-
Notifications
You must be signed in to change notification settings - Fork 1
/
Copy pathdata_helper.py
161 lines (124 loc) · 4.29 KB
/
data_helper.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
import torch
import nltk
from nltk import tokenize
from nltk.tokenize import TweetTokenizer
import json
import numpy as np
import matplotlib.pyplot as plt
import pylab as pl
from tqdm import *
from collections import defaultdict
import operator
import random
from nltk.tokenize import WordPunctTokenizer
import h5py
wpt = WordPunctTokenizer()
min_context_len = 20
max_context_len = 350
min_question_len = 2
max_question_len = 30
max_answer_len = 30
def helper(data_path,voc_path, number_data = None):
data = json.load(open(data_path))
voc = json.load(open(voc_path))
p_set = []
p_len_set = []
p_c_s_e_set = []
q_set = []
q_len_set = []
g_set = []
a_set = []
a_len_set = []
a_b_set = []
p_str_set = []
q_str_set = []
a_str_set = []
k_set = []
for i in tqdm(range(len(data))):
passage = data[i]['passage']
question = data[i]['question']
answer_text = data[i]['answertext']
answer_start = data[i]['answerstart']
key = data[i]['id']
answer_end = answer_start + len(answer_text) - 1
p_tmp = [voc['pad#'] for j in range(max_context_len+1)]
p_len_tmp = 0
p_c_s_e_tmp = [[-1, -1] for j in range(max_context_len+1)]
a_tmp = [voc['pad#'] for j in range(max_answer_len+1)]
a_len_tmp = 0
a_b_tmp = [1, 0]
start = -1
end = -1
for wi, (cs, ce) in enumerate(wpt.span_tokenize(passage)):
if wi == max_context_len: break
if cs <= answer_start and ce >= answer_start:
start = wi
if cs <= answer_end and ce >= answer_end:
end = wi
p_c_s_e_tmp[wi] = [cs, ce]
word = passage[cs:ce].lower()
if word in voc:
p_tmp[wi] = voc[word]
else:
p_tmp[wi] = voc['unk#']
p_len_tmp = wi + 1
p_set.append(p_tmp)
p_len_set.append(p_len_tmp)
p_c_s_e_set.append(p_c_s_e_tmp)
# 在dev中，如果没有发现完整的answer，就不写入了
if not (end == -1 or start == -1):
#在dev中，answer太长，就clip
end = min(end+1, start+max_answer_len)
for j in range(start, end+1):
a_tmp[j-start] = p_tmp[j]
a_len_tmp = end-start+1
a_b_tmp = [start, end]
a_set.append(a_tmp)
a_len_set.append(a_len_tmp)
a_b_set.append(a_b_tmp)
q_tmp = [voc['pad#'] for j in range(max_question_len+2)]
q_len_tmp = 0
gtruth_tmp = [voc['pad#'] for j in range(max_question_len + 2)]
for wi, token in enumerate(wpt.tokenize(question)):
if wi == max_question_len: break
if token.lower() in voc:
q_tmp[wi] = voc[token.lower()]
gtruth_tmp[wi+1] = voc[token.lower()]
else:
q_tmp[wi] = voc['unk#']
gtruth_tmp[wi+1] = voc['unk#']
q_len_tmp = wi + 1
q_tmp[q_len_tmp] = voc['eos#']
gtruth_tmp[0] = voc['go#']
gtruth_tmp[q_len_tmp] = voc['eos#']
# eos
q_len_tmp += 1
q_set.append(q_tmp)
g_set.append(gtruth_tmp)
q_len_set.append(q_len_tmp)
p_str_set.append(passage)
q_str_set.append(question)
a_str_set.append(answer_text)
k_set.append(key)
data = []
for i in range(len(k_set)):
tmp_dict = defaultdict(lambda : 0)
tmp_dict['passage_tokens'] = np.array(p_set[i])
tmp_dict['passage_len'] = p_len_set[i]
tmp_dict['char_start_end'] = np.array(p_c_s_e_set[i])
tmp_dict['question_tokens'] = np.array(q_set[i])
tmp_dict['question_len'] = q_len_set[i]
tmp_dict['ground_truth'] = np.array(g_set[i])
tmp_dict['answer_tokens'] = np.array(a_set[i])
tmp_dict['answer_len'] = a_len_set[i]
tmp_dict['boundary'] = np.array(a_b_set[i])
tmp_dict['passage_str'] = p_str_set[i]
tmp_dict['question_str'] = q_str_set[i]
tmp_dict['answer_str'] = a_str_set[i]
tmp_dict['key'] = k_set[i]
tmp_dict = dict(tmp_dict)
data.append(tmp_dict)
random.shuffle(data)
if number_data != None:
return data[:number_data]
return data