def __init__(self, _name = 'crf_segmenter', verbose = False, global_features = False):
self.name = _name
self.verbose = verbose
self.feature_writer = SegmenterFeatureWriter()
self.global_features = global_features
self.add_classifiers()
class CRFSegmenter:
def __init__(self,
_name='crf_segmenter',
verbose=False,
global_features=False):
self.name = _name
self.verbose = verbose
self.feature_writer = SegmenterFeatureWriter()
self.global_features = global_features
self.add_classifiers()
def add_classifiers(self):
self.classifiers = []
classifier1 = CRFClassifier(name=self.name,
model_type='segmenter',
model_path=paths.SEGMENTER_MODEL_PATH,
model_file='seg.crfsuite',
verbose=self.verbose)
self.add_classifier(classifier1, 'classifier1')
if self.global_features:
classifier2 = CRFClassifier(
name=self.name + "_global_features",
model_type='segmenter',
model_path=paths.SEGMENTER_MODEL_PATH,
model_file='seg_global_features.crfsuite',
verbose=self.verbose)
self.add_classifier(classifier2, 'classifier2')
# classifier1 = MalletCRFClassifier(name= self.name,
# model_type = 'segmenter',
# model_path = paths.CRF_SEGMENTER_MODEL_PATH,
# model_file = 'seg.mallet',
# verbose = self.verbose)
# self.add_classifier(classifier1, 'classifier1')
#
# if self.global_features:
# classifier2 = MalletCRFClassifier(name= self.name + "_global_features",
# model_type = 'segmenter',
# model_path = paths.CRF_SEGMENTER_MODEL_PATH,
# model_file = 'seg_global_features.mallet',
# verbose = self.verbose)
#
#
#
# self.add_classifier(classifier2, 'classifier2')
#
def add_classifier(self, classifier, name):
if name == 'classifier1':
self.classifier = classifier
elif name == 'classifier2':
self.global_features_classifier = classifier
else:
raise Exception('Unknown classifier name', name)
print('Added classifier', name, 'to segmenter', self.name)
self.classifiers.append(classifier)
def unload(self):
for classifier in self.classifiers:
classifier.unload()
def write_features(self, sentence, input_edu_segmentation=None):
features = []
#print sentence.sent_id
# edu_segmentation = sentence.doc.edu_word_segmentation[sentence.sent_id]
# print edu_segmentation
for i in range(len(sentence.tokens) - 1):
token0 = None if i == 0 else sentence.tokens[i - 1]
token1 = sentence.tokens[i]
token2 = sentence.tokens[i + 1]
token3 = None if i == len(
sentence.tokens) - 2 else sentence.tokens[i + 2]
edu_segmentation = input_edu_segmentation
# print edu_segmentation
inst_features = self.feature_writer.write_features(
[token0, token1, token2, token3], edu_segmentation)
feature_str = '\t'.join(list(inst_features))
# print feature_str
features.append('0\t%s' % feature_str)
# feature_str = ' '.join(list(inst_features))
## print feature_str
# features.append('%s 0' % feature_str)
return features
def find_neighbouring_boundary(self,
token_id,
edu_segmentation,
direction='L'):
if direction == 'R':
for j in range(len(edu_segmentation)):
# print edu_segmentation[j]
if edu_segmentation[j][1] > token_id:
return edu_segmentation[j][1]
else:
for j in range(len(edu_segmentation) - 1, -1, -1):
# print edu_segmentation[j]
if edu_segmentation[j][0] < token_id - 1:
return edu_segmentation[j][0]
def segment_sentence(self, sentence, input_edu_segmentation=None):
if len(sentence.tokens) == 1:
edus = [[sentence.tokens[0].word, sentence.raw_text[-3:]]]
sentence.doc.cuts.append(
(len(sentence.doc.edus), len(sentence.doc.edus) + len(edus)))
sentence.start_edu = len(sentence.doc.edus)
sentence.end_edu = len(sentence.doc.edus) + len(edus)
sentence.doc.edu_word_segmentation.append([(0, 1)])
sentence.doc.edus.extend(edus)
return
# print sentence.raw_text
self.feature_writer.cached_subtrees = {}
if input_edu_segmentation:
# print input_edu_segmentation
offset2neighbouring_boundaries = {}
for (j, (start_word,
end_word)) in enumerate(input_edu_segmentation):
for offset in range(start_word, end_word):
if offset == start_word:
l_boundary = input_edu_segmentation[
j - 1][0] if j > 0 else None
else:
l_boundary = start_word
if offset == end_word - 1:
r_boundary = input_edu_segmentation[j + 1][1] if (
j < len(input_edu_segmentation) - 1) else None
else:
r_boundary = end_word
