explosion/spaCy
1
1
Fork 0
mirror of https://github.com/explosion/spaCy.git synced 2025-02-24 15:47:33 +03:00
Code Issues Packages Projects Releases Wiki Activity

[ja] Stash tokenizer output for speed

Browse Source 
Before this commit, the Mecab tokenizer had to be called twice when
creating a Doc- once during tokenization and once during tagging. This
creates a JapaneseDoc wrapper class for Doc that stashes the parsed
tokenizer output to remove redundant processing. -POLM
This commit is contained in:
Paul O'Leary McCann 2017-10-15 23:33:25 +09:00
parent a31d33be06
commit 43eedf73f2
1 changed files with 14 additions and 7 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

21
spacy/ja/__init__.py
View File

@ -16,6 +16,13 @@ from collections import namedtuple
ShortUnitWord = namedtuple('ShortUnitWord', ['surface', 'base_form', 'part_of_speech']) ShortUnitWord = namedtuple('ShortUnitWord', ['surface', 'base_form', 'part_of_speech'])
class JapaneseDoc(Doc):
def __init__(self, detailed_tokens, vocab, words=None, spaces=None, orths_and_spaces=None):
super(JapaneseDoc, self).__init__(vocab, words, spaces, orths_and_spaces)
# This saves tokenizer output so mecab doesn't have to be called again
# when determining POS tags.
self.detailed_tokens = detailed_tokens
def try_mecab_import(): def try_mecab_import():
"""Mecab is required for Japanese support, so check for it. """Mecab is required for Japanese support, so check for it.
@ -34,8 +41,9 @@ class JapaneseTokenizer(object):
self.tokenizer = MeCab.Tagger() self.tokenizer = MeCab.Tagger()
def __call__(self, text): def __call__(self, text):
words = [x.surface for x in detailed_tokens(self.tokenizer, text)] dtokens = detailed_tokens(self.tokenizer, text)
return Doc(self.vocab, words=words, spaces=[False]*len(words)) words = [x.surface for x in dtokens]
return JapaneseDoc(dtokens, self.vocab, words=words, spaces=[False]*len(words))
def resolve_pos(token): def resolve_pos(token):
"""If necessary, add a field to the POS tag for UD mapping. """If necessary, add a field to the POS tag for UD mapping.
@ -91,7 +99,7 @@ class JapaneseTagger(object):
# 1. get raw JP tags # 1. get raw JP tags
# 2. add features to tags as necessary for UD # 2. add features to tags as necessary for UD
dtokens = detailed_tokens(self.tokenizer, tokens.text) dtokens = tokens.detailed_tokens
rawtags = list(map(resolve_pos, dtokens)) rawtags = list(map(resolve_pos, dtokens))
self.tagger.tag_from_strings(tokens, rawtags) self.tagger.tag_from_strings(tokens, rawtags)
@ -112,8 +120,7 @@ class Japanese(Language):
Defaults = JapaneseDefaults Defaults = JapaneseDefaults
def make_doc(self, text): def make_doc(self, text):
words = [str(t) for t in self.tokenizer(text)] jdoc = self.tokenizer(text)
doc = Doc(self.vocab, words=words, spaces=[False]*len(words))
tagger = JapaneseDefaults.create_tagger(self.tokenizer) tagger = JapaneseDefaults.create_tagger(self.tokenizer)
tagger(doc) tagger(jdoc)
return doc return jdoc