-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathdata_loader.py
203 lines (166 loc) · 7.98 KB
/
data_loader.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
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
import torch.utils.data as data
import torch
import numpy as np
import h5py
import json
import random
class load_data(data.Dataset): # torch wrapper
def __init__(self, input_h5, input_vocab, num_val, split_type, negative_sample=20):
print('DataLoader loading: %s' % split_type)
vocab = torch.load(input_vocab)
vocab = dict(vocab)
if num_val <= 0:
num_val = None
# f = json.load(open(input_json, 'r'))
self.src_stoi = vocab['src'].stoi
self.tgt_stoi = vocab['tgt'].stoi
self.src_itos = vocab['src'].itos
self.tgt_itos = vocab['tgt'].itos
print("self.src_itos[:2]", self.src_itos[:10])
print("self.tgt_itos[:2]", self.tgt_itos[:10])
print('Loading txt from %s' %input_h5)
f = h5py.File(input_h5, 'r')
self.src = f['src'][:num_val]
print('%s number of data: %d' % (split_type, len(self.src)), self.src.shape)
self.tgt = f['tgt'][:num_val]
self.src_len = f['src_len'][:num_val]
self.tgt_len = f['tgt_len'][:num_val]
self.tgt_ids = f['tgt_index'][:num_val]
# print("DataLoader", len(self.tgt_ids), self.tgt_ids[-1000:-1])
self.opt_ids = f['opt'][:num_val]
self.opt_list = f['opt_list'][:]
self.opt_len = f['opt_len'][:]
f.close()
print(self.src.shape)
self.src_length = self.src.shape[1]
self.tgt_length = self.tgt.shape[1]
self.src_vocab_size = len(self.src_itos)
self.tgt_vocab_size = len(self.tgt_itos)
# print('src vocab Size: %d' % self.src_vocab_size)
# print('tgt vocab Size: %d' % self.tgt_vocab_size)
self.split = split_type
# self.rnd = 1
self.negative_sample = negative_sample
if not split_type == "train":
self.negative_sample = 100
def __getitem__(self, index):
# print ("self.src_length", self.src_length)
# print ("self.tgt_length", self.tgt_length)
# src = np.zeros([self.src_length], dtype='int64')
src = np.full([self.src_length], self.src_stoi['<blank>'], dtype='int64')
# tgt = np.zeros([self.tgt_length+1], dtype='int64')
tgt = np.full([self.tgt_length+1], self.tgt_stoi['<blank>'], dtype='int64')
tgt_target = np.full([self.tgt_length+1], self.tgt_stoi['<blank>'], dtype='int64')
# tgt_target = np.zeros((self.tgt_length+1), dtype='int64')
# src_ori = np.zeros((self.src_length), dtype='int64')
src_ori = np.full((self.src_length), self.src_stoi['<blank>'], dtype='int64')
# if self.split == "train":
# opt_tgt = np.zeros((self.negative_sample, self.tgt_length+1), dtype='int64')
# else:
opt_tgt = np.full((self.negative_sample, self.tgt_length+1), self.tgt_stoi['<blank>'], dtype='int64')
# opt_tgt = np.full((self.negative_sample, self.tgt_length+1), self.tgt_stoi['<pad>'], dtype='int64')
# opt_tgt = np.full((self.negative_sample, self.tgt_length+1), self.tgt_stoi['</s>'], dtype='int64')
# opt_tgt_target = np.zeros((self.negative_sample, self.tgt_length+1), dtype='int64')
opt_tgt_target = np.full((self.negative_sample, self.tgt_length+1), self.tgt_stoi['<blank>'], dtype='int64')
tgt_len = np.zeros((1), dtype='int64')
src_len = np.zeros((1), dtype='int64')
opt_tgt_len = np.zeros((self.negative_sample), dtype='int64')
tgt_idx = np.zeros((1),dtype='int64')
tgt_ids = np.zeros((1),dtype='int64')
opt_tgt_idx = np.zeros((self.negative_sample), dtype='int64')
# for i in range(self.rnd):
# get the index
# print("index", index)
s_len = self.src_len[index]
# print("s_len", s_len)
t_len = self.tgt_len[index]
# print("t_len", t_len)
# qt_len = s_len + t_len
# if i+1 < self.rnd:
# print ("if i+1 < self.rnd:")
"""把内容放右端"""
src[self.src_length-s_len:] = self.src[index, :s_len]
"""把内容放左端"""
src_ori[:s_len] = self.src[index, :s_len]
tgt[1:t_len+1] = self.tgt[index, :t_len]
# print("DataLoader.py", self.tgt[index, :t_len])
# exit()
tgt[0] = self.tgt_stoi['<s>']
if t_len < self.tgt_length:
tgt[t_len+1] = self.tgt_stoi['</s>']
# else:
# tgt[t_len] = self.tgt_stoi['</s>']
tgt_target[:t_len] = self.tgt[index, :t_len]
tgt_target[t_len] = self.tgt_stoi['</s>']
tgt_len[0] = self.tgt_len[index]
src_len[0] = self.src_len[index]