# l_boundary1 = self.find_neighbouring_boundary(offset + 1, input_edu_segmentation, 'L')
# r_boundary1 = self.find_neighbouring_boundary(offset + 1, input_edu_segmentation, 'R')
##
## print offset, l_boundary, r_boundary, l_boundary1, r_boundary1
# assert l_boundary1 == l_boundary and r_boundary1 == r_boundary
offset2neighbouring_boundaries[offset] = (l_boundary,
r_boundary)
#
# for offset in offset2neighbouring_boundaries:
# print 'sentence', j, offset, offset2neighbouring_boundaries[offset]
else:
offset2neighbouring_boundaries = None
features = self.write_features(sentence,
offset2neighbouring_boundaries)
# print sentence.raw_text
# print len(features)
# for feat in features:
# print feat
# print
if input_edu_segmentation:
seq_prob, predictions = self.global_features_classifier.classify(
features)
else:
seq_prob, predictions = self.classifier.classify(features)
edus = []
edu_word_segmentations = []
start = 0
for i in range(len(predictions)):
pred = int(predictions[i][0])
if pred == 1:
# print i, pred
edu_word_segmentations.append((start, i + 1))
start = i + 1
edu_word_segmentations.append((start, len(sentence.tokens)))
for (start_word, end_word) in edu_word_segmentations:
edu = []
for j in range(start_word, end_word):
edu.extend(
utils.utils.unescape_penn_special_word(
sentence.tokens[j].word).split(' '))
if end_word == len(sentence.tokens):
# print sentence.raw_text
edu.append(sentence.raw_text[-3:])
edus.append(edu)
sentence.doc.cuts.append(
(len(sentence.doc.edus), len(sentence.doc.edus) + len(edus)))
sentence.start_edu = len(sentence.doc.edus)
sentence.end_edu = len(sentence.doc.edus) + len(edus)
sentence.doc.edu_word_segmentation.append(edu_word_segmentations)
sentence.doc.edus.extend(edus)
# print sentence.doc.edu_word_segmentation[-1]
def segment_permutation(self, doc, canonical_doc):
assert len(doc.sentences) == len(canonical_doc.sentences)
doc.edu_word_segmentation = []
doc.cuts = []
doc.edus = []
sentence_order = []
for sent in doc.sentences:
index = 0
# print sent.sent_id, sent.raw_text
while index < len(canonical_doc.sentences):
# print 'canonical', index, canonical_doc.sentences[index].raw_text
if canonical_doc.sentences[
index].raw_text == sent.raw_text and index not in sentence_order:
# print index
# print canonical_doc.sentences[index].raw_text
# print sent.raw_text
# print
break
index += 1
# if index == len(canonical_doc.sentences):
# print sent.sent_id, sent.raw_text
# print canonical_doc.sentences[index - 1].raw_text
# print
# print
# print sent.sent_id, index
sentence_order.append(index)
# print sentence_order
# print sorted(sentence_order)
assert sorted(sentence_order) == range(len(doc.sentences))
for (i, index) in enumerate(sentence_order):
sentence = doc.sentences[i]
sentence.set_unlexicalized_tree(
canonical_doc.sentences[index].unlexicalized_parse_tree)
sentence.set_lexicalized_tree(
canonical_doc.sentences[index].parse_tree)
sentence.tokens = []
for te in canonical_doc.sentences[index].tokens:
token = Token(te.word, te.id, sentence)
token.lemma = te.lemma
sentence.add_token(token)
# print i
# print sentence.raw_text
# print canonical_doc.sentences[index].raw_text
# print
(canonical_start_edu,
canonical_end_edu) = canonical_doc.cuts[index]
edus = canonical_doc.edus[canonical_start_edu:canonical_end_edu]
sentence.start_edu = len(sentence.doc.edus)
sentence.end_edu = len(sentence.doc.edus) + len(edus)
doc.cuts.append(
(len(sentence.doc.edus), len(sentence.doc.edus) + len(edus)))
doc.edus.extend(edus)
doc.edu_word_segmentation.append(
canonical_doc.edu_word_segmentation[index])
doc.start_edu = 0
doc.end_edu = len(doc.edus)
def segment(self, doc):
doc.edu_word_segmentation = []
doc.cuts = []
doc.edus = []
for sentence in doc.sentences:
self.segment_sentence(sentence)
if self.global_features:
init_edu_word_segmentation = doc.edu_word_segmentation
doc.edu_word_segmentation = []
doc.cuts = []
doc.edus = []
for sentence in doc.sentences:
self.segment_sentence(
sentence,
input_edu_segmentation=init_edu_word_segmentation[
sentence.sent_id])
# if init_edu_word_segmentation[sentence.sent_id] != doc.edu_word_segmentation[sentence.sent_id]:
# print sentence.sent_id
# print 'initial segmentation', init_edu_word_segmentation[sentence.sent_id]
# print 'new segmentation', doc.edu_word_segmentation[sentence.sent_id]
# print
doc.start_edu = 0
