)
self.char_size = len(self.data.char2id)+1 #5031
self.emb_dimension = emb_dimension
self.line_batch_size = line_batch_size
self.epochs = epochs
self.initial_lr = initial_lr
self.VCWE_model = VCWEModel(self.emb_size, self.emb_dimension, self.data.wordid2charid, self.char_size)
self.use_cuda = torch.cuda.is_available()
self.device = torch.device("cuda" if self.use_cuda else "cpu")
self.num_train_steps= int(len(self.dataloader) * self.epochs)
if self.use_cuda:
self.VCWE_model.cuda()
def train(self):
self.img_data = torch.from_numpy(self.img_data).to(self.device)
no_decay = ['bias']
optimizer_parameters = [
{'params': [p for n, p in self.VCWE_model.named_parameters() if not any(nd in n for nd in no_decay)], 'weight_decay_rate': 0.01},
{'params': [p for n, p in self.VCWE_model.named_parameters() if any(nd in n for nd in no_decay)], 'weight_decay_rate': 0.0}
]
print("num_train_steps=",self.num_train_steps)
optimizer = VCWEAdam(optimizer_parameters,
lr=self.initial_lr,
warmup=0.1,
t_total=self.num_train_steps)
for epoch in range(self.epochs):
print("Epoch: " + str(epoch + 1))
running_loss = 0.0
for i, sample_batched in enumerate(tqdm(self.dataloader)):
if len(sample_batched[0]) > 1:
pos_u = sample_batched[0].to(self.device)
pos_v = sample_batched[1].to(self.device)
neg_v = sample_batched[2].to(self.device)
lengths = sample_batched[3].to(self.device)
optimizer.zero_grad()
loss = self.VCWE_model.forward(pos_u, pos_v, neg_v, self.img_data)
running_loss += loss.item()
loss.backward()
optimizer.step()
if i > 0 and i % 1000 == 0:
print('loss=', running_loss/1000)
running_loss=0.0
if (epoch+1) % 5 == 0 or (epoch+1) == self.epochs:
self.VCWE_model.save_embedding(self.data.id2word, self.output_dir+"zh_wiki_VCWE_ep"+str(epoch+1)+".txt")
def main():
parser = argparse.ArgumentParser()
parser.add_argument("--input_file",
default="./data/zh_wiki.txt",
type=str,
required=True,
help="The input file that the VCWE model was trained on.")
parser.add_argument("--vocab_file",
default="./data/vocabulary.txt",
type=str,
required=True,
help="The vocabulary file that the VCWE model was trained on.")
parser.add_argument("--img_data_file",
default="./data/char_img_sub_mean.npy",
type=str,
help="The image data file that the VCWE model was trained on.")
parser.add_argument("--char2ix_file",
default="./data/char2ix.npz",
type=str,
help="The character-to-index file corespond to the image data file.")
parser.add_argument("--output_dir",
default="./embedding/",
type=str,
help="The output directory where the embedding file will be written.")
parser.add_argument("--line_batch_size",
default=32,
type=int,
help="Batch size for lines.")
parser.add_argument("--sample_batch_size",
default=128,
type=int,
help="Batch size for samples in a line.")
parser.add_argument("--emb_dim",
default=100,
type=int,
help="Embedding dimensions.")
parser.add_argument("--maxwordlength",
default=5,
type=int,
help="The maximum number of characters in a word.")
parser.add_argument("--neg_num",
default=5,
type=int,
help="The number of negative samplings.")
parser.add_argument("--window_size",
default=5,
type=int,
help="The window size.")
parser.add_argument("--discard",
default=1e-5,
type=int,
help="The sub-sampling threshold.")
parser.add_argument("--learning_rate",
default=0.001,
type=float,
help="The initial learning rate for Adam.")
parser.add_argument("--num_train_epochs",
default=50,
type=int,
help="Total number of training epochs to perform.")
parser.add_argument('--seed',
type=int,
default=12345,
help="random seed for initialization")
args = parser.parse_args()
w2v = Word2VecTrainer(input_file = args.input_file, \
vocabulary_file = args.vocab_file, \
img_data_file = args.img_data_file, \
char2ix_file = args.char2ix_file, \
output_dir = args.output_dir,
maxwordlength = args.maxwordlength,
emb_dimension = args.emb_dim,
line_batch_size = args.line_batch_size,
sample_batch_size = args.sample_batch_size,
neg_num = args.neg_num,
window_size = args.window_size,
discard = args.discard,
epochs = args.num_train_epochs,
initial_lr = args.learning_rate,
seed = args.seed)
w2v.train()
if __name__ == "__main__":
main()