/
parse.py
executable file
·213 lines (183 loc) · 7.55 KB
/
parse.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
204
205
206
207
208
209
210
211
212
213
#!/usr/bin/env python
# -*- coding: utf-8 -*-
from __future__ import print_function, division
import argparse
import json
import errno
import os
import struct
import sys
from base64 import urlsafe_b64decode as b64decode, \
urlsafe_b64encode as b64encode
from functools import partial
from itertools import imap, ifilter
import nltk
import pandas as pd
from google.protobuf import text_format
from mctest_pb2 import AnswerAsWords, QuestionAsWords, StoryAsWords, \
AnswerAsEmbeddings, QuestionAsEmbeddings, StoryAsEmbeddings
DEFAULT_OUTPUT_FORMAT = 'json'
COLUMNS = ['id', 'description', 'story',
'q1', 'a11', 'a12', 'a13', 'a14',
'q2', 'a21', 'a22', 'a23', 'a24',
'q3', 'a31', 'a32', 'a33', 'a34',
'q4', 'a41', 'a42', 'a43', 'a44']
QUESTION_TYPES = ['one', 'multiple']
PUNCTS = ['.', '?', ',', '!', '"', '\'', '$', '%', '^', '&']
def question_text(question):
return question.split(':')[1].strip()
def question_type(question):
question_type, _ = question.split(':')
assert question_type in QUESTION_TYPES
return question_type
def row_to_dict(row, tokenize=None):
return {
'id': row['id'],
'description': row['description'],
'passage': tokenize(row['story']),
'questions': [{
'tokens': tokenize(question_text(row['q%d' % q_number])),
'answers': [tokenize(row['a%d%d' % (q_number, a_number)])
for a_number in xrange(1, 5)],
'type': question_type(row['q%d' % q_number])
} for q_number in xrange(1, 5)]
}
def datapoint_to_tokens(datapoint, include_answers=False):
tokens = datapoint['passage']
for question in datapoint['questions']:
tokens.extend(question['tokens'])
if include_answers:
for answer in datapoint['answers']:
tokens.extend(answer)
return ' '.join(map(lambda t: t.lower(), tokens))
def datapoint_to_proto_as_words(datapoint):
story = StoryAsWords()
story.id = datapoint['id']
story.description = datapoint['description']
story.passage.extend(datapoint['passage'])
for question_dict in datapoint['questions']:
question = story.questions.add()
if question_dict['type'] == 'one':
question.type = QuestionAsWords.ONE
elif question_dict['type'] == 'multiple':
question.type = QuestionAsWords.MULTIPLE
else:
print('Invalid question type: %s' % question_dict['type'],
file=sys.stderr)
sys.exit(3)
question.tokens.extend(question_dict['tokens'])
for answer_list in question_dict['answers']:
answer = question.answers.add()
answer.tokens.extend(answer_list)
return story
def datapoint_to_proto_as_embeddings(datapoint):
story = StoryAsEmbeddings()
story.id = datapoint['id']
story.description = datapoint['description']
for passage_vec in datapoint['passage']:
embed = story.passage.add()
embed.value.extend(list(passage_vec))
for question_dict in datapoint['questions']:
question = story.questions.add()
if question_dict['type'] == 'one':
question.type = QuestionAsWords.ONE
elif question_dict['type'] == 'multiple':
question.type = QuestionAsWords.MULTIPLE
else:
print('Invalid question type: %s' % question_dict['type'],
file=sys.stderr)
sys.exit(3)
for token_vec in question_dict['tokens']:
question.tokens.add().value.extend(list(token_vec))
for answer_list in question_dict['answers']:
answer = question.answers.add()
for answer_vec in answer_list:
answer.tokens.add().value.extend(list(answer_vec))
return story
def length_prefix_proto(proto):
serialized = proto.SerializeToString()
return struct.pack('I', len(serialized)) + serialized
def parse_proto_stream(stream, proto_type=StoryAsWords):
while True:
proto_size_bin = stream.read(4)
if len(proto_size_bin) != 4:
if len(proto_size_bin) == 0:
return
print('Invalid read: rubbish at the end of the file?',
file=sys.stderr)
return
proto_size = struct.unpack_from('I', proto_size_bin)[0]
story = proto_type()
story.ParseFromString(stream.read(proto_size))
yield story
if __name__ == '__main__':
parser = argparse.ArgumentParser(
description='Converts raw TSV files from the MCTest dataset')
_arg = parser.add_argument
_arg('-o', type=str, action='store', metavar='FORMAT',
default=DEFAULT_OUTPUT_FORMAT,
help='Output format: json, proto, proto_text, token_list (default=%s)' %
DEFAULT_OUTPUT_FORMAT)
_arg('--rm-stop', type=str, action='store', metavar='FILE',
help='Remove stop words specified by file (one word per line).')
_arg('--rm-punct', action='store_true',
help='Remove punctuation when tokenizing.')
_arg('--model-file', type=str, action='store', metavar='FILE', default=None,
help='File with word2vec model. If provided, makes it output' \
'embeddings.')
_arg('-i', type=str, action='store', metavar='FILE', default=None,
help='Input file (TSV).')
args = parser.parse_args()
token_mappers = []
if args.rm_stop:
stopwords = open(args.rm_stop, 'r').read().split('\n')
stopwords = set(map(lambda x: x.strip().rstrip(), stopwords))
token_mappers.append(lambda x: x if x.lower() not in stopwords else None)
if args.rm_punct:
token_mappers.append(lambda x: x if x not in PUNCTS else None)
as_embeddings = args.model_file is not None
if args.model_file:
import word2vec
embedding_model = word2vec.load(args.model_file)
def to_embeddings(token):
try:
return embedding_model[token.lower()]
except KeyError as e:
print('WARNING: "%s" missing from vocabulary.' % token.lower(),
file=sys.stderr)
return None
token_mappers.append(to_embeddings)
def tokenize(text):
if not isinstance(text, basestring):
text = str(text)
text = text.replace('\\newline', ' ')
mapped = nltk.word_tokenize(text)
for mapper in token_mappers:
mapped = filter(lambda x: x is not None, map(mapper, mapped))
return mapped
data_in = open(args.i, 'r') if args.i else sys.stdin
df = pd.read_csv(data_in, sep='\t', names=COLUMNS)
for row in (df.ix[i] for i in df.index):
datapoint = row_to_dict(row, tokenize)
try:
serialized = None
if args.o == 'json':
serialized = json.dumps(datapoint) + '\n'
elif args.o == 'proto' or args.o == 'proto_text':
proto = datapoint_to_proto_as_embeddings(datapoint) \
if as_embeddings else \
datapoint_to_proto_as_words(datapoint)
serialized = length_prefix_proto(proto) if args.o == 'proto' \
else text_format.MessageToString(proto)
elif args.o == 'token_list':
serialized = datapoint_to_tokens(datapoint) + ' '
else:
print('Unknown output format "%s"' % args.o,
file=sys.stderr)
sys.exit(2)
assert serialized
sys.stdout.write(serialized)
except IOError as e:
if e.errno == errno.EPIPE:
sys.exit(0)
raise e