-
Notifications
You must be signed in to change notification settings - Fork 3
/
train_eatn.py
138 lines (104 loc) · 7.1 KB
/
train_eatn.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
import argparse
import logging
import pandas as pd
import numpy as np
import torch
import torch.nn as nn
import torch.optim as optim
from torch.autograd import Variable
from torch.nn import functional as F
import sys
sys.path.append('../')
from pre_models import TextNormalization, MultiLabel_Vectorizer, DocProcessor
from models import EATN
from pre_models import BertMultiTrainer, BertPredictor
from pre_models import CrossAspectProcessor,get_cross_aspect_batch,accuracy
from pre_models import BertCrossFeatures
from pytorch_pretrained_bert import BertTokenizer
def fill_token(sent, term):
#print(sent, term)
return sent.replace('$T$', term)
def main():
parser = argparse.ArgumentParser()
#parser.add_argument('--model_name', default='bert_aspect', type=str)
# aspect-term dataset
prefix_train = '../data_after_process/aspect_term/2014_laptop/'
# prefix_train = '../data_after_process/aspect_term/2014_resturant/'
# prefix_train = '../data_after_process/aspect_term/2014_twitter/'
prefix_test = '../data_after_process/aspect_term/2014_resturant/'
# prefix_train = '../data_after_process/aspect_term/2014_laptop/'
# prefix_train = '../data_after_process/aspect_term/2014_twitter/'
# aspect-category dataset
# prefix_train = '../data_after_process/aspect_term/restaurant/'
# prefix_train = '../data_after_process/aspect_term/hotel/'
# prefix_train = '../data_after_process/aspect_term/beautyspa/'
# prefix_train = '../data_after_process/aspect_category/restaurant/'
# prefix_train = '../data_after_process/aspect_category/hotel/'
# prefix_train = '../data_after_process/aspect_category/beautyspa/'
source_train = prefix_train + 'train.csv'
target_train = prefix_test + 'train.csv'
source_dev = prefix_train + 'test.csv'
target_dev = prefix_test + 'test.csv'
source_train_df = pd.read_csv(source_train)
target_train_df = pd.read_csv(target_train)
source_train_df['sent'] = source_train_df.apply(lambda x: fill_token(x[1], x[2]), axis=1)
target_train_df['sent'] = target_train_df.apply(lambda x: fill_token(x[1], x[2]), axis=1)
logging.info('source_train_df shape {}, target_train_df shape {}'.format(source_train_df.shape, target_train_df.shape))
source_dev_df = pd.read_csv(source_dev)
target_dev_df = pd.read_csv(target_dev)
source_dev_df['sent'] = source_dev_df.apply(lambda x: fill_token(x[1], x[2]), axis=1)
target_dev_df['sent'] = target_dev_df.apply(lambda x: fill_token(x[1], x[2]), axis=1)
logging.info('source_dev_df shape {}, target_dev_df shape {}'.format(source_dev_df.shape,target_dev_df.shape))
# ps_proc = TextPhraseSimProcessor(labels=None, feature_columns=['sent', 'term'], label_columns='polarity')
ps_proc = CrossAspectProcessor(labels=None, source_feature_columns=['sent','term'], source_aspect_labels='polarity',
target_feature_columns=['sent','term'], target_aspect_labels='polarity')
logging.info('start get examples')
source_train_examples,target_train_examples = ps_proc.get_train_examples(source_df=source_train_df, target_df=target_train_df, size=-1, labels_available=True)
source_dev_examples, target_dev_examples = ps_proc.get_dev_examples(source_df=source_dev_df,target_df=target_dev_df, size=-1, labels_available=True)
logging.info('source_train_examples {}, target_train_examples {}'.format(len(source_train_examples), len(target_train_examples)))
logging.info('source_dev_examples {}, target_dev_examples {}'.format(len(source_dev_examples), len(target_dev_examples)))
# source_dev_examples, target_dev_examples = ps_proc.get_dev_examples(source_df=source_train_df,target_df=target_train_df, size=100, labels_available=True)
#tokenizer = BertTokenizer.from_pretrained('bert-base-uncased', do_lower_case=True)
tokenizer = BertTokenizer.from_pretrained('bert-base-uncased-vocab.txt', do_lower_case=True)
max_seq_len = 300
# change this label_map based on datset
domain_label_map = {'laptop':0, 'restaurant':1}
sentiment_label_map = {-1:0, 0:1, 1:2}
#inverse_domain_label_map = {val:key for key, val in domain_label_map.items()}
domain_n_labels = 2
sentiment_n_labels = 3
params = {'aspect_n_labels':aspect_n_labels,'domain_n_labels':domain_n_labels,'sentiment_label_map':sentiment_label_map,'domain_label_map':domain_label_map,
'batch_size':16, 'n_epochs':25, 'seq_len':max_seq_len, 'n_workers':-1, 'lr':0.0001, 'keep_prob':0.1, 'heads':12}
#train_data = get_cross_aspect_batch(source_train_examples, domain_label_map, sentiment_label_map, max_seq_len, tokenizer, target_train_examples,
# output_mode="classification",label_available=True, batch_size=16, num_workers=-1)
#dev_data = get_cross_aspect_batch(source_dev_examples, domain_label_map, sentiment_label_map, max_seq_len, tokenizer, target_dev_examples,
# output_mode="classification",label_available=True, batch_size=16, num_workers=-1)
bert_cross_proc = BertCrossFeatures(tokenizer, max_seq_len, domain_label_map, sentiment_label_map)
train_data = get_cross_aspect_batch(source_train_examples,target_train_examples,bert_cross_proc,label_available=True, batch_size=params['batch_size'], num_workers=-1)
logging.info('train_data {}'.format(train_data[0]))
dev_data = get_cross_aspect_batch(source_dev_examples,target_dev_examples,bert_cross_proc,label_available=True, batch_size=params['batch_size'], num_workers=-1)
logging.info('dev_data {}'.format(dev_data[0]))
bert_cross_aspect_model = EATN.from_pretrained('../home/kkzhang/bert_pytorch_model/bert_base', params)
# bert_aspect_model = BertForAspect.from_pretrained('bert-base-uncased', params)
#device = torch.device('cpu')
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
print(device)
if device!='cpu' and torch.cuda.device_count()> 1:
logging.info('has {} gpus'.format(torch.cuda.device_count()))
bert_cross_aspect_model = nn.DataParallel(bert_cross_aspect_model)
bert_cross_aspect_model.to(device)
#loss_fn = F.cross_entropy
loss_fn = None
# maintain all metrics required in this dictionary- these are used in the training and evaluation loops
metrics = {
'accuracy': accuracy,
# could add more metrics such as accuracy for each token type
}
optimizer = optim.Adam(filter(lambda p: p.requires_grad, bert_cross_aspect_model.parameters()), lr=params['lr'])
bert_cross_aspect_trainer = BertMultiTrainer(device,batch_size=params['batch_size'],n_epochs=params['n_epochs'],min_clip_val=-1.0, max_clip_val=1.0)
bert_cross_aspect_trainer.train_and_evaluate(bert_cross_aspect_model,train_data,dev_data,optimizer,metrics,loss_fn=loss_fn,model_dir='./cross_aspect_results/')
bert_aspect_pred = BertPredictor(device, model=bert_aspect_model, max_seq_length=max_seq_len, tokenizer=tokenizer, X_proc=ps_proc, target_int2label_dict=inverse_label_map,
target_label2int_dict=label_map)
if __name__ == '__main__':
logging.basicConfig(level=logging.INFO)
main()