-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathdata.py
176 lines (157 loc) · 7.96 KB
/
data.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
169
170
171
172
173
174
175
176
import math
import torch
import numpy as np
from transformers import *
from utils import context_models
START_TAG = 'START'
STOP_TAG = 'STOP'
PAD_TAG = 'PAD'
sentiment2id = {'negative': 3, 'neutral': 4, 'positive': 5}
biotags2id = {'O': 0, 'B': 1, 'I': 2}
label2idx = {'O': 0, 'B': 1, 'I': 2, START_TAG: 3, STOP_TAG: 4, PAD_TAG: 5}
idx2labels = ['O', 'B', 'I', START_TAG, STOP_TAG, PAD_TAG]
def get_spans(tags):
"""
for spans
"""
tags = tags.strip().split('<tag>')
length = len(tags)
spans = []
start = -1
for i in range(length):
if tags[i].endswith('B'):
if start != -1:
spans.append([start, i - 1])
start = i
elif tags[i].endswith('O'):
if start != -1:
spans.append([start, i - 1])
start = -1
if start != -1:
spans.append([start, length - 1])
return spans
class Instance(object):
def __init__(self, tokenizer, sentence_pack, args):
self.id = sentence_pack['id']
self.sentence = sentence_pack['sentence']
self.last_review = sentence_pack['split_idx']
self.tokens = self.sentence.strip().split(' <sentsep> ')
self.sen_length = len(self.tokens)
self.bert_tokens = []
self.num_tokens = []
for i, sent in enumerate(self.tokens):
word_tokens = tokenizer.tokenize(" " + sent)
input_ids = tokenizer.convert_tokens_to_ids(
[tokenizer.cls_token] + word_tokens + [tokenizer.sep_token])
self.bert_tokens.append(input_ids)
self.num_tokens.append(min(len(word_tokens), args.max_bert_token-1))
# self.bert_tokens = tokenizer.encode(self.sentence)
self.length = len(self.bert_tokens)
self.token_range = [[i, i] for i in range(self.length)]
self.tags = torch.zeros(self.length, self.length).long()
self.bio = torch.zeros(self.length).long()
self.type_idx = torch.zeros(self.length).long()
self.type_idx[self.last_review+1:self.length] = 1
self.bio[:] = label2idx['O']
if args.cls_method == 'binary':
self.tags[:self.last_review+1, :self.last_review+1] = -1
self.tags[self.last_review+1:, self.last_review+1:] = -1
self.tags[self.last_review+1:, :self.last_review+1] = -1
for triple in sentence_pack['triples']:
aspect = triple['target_tags']
opinion = triple['opinion_tags']
aspect_span = get_spans(aspect)
opinion_span = get_spans(opinion)
for l, r in aspect_span:
for i in range(l, r+1):
if i == l:
self.bio[i] = biotags2id['B']
else:
self.bio[i] = biotags2id['I']
if args.cls_method == 'multiclass':
for j in range(l, r+1):
self.tags[i][j] = 1
for l, r in opinion_span:
for i in range(l, r+1):
if i == l:
self.bio[i] = biotags2id['B']
else:
self.bio[i] = biotags2id['I']
if args.cls_method == 'multiclass':
for j in range(l, r+1):
self.tags[i][j] = 2
for al, ar in aspect_span:
for pl, pr in opinion_span:
for i in range(al, ar+1):
for j in range(pl, pr+1):
if args.task == 'pair':
if args.cls_method == 'binary':
self.tags[i][j] = 1
# self.tags[j][i] = 1
else:
self.tags[i][j] = 3
# self.tags[j][i] = 3
elif args.task == 'triplet':
self.tags[i][j] = sentiment2id[triple['sentiment']]
def load_data_instances(sentence_packs, args):
instances = list()
tokenizer = context_models[args.bert_tokenizer_path]['tokenizer'].from_pretrained(args.bert_tokenizer_path)
if args.num_instances != -1:
for sentence_pack in sentence_packs[:args.num_instances]:
instances.append(Instance(tokenizer, sentence_pack, args))
else:
for sentence_pack in sentence_packs:
instances.append(Instance(tokenizer, sentence_pack, args))
return instances
class DataIterator(object):
def __init__(self, instances, args):
self.instances = instances
self.args = args
self.batch_count = math.ceil(len(instances)/args.batch_size)
self.max_bert_token = args.max_bert_token
def get_batch(self, index):
sentence_ids = []
sentences = []
sens_lens = []
lengths = []
last_review_indice = []
batch_size = min((index + 1) * self.args.batch_size, len(self.instances)) - index * self.args.batch_size
max_num_sents = max([self.instances[i].length for i in range(index * self.args.batch_size,
min((index + 1) * self.args.batch_size, len(self.instances)))])
max_sent_length = min(max([max(map(len, self.instances[i].bert_tokens)) for i in range(index * self.args.batch_size,
min((index + 1) * self.args.batch_size, len(self.instances)))]), self.max_bert_token)
bert_tokens = torch.zeros(batch_size, max_num_sents, max_sent_length, dtype=torch.long)
attn_masks = torch.zeros(batch_size, max_num_sents, max_sent_length, dtype=torch.long)
masks = torch.zeros(batch_size, max_num_sents, dtype=torch.long)
tags = -torch.ones(batch_size, max_num_sents, max_num_sents).long()
biotags = torch.full((batch_size, max_num_sents), label2idx[PAD_TAG]).long()
type_idx = torch.zeros(batch_size, max_num_sents).long()
num_tokens = torch.ones(batch_size, max_num_sents).long()
for i in range(index * self.args.batch_size,
min((index + 1) * self.args.batch_size, len(self.instances))):
sentence_ids.append(self.instances[i].id)
sentences.append(self.instances[i].sentence)
sens_lens.append(self.instances[i].sen_length)
lengths.append(self.instances[i].length)
last_review_indice.append(self.instances[i].last_review)
masks[i-index * self.args.batch_size, :self.instances[i].length] = 1
tags[i-index * self.args.batch_size, :self.instances[i].length, :self.instances[i].length] = self.instances[i].tags
biotags[i-index * self.args.batch_size, :self.instances[i].length] = self.instances[i].bio
type_idx[i-index * self.args.batch_size, :self.instances[i].length] = self.instances[i].type_idx
num_tokens[i-index * self.args.batch_size, :self.instances[i].length] = torch.LongTensor(self.instances[i].num_tokens)
for j in range(self.instances[i].length):
length_filled = min(self.max_bert_token, len(self.instances[i].bert_tokens[j]))
bert_tokens[i-index * self.args.batch_size, j, :length_filled] = \
torch.LongTensor(self.instances[i].bert_tokens[j][:length_filled])
attn_masks[i-index * self.args.batch_size, j, :length_filled] = 1
bert_tokens = bert_tokens.to(self.args.device)
attn_masks = attn_masks.to(self.args.device)
tags = tags.to(self.args.device)
biotags = biotags.to(self.args.device)
lengths = torch.tensor(lengths).to(self.args.device)
masks = masks.to(self.args.device)
type_idx = type_idx.to(self.args.device)
if self.args.token_embedding:
return sentence_ids, (bert_tokens, attn_masks, num_tokens), lengths, sens_lens, tags, biotags, masks, last_review_indice, type_idx
else:
return sentence_ids, (bert_tokens, attn_masks), lengths, sens_lens, tags, biotags, masks, last_review_indice, type_idx