2015-01-09 17:53:26 +00:00
|
|
|
#!/usr/bin/env python
|
|
|
|
from __future__ import division
|
|
|
|
from __future__ import unicode_literals
|
|
|
|
|
|
|
|
import os
|
|
|
|
from os import path
|
|
|
|
import shutil
|
|
|
|
import codecs
|
|
|
|
import random
|
|
|
|
|
|
|
|
import plac
|
|
|
|
import cProfile
|
|
|
|
import pstats
|
2015-05-23 15:21:25 +00:00
|
|
|
import re
|
2015-01-09 17:53:26 +00:00
|
|
|
|
|
|
|
import spacy.util
|
|
|
|
from spacy.en import English
|
|
|
|
from spacy.en.pos import POS_TEMPLATES, POS_TAGS, setup_model_dir
|
|
|
|
|
|
|
|
from spacy.syntax.util import Config
|
2015-05-24 19:35:02 +00:00
|
|
|
from spacy.gold import read_json_file
|
|
|
|
from spacy.gold import GoldParse
|
2015-01-09 17:53:26 +00:00
|
|
|
|
2015-03-10 17:00:23 +00:00
|
|
|
from spacy.scorer import Scorer
|
|
|
|
|
2015-01-09 17:53:26 +00:00
|
|
|
|
2015-05-24 00:50:14 +00:00
|
|
|
def add_noise(c, noise_level):
|
|
|
|
if random.random() >= noise_level:
|
|
|
|
return c
|
|
|
|
elif c == ' ':
|
|
|
|
return '\n'
|
|
|
|
elif c == '\n':
|
|
|
|
return ' '
|
|
|
|
elif c in ['.', "'", "!", "?"]:
|
|
|
|
return ''
|
|
|
|
else:
|
|
|
|
return c.lower()
|
|
|
|
|
|
|
|
|
2015-05-30 23:11:11 +00:00
|
|
|
def score_model(scorer, nlp, raw_text, annot_tuples, train_tags=None):
|
2015-05-27 17:14:02 +00:00
|
|
|
if raw_text is None:
|
|
|
|
tokens = nlp.tokenizer.tokens_from_list(annot_tuples[1])
|
|
|
|
else:
|
2015-05-31 04:49:06 +00:00
|
|
|
tokens = nlp.tokenizer(raw_text)
|
2015-05-30 23:11:11 +00:00
|
|
|
if train_tags is not None:
|
|
|
|
key = hash(tokens.string)
|
|
|
|
nlp.tagger.tag_from_strings(tokens, train_tags[key])
|
|
|
|
else:
|
|
|
|
nlp.tagger(tokens)
|
|
|
|
|
|
|
|
nlp.entity(tokens)
|
|
|
|
nlp.parser(tokens)
|
2015-05-27 17:14:02 +00:00
|
|
|
gold = GoldParse(tokens, annot_tuples)
|
|
|
|
scorer.score(tokens, gold, verbose=False)
|
|
|
|
|
|
|
|
|
2015-05-30 03:23:02 +00:00
|
|
|
def _merge_sents(sents):
|
|
|
|
m_deps = [[], [], [], [], [], []]
|
|
|
|
m_brackets = []
|
|
|
|
i = 0
|
|
|
|
for (ids, words, tags, heads, labels, ner), brackets in sents:
|
|
|
|
m_deps[0].extend(id_ + i for id_ in ids)
|
|
|
|
m_deps[1].extend(words)
|
|
|
|
m_deps[2].extend(tags)
|
|
|
|
m_deps[3].extend(head + i for head in heads)
|
|
|
|
m_deps[4].extend(labels)
|
|
|
|
m_deps[5].extend(ner)
|
|
|
|
m_brackets.extend((b['first'] + i, b['last'] + i, b['label']) for b in brackets)
|
|
|
|
i += len(ids)
|
|
|
|
return [(m_deps, m_brackets)]
|
|
|
|
|
2015-05-30 23:11:11 +00:00
|
|
|
|
|
|
|