opt_ids = self.opt_ids[index] # since python start from 0
# print("opt_ids", opt_ids)
# random select the negative samples.
tgt_idx[0] = opt_ids[self.tgt_ids[index]]
# print("load_data self.tgt_ids[index] index", index, self.tgt_ids[index])
# print("load_data, tgt_idx", tgt_idx)
tgt_ids[0] = self.tgt_ids[index]
# print("tgt_ids", tgt_idx)
# exclude the gt index.
# opt_ids = np.delete(opt_ids, tgt_idx, 0)
if self.split == "train":
# print ("train")
opt_ids = np.delete(opt_ids, self.tgt_ids[index], 0)
# opt_ids = np.delete(opt_ids, tgt_idx[0], 0)
# print("opt_ids", opt_ids)
# print("self.negative_sample", self.negative_sample)
random.shuffle(opt_ids)
for j in range(self.negative_sample):
ids = opt_ids[j]
# print("j ids", ids)
opt_tgt_idx[j] = ids
opt_len = self.opt_len[ids]
# print("DataLoader opt_len", opt_len)
opt_tgt_len[j] = opt_len
opt_tgt[j, 1:opt_len+1] = self.opt_list[ids,:opt_len]
opt_tgt[j, 0] = self.tgt_stoi['<s>']
if opt_len < self.tgt_length:
opt_tgt[j, opt_len+1] = self.tgt_stoi['</s>']
# else:
# opt_tgt[j, opt_len] = self.tgt_stoi['</s>']
opt_tgt_target[j, :opt_len] = self.opt_list[ids,:opt_len]
opt_tgt_target[j, opt_len] = self.tgt_stoi['</s>']
else:
# print ("opt_ids", len(opt_ids))
for j, ids in enumerate(opt_ids):
opt_len = self.opt_len[ids]
# print("DataLoader opt_len", opt_len)
opt_tgt[j, 1:opt_len+1] = self.opt_list[ids,:opt_len]
opt_tgt[j, 0] = self.tgt_stoi['<s>']
if opt_len < self.tgt_length:
opt_tgt[j, opt_len+1] = self.tgt_stoi['</s>']
# else:
# opt_tgt[j, opt_len] = self.tgt_stoi['</s>']
if opt_len > 50:
print("DataLoader.py opt_len", opt_len)
exit()
opt_tgt_target[j, :opt_len] = self.opt_list[ids,:opt_len]
opt_tgt_target[j, opt_len] = self.tgt_stoi['</s>']
opt_tgt_idx[j] = opt_ids[j]
opt_tgt_len[j] = opt_len
src = torch.from_numpy(src)
tgt = torch.from_numpy(tgt)
tgt_target = torch.from_numpy(tgt_target)
src_ori = torch.from_numpy(src_ori)
tgt_len = torch.from_numpy(tgt_len)
src_len = torch.from_numpy(src_len)
opt_tgt_len = torch.from_numpy(opt_tgt_len)
opt_tgt = torch.from_numpy(opt_tgt)
opt_tgt_target = torch.from_numpy(opt_tgt_target)
tgt_idx = torch.from_numpy(tgt_idx)
tgt_ids = torch.from_numpy(tgt_ids)
opt_tgt_idx = torch.from_numpy(opt_tgt_idx)
# print src
# print tgt
if not self.split == "train":
# return src, tgt, tgt_target, tgt_len, tgt_idx, src_ori, opt_tgt, opt_tgt_target, opt_tgt_len
return src_len, src, tgt, tgt_target, tgt_ids, src_ori, opt_tgt, opt_tgt_target, opt_tgt_len, opt_tgt_idx
return src_len, src, tgt, tgt_target, tgt_len, tgt_idx, src_ori, opt_tgt, opt_tgt_target, opt_tgt_len, opt_tgt_idx, tgt_ids
def __len__(self):
return self.src.shape[0]