File size: 10,864 Bytes
158b61b |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 |
#!/usr/bin/env python
# -*- coding: utf-8 -*-
# Author: Rico Sennrich
# flake8: noqa
"""Use operations learned with learn_bpe.py to encode a new text.
The text will not be smaller, but use only a fixed vocabulary, with rare words
encoded as variable-length sequences of subword units.
Reference:
Rico Sennrich, Barry Haddow and Alexandra Birch (2015). Neural Machine Translation of Rare Words with Subword Units.
Proceedings of the 54th Annual Meeting of the Association for Computational Linguistics (ACL 2016). Berlin, Germany.
"""
# This file is retrieved from https://github.com/rsennrich/subword-nmt
from __future__ import unicode_literals, division
import sys
import codecs
import io
import argparse
import json
import re
from collections import defaultdict
# hack for python2/3 compatibility
from io import open
argparse.open = open
class BPE(object):
def __init__(self, codes, separator='@@', vocab=None, glossaries=None):
# check version information
firstline = codes.readline()
if firstline.startswith('#version:'):
self.version = tuple([int(x) for x in re.sub(
r'(\.0+)*$', '', firstline.split()[-1]).split(".")])
else:
self.version = (0, 1)
codes.seek(0)
self.bpe_codes = [tuple(item.split()) for item in codes]
# some hacking to deal with duplicates (only consider first instance)
self.bpe_codes = dict(
[(code, i) for (i, code) in reversed(list(enumerate(self.bpe_codes)))])
self.bpe_codes_reverse = dict(
[(pair[0] + pair[1], pair) for pair, i in self.bpe_codes.items()])
self.separator = separator
self.vocab = vocab
self.glossaries = glossaries if glossaries else []
self.cache = {}
def segment(self, sentence):
"""segment single sentence (whitespace-tokenized string) with BPE encoding"""
output = []
for word in sentence.split():
new_word = [out for segment in self._isolate_glossaries(word)
for out in encode(segment,
self.bpe_codes,
self.bpe_codes_reverse,
self.vocab,
self.separator,
self.version,
self.cache,
self.glossaries)]
for item in new_word[:-1]:
output.append(item + self.separator)
output.append(new_word[-1])
return ' '.join(output)
def _isolate_glossaries(self, word):
word_segments = [word]
for gloss in self.glossaries:
word_segments = [out_segments for segment in word_segments
for out_segments in isolate_glossary(segment, gloss)]
return word_segments
def create_parser():
parser = argparse.ArgumentParser(
formatter_class=argparse.RawDescriptionHelpFormatter,
description="learn BPE-based word segmentation")
parser.add_argument(
'--input', '-i', type=argparse.FileType('r'), default=sys.stdin,
metavar='PATH',
help="Input file (default: standard input).")
parser.add_argument(
'--codes', '-c', type=argparse.FileType('r'), metavar='PATH',
required=True,
help="File with BPE codes (created by learn_bpe.py).")
parser.add_argument(
'--output', '-o', type=argparse.FileType('w'), default=sys.stdout,
metavar='PATH',
help="Output file (default: standard output)")
parser.add_argument(
'--separator', '-s', type=str, default='@@', metavar='STR',
help="Separator between non-final subword units (default: '%(default)s'))")
parser.add_argument(
'--vocabulary', type=argparse.FileType('r'), default=None,
metavar="PATH",
help="Vocabulary file (built with get_vocab.py). If provided, this script reverts any merge operations that produce an OOV.")
parser.add_argument(
'--vocabulary-threshold', type=int, default=None,
metavar="INT",
help="Vocabulary threshold. If vocabulary is provided, any word with frequency < threshold will be treated as OOV")
parser.add_argument(
'--glossaries', type=str, nargs='+', default=None,
metavar="STR",
help="Glossaries. The strings provided in glossaries will not be affected" +
"by the BPE (i.e. they will neither be broken into subwords, nor concatenated with other subwords")
return parser
def get_pairs(word):
"""Return set of symbol pairs in a word.