def get_train_tags(Language, model_dir, docs, gold_preproc):
|
|
|
|
taggings = {}
|
|
|
|
for train_part, test_part in get_partitions(docs, 5):
|
|
|
|
nlp = _train_tagger(Language, model_dir, train_part, gold_preproc)
|
|
|
|
for tokens in _tag_partition(nlp, test_part):
|
|
|
|
taggings[hash(tokens.string)] = [w.tag_ for w in tokens]
|
|
|
|
return taggings
|
|
|
|
|
|
|
|
def get_partitions(docs, n_parts):
|
2015-05-30 23:25:02 +00:00
|
|
|
random.shuffle(docs)
|
2015-05-30 23:11:11 +00:00
|
|
|
n_test = len(docs) / n_parts
|
|
|
|
n_train = len(docs) - n_test
|
|
|
|
for part in range(n_parts):
|
|
|
|
start = int(part * n_test)
|
|
|
|
end = int(start + n_test)
|
|
|
|
yield docs[:start] + docs[end:], docs[start:end]
|
|
|
|
|
|
|
|
|
|
|
|
def _train_tagger(Language, model_dir, docs, gold_preproc=False, n_iter=5):
|
|
|
|
pos_model_dir = path.join(model_dir, 'pos')
|
|
|
|
if path.exists(pos_model_dir):
|
|
|
|
shutil.rmtree(pos_model_dir)
|
|
|
|
os.mkdir(pos_model_dir)
|
|
|
|
setup_model_dir(sorted(POS_TAGS.keys()), POS_TAGS, POS_TEMPLATES, pos_model_dir)
|
|
|
|
|
|
|
|
nlp = Language(data_dir=model_dir)
|
|
|
|
|
|
|
|
print "Itn.\tTag %"
|
|
|
|
for itn in range(n_iter):
|
|
|
|
scorer = Scorer()
|
|
|
|
correct = 0
|
|
|
|
total = 0
|
|
|
|
for raw_text, sents in docs:
|
|
|
|
if gold_preproc:
|
|
|
|
raw_text = None
|
|
|
|
else:
|
|
|
|
sents = _merge_sents(sents)
|
|
|
|
for annot_tuples, ctnt in sents:
|
|
|
|
if raw_text is None:
|
|
|
|
tokens = nlp.tokenizer.tokens_from_list(annot_tuples[1])
|
|
|
|
else:
|
|
|
|
tokens = nlp.tokenizer(raw_text)
|
|
|
|
gold = GoldParse(tokens, annot_tuples)
|
|
|
|
correct += nlp.tagger.train(tokens, gold.tags)
|
|
|
|
total += len(tokens)
|
|
|
|
random.shuffle(docs)
|
|
|
|
print itn, '%.3f' % (correct / total)
|
|
|
|
nlp.tagger.model.end_training()
|
|
|
|
nlp.vocab.strings.dump(path.join(model_dir, 'vocab', 'strings.txt'))
|
|
|
|
return nlp
|
|
|
|
|
|
|
|
|
|
|
|
def _tag_partition(nlp, docs, gold_preproc=False):
|
|
|
|
for raw_text, sents in docs:
|
|
|
|
if gold_preproc:
|
|
|
|
raw_text = None
|
|
|
|
else:
|
|
|
|
sents = _merge_sents(sents)
|
|
|
|
for annot_tuples, _ in sents:
|
|
|
|
if raw_text is None:
|
|
|
|
tokens = nlp.tokenizer.tokens_from_list(annot_tuples[1])
|
|
|
|
else:
|
|
|
|
tokens = nlp.tokenizer(raw_text)
|
|
|
|
|
|
|
|
nlp.tagger(tokens)
|
|
|
|
yield tokens
|
|
|
|
|
|
|
|
|
|
|
|
def train(Language, gold_tuples, model_dir, n_iter=15, feat_set=u'basic',
|
|
|
|
seed=0, gold_preproc=False, n_sents=0, corruption_level=0,
|
2015-06-02 00:01:33 +00:00
|
|
|