doc.end_edu = len(doc.edus)
class CRFSegmenter:
def __init__(self, _name = 'crf_segmenter', verbose = False, global_features = False):
self.name = _name
self.verbose = verbose
self.feature_writer = SegmenterFeatureWriter()
self.global_features = global_features
self.add_classifiers()
def add_classifiers(self):
self.classifiers = []
classifier1 = CRFClassifier(name= self.name,
model_type = 'segmenter',
model_path = paths.SEGMENTER_MODEL_PATH,
model_file = 'seg.crfsuite',
verbose = self.verbose)
self.add_classifier(classifier1, 'classifier1')
if self.global_features:
classifier2 = CRFClassifier(name= self.name + "_global_features",
model_type = 'segmenter',
model_path = paths.SEGMENTER_MODEL_PATH,
model_file = 'seg_global_features.crfsuite',
verbose = self.verbose)
self.add_classifier(classifier2, 'classifier2')
# classifier1 = MalletCRFClassifier(name= self.name,
# model_type = 'segmenter',
# model_path = paths.CRF_SEGMENTER_MODEL_PATH,
# model_file = 'seg.mallet',
# verbose = self.verbose)
# self.add_classifier(classifier1, 'classifier1')
#
# if self.global_features:
# classifier2 = MalletCRFClassifier(name= self.name + "_global_features",
# model_type = 'segmenter',
# model_path = paths.CRF_SEGMENTER_MODEL_PATH,
# model_file = 'seg_global_features.mallet',
# verbose = self.verbose)
#
#
#
# self.add_classifier(classifier2, 'classifier2')
#
def add_classifier(self, classifier, name):
if name == 'classifier1':
self.classifier = classifier
elif name == 'classifier2':
self.global_features_classifier = classifier
else:
raise Exception('Unknown classifier name', name)
print 'Added classifier', name, 'to segmenter', self.name
self.classifiers.append(classifier)
def unload(self):
for classifier in self.classifiers:
classifier.unload()
def write_features(self, sentence, input_edu_segmentation = None):
features = []
#print sentence.sent_id
# edu_segmentation = sentence.doc.edu_word_segmentation[sentence.sent_id]
# print edu_segmentation
for i in range(len(sentence.tokens) - 1):
token0 = None if i == 0 else sentence.tokens[i - 1]
token1 = sentence.tokens[i]
token2 = sentence.tokens[i + 1]
token3 = None if i == len(sentence.tokens) - 2 else sentence.tokens[i + 2]
edu_segmentation = input_edu_segmentation
# print edu_segmentation
inst_features = self.feature_writer.write_features([token0, token1, token2, token3], edu_segmentation)
feature_str = '\t'.join(list(inst_features))
# print feature_str
features.append('0\t%s' % feature_str)
# feature_str = ' '.join(list(inst_features))
## print feature_str
# features.append('%s 0' % feature_str)
return features
def find_neighbouring_boundary(self, token_id, edu_segmentation, direction = 'L'):
if direction == 'R':
for j in range(len(edu_segmentation)):
# print edu_segmentation[j]
if edu_segmentation[j][1] > token_id:
return edu_segmentation[j][1]
else:
for j in range(len(edu_segmentation) - 1, -1, -1):
# print edu_segmentation[j]
if edu_segmentation[j][0] < token_id - 1:
return edu_segmentation[j][0]
def segment_sentence(self, sentence, input_edu_segmentation = None):
if len(sentence.tokens) == 1:
edus = [[sentence.tokens[0].word, sentence.raw_text[-3 : ]]]
sentence.doc.cuts.append((len(sentence.doc.edus), len(sentence.doc.edus) + len(edus)))
sentence.start_edu = len(sentence.doc.edus)
sentence.end_edu = len(sentence.doc.edus) + len(edus)
sentence.doc.edu_word_segmentation.append([(0, 1)])
sentence.doc.edus.extend(edus)
return
# print sentence.raw_text
self.feature_writer.cached_subtrees = {}
if input_edu_segmentation:
# print input_edu_segmentation
offset2neighbouring_boundaries = {}
for (j, (start_word, end_word)) in enumerate(input_edu_segmentation):
for offset in range(start_word, end_word):
if offset == start_word:
l_boundary = input_edu_segmentation[j - 1][0] if j > 0 else None
else:
l_boundary = start_word
if offset == end_word - 1:
r_boundary = input_edu_segmentation[j + 1][1] if (j < len(input_edu_segmentation) - 1) else None
else:
r_boundary = end_word
