-
Notifications
You must be signed in to change notification settings - Fork 1
/
data.py
191 lines (154 loc) · 6.99 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
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
import numpy as np
import os
import struct
from torch.utils.data import Dataset
from tensorflow.core.example import example_pb2
from fairseq.tasks import FairseqTask
from fairseq.data import data_utils, Dictionary
import torch
import random
import string
def collate(
samples, pad_idx, eos_idx, left_pad_source=True, left_pad_target=False,
input_feeding=True,
):
# taken from https://github.com/pytorch/fairseq/blob/master/fairseq/data/language_pair_dataset.py
if len(samples) == 0:
return {}
def merge(key, left_pad, move_eos_to_beginning=False):
return data_utils.collate_tokens(
[s[key] for s in samples],
pad_idx, eos_idx, left_pad, move_eos_to_beginning,
)
id = torch.LongTensor([s['id'] for s in samples])
src_tokens = merge('article', left_pad=left_pad_source)
# sort by descending source length
src_lengths = torch.LongTensor([s['article'].numel() for s in samples])
src_lengths, sort_order = src_lengths.sort(descending=True)
id = id.index_select(0, sort_order)
src_tokens = src_tokens.index_select(0, sort_order)
prev_output_tokens = None
target = merge('summary', left_pad=left_pad_target)
target = target.index_select(0, sort_order)
ntokens = sum(len(s['summary']) for s in samples)
if input_feeding:
# we create a shifted version of targets for feeding the
# previous output token(s) into the next decoder step
prev_output_tokens = merge(
'summary',
left_pad=left_pad_target,
move_eos_to_beginning=True,
)
prev_output_tokens = prev_output_tokens.index_select(0, sort_order)
batch = {
'id': id,
'nsentences': len(samples),
'ntokens': ntokens,
'net_input': {
'src_tokens': src_tokens,
'src_lengths': src_lengths,
},
'target': target,
}
if prev_output_tokens is not None:
batch['net_input']['prev_output_tokens'] = prev_output_tokens
return batch
class SummaryDataset(Dataset):
'''
'''
def __init__(self, datapath, dictionary, max_article_size=10000, max_summary_size=10000, max_elements=None):
self.datapath = datapath
self.dictionary = dictionary
self.max_article_size = max_article_size
self.max_summary_size = max_summary_size
self.max_elements= max_elements
self.articles = []
self.summaries = []
self.articles_len = []
self.summaries_len = []
self.preprocess()
def preprocess(self):
''' Import the dataset from the binary files.
Code taken and adapted from: https://github.com/abisee/pointer-generator/blob/master/data.py '''
filelist = os.listdir(self.datapath) # get the list of datafiles
filelist = [os.path.join(self.datapath, f) for f in filelist]
filelist.sort()
assert filelist, ('Error: Empty filelist at %s' %
self.datapath) # check filelist isn't empty
for f in filelist:
reader = open(f, 'rb')
while True:
len_bytes = reader.read(8)
if not len_bytes:
break # finished reading this file
str_len = struct.unpack('q', len_bytes)[0]
example_str = struct.unpack(
'%ds' % str_len, reader.read(str_len))[0]
tf_example = example_pb2.Example.FromString(example_str)
examples = []
for key in tf_example.features.feature:
examples.append(
'%s' % (tf_example.features.feature[key].bytes_list.value[0]))
examples[0] = examples[0][2:-1].split()[:self.max_article_size]
examples[1] = [w for w in examples[1][2:-1].split()
if (w != self.dictionary.eos_word and w != '<s>')]
examples[1] = examples[1][:self.max_summary_size]
self.articles.append(examples[0])
self.summaries.append(examples[1])
self.articles_len = np.array([len(a) for a in self.articles], dtype='long')
self.summaries_len = np.array([len(s) for s in self.summaries], dtype='long')
if self.max_elements is not None:
self.articles_len = self.articles_len[:self.max_elements]
self.summaries_len = self.summaries_len[:self.max_elements]
self.articles = self.articles[:self.max_elements]
self.summaries = self.summaries[:self.max_elements]
def tokenize(self, text):
return torch.LongTensor([self.dictionary.index(sym) for sym in text] + [self.dictionary.eos_index])
def ordered_indices(self, shuffle=False):
#TODO use it
"""Return an ordered list of indices. Batches will be constructed based
on this order."""
if shuffle:
indices = np.random.permutation(len(self))
else:
indices = np.arange(len(self))
indices = indices[np.argsort(self.summaries_len[indices], kind='mergesort')]
return indices[np.argsort(self.articles_len[indices], kind='mergesort')]
def __getitem__(self, index):
article, summary = self.articles[index], self.summaries[index]
article, summary = self.tokenize(article), self.tokenize(summary)
item = {'id': index, 'article': article, 'summary': summary}
return item
def __len__(self):
return len(self.articles)
class DummySummaryDataset(Dataset):
def __init__(self,n_summaries,article_size,summary_size,dictionary):
self.article = [random.choice(string.ascii_letters) for _ in range(article_size)]
self.summary = [random.choice(string.ascii_letters) for _ in range(summary_size)]
self.articles = [self.article for _ in range(n_summaries)]
self.summaries = [self.summary for _ in range(n_summaries)]
self.dictionary = dictionary
def tokenize(self, text):
return torch.LongTensor([self.dictionary.index(sym) for sym in text] + [self.dictionary.eos_index])
def __getitem__(self, index):
article, summary = self.articles[index], self.summaries[index]
article, summary = self.tokenize(article), self.tokenize(summary)
item = {'id': index, 'article': article, 'summary': summary}
return item
def __len__(self):
return len(self.articles)
##TODO maybe integrate totally into FairSeq for later use
class SummarizationTask(FairseqTask):
##TODO finish this and design in in the same way as https://github.com/pytorch/fairseq/blob/master/fairseq/tasks/language_modeling.py
def __init__(self, args, dictionary):
super().__init__(args)
self.dictionary = dictionary
@property
def source_dictionary(self):
return self.dictionary
@property
def target_dictionary(self):
return self.dictionary
if __name__ == "__main__":
dataset = SummaryDataset(datapath='datasets/cnn_debug/train', dictionary=Dictionary.load('datasets/vocab'))
print(dataset[3])