train_tags=None, beam_width=1):
|
2015-01-09 17:53:26 +00:00
|
|
|
dep_model_dir = path.join(model_dir, 'deps')
|
|
|
|
pos_model_dir = path.join(model_dir, 'pos')
|
2015-03-09 05:46:53 +00:00
|
|
|
ner_model_dir = path.join(model_dir, 'ner')
|
2015-01-09 17:53:26 +00:00
|
|
|
if path.exists(dep_model_dir):
|
|
|
|
shutil.rmtree(dep_model_dir)
|
|
|
|
if path.exists(pos_model_dir):
|
|
|
|
shutil.rmtree(pos_model_dir)
|
2015-03-09 05:46:53 +00:00
|
|
|
if path.exists(ner_model_dir):
|
|
|
|
shutil.rmtree(ner_model_dir)
|
2015-01-09 17:53:26 +00:00
|
|
|
os.mkdir(dep_model_dir)
|
|
|
|
os.mkdir(pos_model_dir)
|
2015-03-09 05:46:53 +00:00
|
|
|
os.mkdir(ner_model_dir)
|
|
|
|
|
|
|
|
setup_model_dir(sorted(POS_TAGS.keys()), POS_TAGS, POS_TEMPLATES, pos_model_dir)
|
|
|
|
|
2015-01-09 17:53:26 +00:00
|
|
|
Config.write(dep_model_dir, 'config', features=feat_set, seed=seed,
|
2015-06-01 22:54:12 +00:00
|
|
|
labels=Language.ParserTransitionSystem.get_labels(gold_tuples),
|
2015-06-02 00:01:33 +00:00
|
|
|
beam_width=beam_width)
|
2015-03-09 11:06:01 +00:00
|
|
|
Config.write(ner_model_dir, 'config', features='ner', seed=seed,
|
2015-06-02 00:01:33 +00:00
|
|
|
labels=Language.EntityTransitionSystem.get_labels(gold_tuples),
|
|
|
|
beam_width=1)
|
2015-03-09 05:46:53 +00:00
|
|
|
|
2015-03-10 17:00:23 +00:00
|
|
|
if n_sents > 0:
|
|
|
|
gold_tuples = gold_tuples[:n_sents]
|
2015-05-30 23:11:11 +00:00
|
|
|
|
2015-04-08 20:48:26 +00:00
|
|
|
nlp = Language(data_dir=model_dir)
|
2015-03-10 17:00:23 +00:00
|
|
|
|
2015-05-24 19:35:02 +00:00
|
|
|
print "Itn.\tP.Loss\tUAS\tNER F.\tTag %\tToken %"
|
2015-01-09 17:53:26 +00:00
|
|
|
for itn in range(n_iter):
|
2015-03-10 17:00:23 +00:00
|
|
|
scorer = Scorer()
|
2015-05-24 19:35:02 +00:00
|
|
|
loss = 0
|
2015-05-29 23:25:46 +00:00
|
|
|
for raw_text, sents in gold_tuples:
|
2015-05-30 03:23:02 +00:00
|
|
|
if gold_preproc:
|
|
|
|
raw_text = None
|
|
|
|
else:
|
2015-05-29 23:25:46 +00:00
|
|
|
sents = _merge_sents(sents)
|
|
|
|
for annot_tuples, ctnt in sents:
|
2015-06-05 00:29:03 +00:00
|
|
|
if len(annot_tuples[1]) == 1:
|
|
|
|
continue
|
2015-05-30 23:11:11 +00:00
|
|
|
score_model(scorer, nlp, raw_text, annot_tuples, train_tags)
|
2015-05-30 03:23:02 +00:00
|
|
|
if raw_text is None:
|
2015-05-29 23:25:46 +00:00
|
|
|
tokens = nlp.tokenizer.tokens_from_list(annot_tuples[1])
|
|
|
|
else:
|
|
|
|
tokens = nlp.tokenizer(raw_text)
|
2015-05-30 23:11:11 +00:00
|
|
|
if train_tags is not None:
|
|
|
|
sent_id = hash(tokens.string)
|
|
|
|
nlp.tagger.tag_from_strings(tokens, train_tags[sent_id])