# l_boundary1 = self.find_neighbouring_boundary(offset + 1, input_edu_segmentation, 'L')
# r_boundary1 = self.find_neighbouring_boundary(offset + 1, input_edu_segmentation, 'R')
##
## print offset, l_boundary, r_boundary, l_boundary1, r_boundary1
# assert l_boundary1 == l_boundary and r_boundary1 == r_boundary
offset2neighbouring_boundaries[offset] = (l_boundary, r_boundary)
#
# for offset in offset2neighbouring_boundaries:
# print 'sentence', j, offset, offset2neighbouring_boundaries[offset]
else:
offset2neighbouring_boundaries = None
features = self.write_features(sentence, offset2neighbouring_boundaries)
# print sentence.raw_text
# print len(features)
# for feat in features:
# print feat
# print
if input_edu_segmentation:
seq_prob, predictions = self.global_features_classifier.classify(features)
else:
seq_prob, predictions = self.classifier.classify(features)
edus = []
edu_word_segmentations = []
start = 0
for i in range(len(predictions)):
pred = int(predictions[i][0])
if pred == 1:
# print i, pred
edu_word_segmentations.append((start, i + 1))
start = i + 1
edu_word_segmentations.append((start, len(sentence.tokens)))
for (start_word, end_word) in edu_word_segmentations:
edu = []
for j in range(start_word, end_word):
edu.extend(utils.utils.unescape_penn_special_word(sentence.tokens[j].word).split(' '))
if end_word == len(sentence.tokens):
# print sentence.raw_text
edu.append(sentence.raw_text[-3 : ])
edus.append(edu)
sentence.doc.cuts.append((len(sentence.doc.edus), len(sentence.doc.edus) + len(edus)))
sentence.start_edu = len(sentence.doc.edus)
sentence.end_edu = len(sentence.doc.edus) + len(edus)
sentence.doc.edu_word_segmentation.append(edu_word_segmentations)
sentence.doc.edus.extend(edus)
# print sentence.doc.edu_word_segmentation[-1]
def segment_permutation(self, doc, canonical_doc):
assert len(doc.sentences) == len(canonical_doc.sentences)
doc.edu_word_segmentation = []
doc.cuts = []
doc.edus = []
sentence_order = []
for sent in doc.sentences:
index = 0
# print sent.sent_id, sent.raw_text
while index < len(canonical_doc.sentences):
# print 'canonical', index, canonical_doc.sentences[index].raw_text
if canonical_doc.sentences[index].raw_text == sent.raw_text and index not in sentence_order:
# print index
# print canonical_doc.sentences[index].raw_text
# print sent.raw_text
# print
break
index += 1
# if index == len(canonical_doc.sentences):
# print sent.sent_id, sent.raw_text
# print canonical_doc.sentences[index - 1].raw_text
# print
# print
# print sent.sent_id, index
sentence_order.append(index)
# print sentence_order
# print sorted(sentence_order)
assert sorted(sentence_order) == range(len(doc.sentences))
for (i, index) in enumerate(sentence_order):
sentence = doc.sentences[i]
sentence.set_unlexicalized_tree(canonical_doc.sentences[index].unlexicalized_parse_tree)
sentence.set_lexicalized_tree(canonical_doc.sentences[index].parse_tree)
sentence.tokens = []
for te in canonical_doc.sentences[index].tokens:
token = Token(te.word, te.id, sentence)
token.lemma = te.lemma
sentence.add_token(token)
# print i
# print sentence.raw_text
# print canonical_doc.sentences[index].raw_text
# print
(canonical_start_edu, canonical_end_edu) = canonical_doc.cuts[index]
edus = canonical_doc.edus[canonical_start_edu : canonical_end_edu]
sentence.start_edu = len(sentence.doc.edus)
sentence.end_edu = len(sentence.doc.edus) + len(edus)
doc.cuts.append((len(sentence.doc.edus), len(sentence.doc.edus) + len(edus)))
doc.edus.extend(edus)
doc.edu_word_segmentation.append(canonical_doc.edu_word_segmentation[index])
doc.start_edu = 0
doc.end_edu = len(doc.edus)
def segment(self, doc):
doc.edu_word_segmentation = []
doc.cuts = []
doc.edus = []
for sentence in doc.sentences:
self.segment_sentence(sentence)
if self.global_features:
init_edu_word_segmentation = doc.edu_word_segmentation
# print init_edu_word_segmentation
doc.edu_word_segmentation = []
doc.cuts = []
doc.edus = []
for sentence in doc.sentences:
self.segment_sentence(sentence, input_edu_segmentation = init_edu_word_segmentation[sentence.sent_id])
# if init_edu_word_segmentation[sentence.sent_id] != doc.edu_word_segmentation[sentence.sent_id]:
# print sentence.sent_id
# print 'initial segmentation', init_edu_word_segmentation[sentence.sent_id]
# print 'new segmentation', doc.edu_word_segmentation[sentence.sent_id]
# print
doc.start_edu = 0
doc.end_edu = len(doc.edus)
