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
|
from nltk.classify import iis
import yaml
import os
class SequentialClassifier(object):
def __init__(self, left=2, right=0):
#left = look back
#right = look forward
self._model = []
self._left = left
self._right = right
self._leftcontext = [None] * (left)
self._history = self._leftcontext
self._rightcontext = [None] * (right)
def size(self):
return len(self._model)
def classify(self, featuresets):
if self.size() == 0:
raise ValueError, 'Tagger is not trained'
for i, featureset in enumerate(featuresets):
#if i >= self._left:
#self._leftcontext = sequence[i-self._left : i]
#else:
#self._leftcontext = sequence[:i]
self._rightcontext = sequence[i+1 : i+1+self._right]
label = self.classify_one(featureset)
featureset['label'] = label
del self._leftcontext[0]
self._leftcontext.append(featureset)
yield label
def classify_one(self, featureset):
"""
Classify a single featureset.
"""
return self._model([featureset][0])
def contexts(self, sequence):
"""
Build a generator of triples (left context, item, right context).
@param sequence: Input sequence
@type sequence: C{list}
@rtype: C{generator} of triples (left_context, token, right_contex)
"""
for i in range(len(sequence)):
if i >= self._left:
left_context = sequence[i - self._left:i]
else:
left_context = sequence[:i]
right_context = sequence[i+1 : i+1+self._right]
yield (left_context, sequence[i], right_context)
def detect_features(self, context):
from string import join
left_context, item, right_context = context
features = {}
token = item['token']
features['cur_token(%s)' % token] = True
features['is_title'] = token.istitle()
features['is_digit'] = token.isdigit()
features['is_upper'] = token.isupper()
features['POS(%s)' % item['POS']] = True
if left_context == []:
features['initword'] = True
else:
left_labels = join([item['label'] for item in left_context], '_')
features['left_labels(%s)' % left_labels] = True
return features
def save_features(self, training_data, filename):
stream = open(filename,'w')
yaml.dump_all(training_data, stream)
print "Saving features to %s" % os.path.abspath(filename)
stream.close()
def corpus2training_data(self, training_corpus, model_name='default', save=False):
dict_corpus = tabular2dict(training_corpus, KEYS)
contexts = self.contexts(dict_corpus)
print "Detecting features"
training_data = [(self.detect_features(c), c[1]['label']) for c in contexts]
if save:
feature_file = model_name + '.yaml'
self.save_features(training_data, feature_file)
else:
return training_data
def train(self, training_corpus, classifier=iis):
"""
Train a classifier.
"""
if self.size() != 0:
raise ValueError, 'Classifier is already trained'
training_data = self.corpus2training_data(training_corpus)
print "Training classifier"
self._model = iis(training_data)
def tabular2dict(tabular, keys):
"""
Utility function to turn tabular format CONLL data into a
sequence of dictionaries.
@param tabular: tabular input
@param keys: a dictionary that maps field positions into feature names
@rtype: C{list} of featuresets
"""
tokendicts = []
lines = tabular.splitlines()
for line in lines:
line = line.strip()
line = line.split()
if line:
tokendict = {}
for i in range(len(line)):
key = keys[i]
tokendict [key] = line[i]
tokendicts.append(tokendict )
return tokendicts
KEYS = {0: 'token', 1: 'POS', 2: 'label'}
def demo():
tabtrain = \
"""Het Art O
Hof N B-ORG
van Prep I-ORG
Cassatie N I-ORG
verbrak V O
het Art O
arrest N O
"""
tabtest = \
"""Het Art
Hof N
van Prep
Cassatie N
verbrak V
het Art
arrest N
"""
test = tabular2dict(tabtest, KEYS)
train = tabular2dict(tabtrain, KEYS)
sc = SequentialClassifier(2, 0)
sc.train(tabtrain)
sc.classify(tabtest)
demo()
|