|
|
|
|
else:
|
|
|
|
nlp.tagger(tokens)
|
|
|
|
gold = GoldParse(tokens, annot_tuples, make_projective=True)
|
2015-06-02 00:01:33 +00:00
|
|
|
loss += nlp.parser.train(tokens, gold)
|
2015-05-30 23:11:11 +00:00
|
|
|
|
2015-05-29 23:25:46 +00:00
|
|
|
nlp.entity.train(tokens, gold)
|
|
|
|
nlp.tagger.train(tokens, gold.tags)
|
2015-05-27 01:16:21 +00:00
|
|
|
random.shuffle(gold_tuples)
|
2015-05-24 19:35:02 +00:00
|
|
|
print '%d:\t%d\t%.3f\t%.3f\t%.3f\t%.3f' % (itn, loss, scorer.uas, scorer.ents_f,
|
2015-05-23 15:21:25 +00:00
|
|
|
scorer.tags_acc,
|
|
|
|
scorer.token_acc)
|
2015-01-09 17:53:26 +00:00
|
|
|
nlp.parser.model.end_training()
|
2015-03-10 17:00:23 +00:00
|
|
|
nlp.entity.model.end_training()
|
2015-01-09 17:53:26 +00:00
|
|
|
nlp.tagger.model.end_training()
|
2015-03-25 00:08:24 +00:00
|
|
|
nlp.vocab.strings.dump(path.join(model_dir, 'vocab', 'strings.txt'))
|
2015-01-09 17:53:26 +00:00
|
|
|
|
|
|
|
|
2015-05-30 03:23:02 +00:00
|
|
|
def evaluate(Language, gold_tuples, model_dir, gold_preproc=False, verbose=False):
|
2015-04-08 20:48:26 +00:00
|
|
|
nlp = Language(data_dir=model_dir)
|
2015-03-10 17:00:23 +00:00
|
|
|
scorer = Scorer()
|
2015-05-29 23:25:46 +00:00
|
|
|
for raw_text, sents in gold_tuples:
|
2015-05-30 03:23:02 +00:00
|
|
|
if gold_preproc:
|
|
|
|
raw_text = None
|
|
|
|
else:
|
|
|
|
sents = _merge_sents(sents)
|
2015-05-29 23:25:46 +00:00
|
|
|
for annot_tuples, brackets in sents:
|
2015-05-30 03:23:02 +00:00
|
|
|
if raw_text is None:
|
2015-05-29 23:25:46 +00:00
|
|
|
tokens = nlp.tokenizer.tokens_from_list(annot_tuples[1])
|
|
|
|
nlp.tagger(tokens)
|
|
|
|
nlp.entity(tokens)
|
|
|
|
nlp.parser(tokens)
|
|
|
|
else:
|
|
|
|
tokens = nlp(raw_text, merge_mwes=False)
|
|
|
|
gold = GoldParse(tokens, annot_tuples)
|
|
|
|
scorer.score(tokens, gold, verbose=verbose)
|
2015-03-10 17:00:23 +00:00
|
|
|
return scorer
|
2015-03-08 05:17:12 +00:00
|
|
|
|
|
|
|
|
2015-03-20 00:14:20 +00:00
|
|
|
def write_parses(Language, dev_loc, model_dir, out_loc):
|
|
|
|
nlp = Language()
|
|
|
|
gold_tuples = read_docparse_file(dev_loc)
|
|
|
|
scorer = Scorer()
|
|
|
|
out_file = codecs.open(out_loc, 'w', 'utf8')
|
|
|
|
for raw_text, segmented_text, annot_tuples in gold_tuples:
|
|
|
|
tokens = nlp(raw_text)
|
|
|
|
for t in tokens:
|
|
|
|
out_file.write(
|
|
|
|
'%s\t%s\t%s\t%s\n' % (t.orth_, t.tag_, t.head.orth_, t.dep_)
|
|
|
|
)
|
|
|
|
return scorer
|
|
|
|
|
|
|
|
|
2015-02-23 19:05:04 +00:00
|
|
|
@plac.annotations(
|
2015-05-29 01:52:55 +00:00
|
|
|