word is represented as tuple of symbols (symbols being variable-length strings)
"""
pairs = set()
prev_char = word[0]
for char in word[1:]:
pairs.add((prev_char, char))
prev_char = char
return pairs
def encode(orig, bpe_codes, bpe_codes_reverse, vocab, separator, version, cache, glossaries=None):
"""Encode word based on list of BPE merge operations, which are applied consecutively
"""
if orig in cache:
return cache[orig]
if orig in glossaries:
cache[orig] = (orig,)
return (orig,)
if version == (0, 1):
word = tuple(orig) + ('</w>',)
elif version == (0, 2): # more consistent handling of word-final segments
word = tuple(orig[:-1]) + (orig[-1] + '</w>',)
else:
raise NotImplementedError
pairs = get_pairs(word)
if not pairs:
return orig
while True:
bigram = min(pairs, key=lambda pair: bpe_codes.get(pair, float('inf')))
if bigram not in bpe_codes:
break
first, second = bigram
new_word = []
i = 0
while i < len(word):
try:
j = word.index(first, i)
new_word.extend(word[i:j])
i = j
except:
new_word.extend(word[i:])
break
if word[i] == first and i < len(word) - 1 and word[i + 1] == second:
new_word.append(first + second)
i += 2
else:
new_word.append(word[i])
i += 1
new_word = tuple(new_word)
word = new_word
if len(word) == 1:
break
else:
pairs = get_pairs(word)
# don't print end-of-word symbols
if word[-1] == '</w>':
word = word[:-1]
elif word[-1].endswith('</w>'):
word = word[:-1] + (word[-1].replace('</w>', ''),)
if vocab:
word = check_vocab_and_split(word, bpe_codes_reverse, vocab, separator)
cache[orig] = word
return word
def recursive_split(segment, bpe_codes, vocab, separator, final=False):
"""Recursively split segment into smaller units (by reversing BPE merges)
until all units are either in-vocabulary, or cannot be split futher."""
try:
if final:
left, right = bpe_codes[segment + '</w>']
right = right[:-4]
else:
left, right = bpe_codes[segment]
except:
#sys.stderr.write('cannot split {0} further.\n'.format(segment))
yield segment
return
if left + separator in vocab:
yield left
else:
for item in recursive_split(left, bpe_codes, vocab, separator, False):
yield item
if (final and right in vocab) or (not final and right + separator in vocab):
yield right
else:
for item in recursive_split(right, bpe_codes, vocab, separator, final):
yield item
def check_vocab_and_split(orig, bpe_codes, vocab, separator):
"""Check for each segment in word if it is in-vocabulary,
and segment OOV segments into smaller units by reversing the BPE merge operations"""
out = []
for segment in orig[:-1]:
if segment + separator in vocab:
out.append(segment)
else:
#sys.stderr.write('OOV: {0}\n'.format(segment))
for item in recursive_split(segment, bpe_codes, vocab, separator, False):
out.append(item)
segment = orig[-1]
if segment in vocab:
out.append(segment)
else:
#sys.stderr.write('OOV: {0}\n'.format(segment))
for item in recursive_split(segment, bpe_codes, vocab, separator, True):
out.append(item)
return out
def read_vocabulary(vocab_file, threshold):
"""read vocabulary file produced by get_vocab.py, and filter according to frequency threshold.
"""
vocabulary = set()
for line in vocab_file:
word, freq = line.split()
freq = int(freq)
if threshold == None or freq >= threshold:
vocabulary.add(word)
return vocabulary
def isolate_glossary(word, glossary):
"""
Isolate a glossary present inside a word.
Returns a list of subwords. In which all 'glossary' glossaries are isolated
For example, if 'USA' is the glossary and '1934USABUSA' the word, the return value is:
['1934', 'USA', 'B', 'USA']
"""
if word == glossary or glossary not in word:
return [word]
else:
splits = word.split(glossary)
segments = [segment.strip() for split in splits[:-1]
for segment in [split, glossary] if segment != '']
return segments + [splits[-1].strip()] if splits[-1] != '' else segments
if __name__ == '__main__':
# python 2/3 compatibility
if sys.version_info < (3, 0):
sys.stderr = codecs.getwriter('UTF-8')(sys.stderr)
sys.stdout = codecs.getwriter('UTF-8')(sys.stdout)
sys.stdin = codecs.getreader('UTF-8')(sys.stdin)
else:
sys.stdin = io.TextIOWrapper(sys.stdin.buffer, encoding='utf-8')
sys.stderr = io.TextIOWrapper(sys.stderr.buffer, encoding='utf-8')
sys.stdout = io.TextIOWrapper(
sys.stdout.buffer, encoding='utf-8', write_through=True, line_buffering=True)
parser = create_parser()
args = parser.parse_args()
# read/write files as UTF-8
args.codes = codecs.open(args.codes.name, encoding='utf-8')
if args.input.name != '<stdin>':
args.input = codecs.open(args.input.name, encoding='utf-8')
if args.output.name != '<stdout>':
args.output = codecs.open(args.output.name, 'w', encoding='utf-8')
if args.vocabulary:
args.vocabulary = codecs.open(args.vocabulary.name, encoding='utf-8')
if args.vocabulary:
vocabulary = read_vocabulary(
args.vocabulary, args.vocabulary_threshold)
else:
vocabulary = None
bpe = BPE(args.codes, args.separator, vocabulary, args.glossaries)
for line in args.input:
args.output.write(bpe.segment(line).strip())
args.output.write('\n')
|