train_loc=("Location of training file or directory"),
|
|
|
|
dev_loc=("Location of development file or directory"),
|
2015-05-24 00:50:14 +00:00
|
|
|
corruption_level=("Amount of noise to add to training data", "option", "c", float),
|
2015-05-29 23:25:46 +00:00
|
|
|
gold_preproc=("Use gold-standard sentence boundaries in training?", "flag", "g", bool),
|
2015-02-23 19:05:04 +00:00
|
|
|
model_dir=("Location of output model directory",),
|
2015-03-20 00:14:20 +00:00
|
|
|
out_loc=("Out location", "option", "o", str),
|
2015-03-14 15:09:55 +00:00
|
|
|
n_sents=("Number of training sentences", "option", "n", int),
|
2015-05-24 19:35:02 +00:00
|
|
|
n_iter=("Number of training iterations", "option", "i", int),
|
2015-06-01 22:54:12 +00:00
|
|
|
beam_width=("Number of candidates to maintain in the beam", "option", "k", int),
|
2015-03-14 15:09:55 +00:00
|
|
|
verbose=("Verbose error reporting", "flag", "v", bool),
|
2015-03-24 03:25:38 +00:00
|
|
|
debug=("Debug mode", "flag", "d", bool)
|
2015-02-23 19:05:04 +00:00
|
|
|
)
|
2015-05-24 19:35:02 +00:00
|
|
|
def main(train_loc, dev_loc, model_dir, n_sents=0, n_iter=15, out_loc="", verbose=False,
|
2015-06-01 22:54:12 +00:00
|
|
|
debug=False, corruption_level=0.0, gold_preproc=False, beam_width=1):
|
2015-05-28 20:40:26 +00:00
|
|
|
gold_train = list(read_json_file(train_loc))
|
2015-05-31 13:21:56 +00:00
|
|
|
#taggings = get_train_tags(English, model_dir, gold_train, gold_preproc)
|
|
|
|
taggings = None
|
2015-05-28 20:40:26 +00:00
|
|
|
train(English, gold_train, model_dir,
|
|
|
|
feat_set='basic' if not debug else 'debug',
|
2015-05-29 23:25:46 +00:00
|
|
|
gold_preproc=gold_preproc, n_sents=n_sents,
|
2015-05-30 23:11:11 +00:00
|
|
|
corruption_level=corruption_level, n_iter=n_iter,
|
2015-06-01 22:54:12 +00:00
|
|
|
train_tags=taggings, beam_width=beam_width)
|
2015-05-30 03:23:02 +00:00
|
|
|
if out_loc:
|
|
|
|
write_parses(English, dev_loc, model_dir, out_loc)
|
2015-05-27 17:14:02 +00:00
|
|
|
scorer = evaluate(English, list(read_json_file(dev_loc)),
|
2015-05-29 23:25:46 +00:00
|
|
|
model_dir, gold_preproc=gold_preproc, verbose=verbose)
|
2015-05-23 15:21:25 +00:00
|
|
|
print 'TOK', 100-scorer.token_acc
|
2015-03-10 17:00:23 +00:00
|
|
|
print 'POS', scorer.tags_acc
|
|
|
|
print 'UAS', scorer.uas
|
|
|
|
print 'LAS', scorer.las
|
|
|
|
|
|
|
|
print 'NER P', scorer.ents_p
|
|
|
|
print 'NER R', scorer.ents_r
|
|
|
|
print 'NER F', scorer.ents_f
|
2015-04-19 08:31:31 +00:00
|
|
|
|
2015-01-09 17:53:26 +00:00
|
|
|
|
|
|
|
if __name__ == '__main__':
|
|
|
|
plac.call(main)
|