forked from roshan-research/hazm
-
Notifications
You must be signed in to change notification settings - Fork 1
/
data.py
executable file
·247 lines (189 loc) · 10.4 KB
/
data.py
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
# coding: utf-8
from __future__ import print_function, unicode_literals
import codecs, subprocess, random
from collections import Counter
from itertools import islice
from nltk.tag import untag
from sklearn.model_selection import train_test_split
from hazm import *
from hazm.Chunker import tree2brackets
from hazm.PeykareReader import coarse_pos_e as peykare_coarse_pos_e
from hazm.DadeganReader import coarse_pos_e as dadegan_coarse_pos_e
def create_words_file(dic_file='resources/persian.dic', output='hazm/data/words.dat'):
""" prepares list of persian word words from [Virastyar](https://sourceforge.net/projects/virastyar/) dic file. """
dic_words = [line.strip().replace(', ', ',').split('\t') for line in codecs.open(dic_file, encoding='utf-8') if len(line.strip().split('\t')) == 3]
dic_words = filter(lambda item: not item[2].startswith('V') and 'NEG' not in item[2], dic_words)
dic_words = ['\t'.join(item) for item in sorted(dic_words, key=lambda item: item[0])]
print(*dic_words, sep='\n', file=codecs.open(output, 'w', 'utf-8'))
print(output, 'created')
def evaluate_lemmatizer(conll_file='resources/train.conll', peykare_root='corpora/peykare'):
lemmatizer = Lemmatizer()
errors = []
with codecs.open('resources/lemmatizer_errors.txt', 'w', 'utf8') as output:
dadegan = DadeganReader(conll_file)
for tree in dadegan.trees():
for node in tree.nodelist[1:]:
word, lemma, pos = node['word'], node['lemma'], node['mtag']
if lemmatizer.lemmatize(word, pos) != lemma:
errors.append((word, lemma, pos, lemmatizer.lemmatize(word, pos)))
print(len(errors), 'errors', file=output)
counter = Counter(errors)
for item, count in sorted(counter.items(), key=lambda t: t[1], reverse=True):
print(count, *item, file=output)
missed = []
with codecs.open('resources/lemmatizer_missed.txt', 'w', 'utf8') as output:
peykare = PeykareReader(peykare_root)
for sentence in peykare.sents():
for word in sentence:
if word[1] == 'V':
if word[0] == lemmatizer.lemmatize(word[0]):
missed.append(word[0])
print(len(missed), 'missed', file=output)
counter = Counter(missed)
for item, count in sorted(counter.items(), key=lambda t: t[1], reverse=True):
print(count, item, file=output)
def evaluate_normalizer(tnews_root='corpora/tnews'):
tnews = TNewsReader(root=tnews_root)
normalizer = Normalizer(persian_style=False, persian_numbers=False, remove_diacritics=False, token_based=False, affix_spacing=True)
token_normalizer = Normalizer(persian_style=False, persian_numbers=False, remove_diacritics=False, token_based=True, affix_spacing=False)
with codecs.open('resources/normalized.txt', 'w', 'utf8') as output1, codecs.open('resources/normalized_token_based.txt', 'w', 'utf8') as output2:
random.seed(0)
for text in tnews.texts():
if random.randint(0, 100) != 0:
continue
for sentence in sent_tokenize(text):
print(normalizer.normalize(sentence), '\n', file=output1)
print(token_normalizer.normalize(sentence), '\n', file=output2)
def evaluate_informal_normalizer(sentipars_root='corpora/sentipers'):
sentipers = SentiPersReader(root=sentipars_root)
normalizer = Normalizer()
informal_normalizer = InformalNormalizer()
output = codecs.open('resources/normalized.txt', 'w', 'utf8')
for comments in sentipers.comments():
for comment in comments:
for sentence in comment:
print(normalizer.normalize(sentence), file=output)
sents = informal_normalizer.normalize(sentence)
sents = [[word[0] for word in sent] for sent in sents]
sents = [' '.join(sent) for sent in sents]
text = '\n'.join(sents)
text = normalizer.normalize(text)
print(text, file=output)
print(file=output)
def evaluate_chunker(treebank_root='corpora/treebank'):
treebank = TreebankReader(treebank_root, join_clitics=True, join_verb_parts=True)
chunker = Chunker()
chunked_trees = list(treebank.chunked_trees())
print(chunker.evaluate(chunked_trees))
output = codecs.open('resources/chunker_errors.txt', 'w', 'utf8')
for sentence, gold in zip(treebank.sents(), chunked_trees):
chunked = chunker.parse(sentence)
if chunked != gold:
print(tree2brackets(chunked), file=output)
print(tree2brackets(gold), file=output)
print(file=output)
def train_postagger(peykare_root='corpora/peykare', model_file='resources/postagger.model', test_size=.1, sents_limit=None, pos_map=peykare_coarse_pos_e):
tagger = POSTagger(type='crf', algo='rprop', compact=True, patterns=[
'*',
'u:wll=%x[-2,0]',
'u:wl=%x[-1,0]',
'u:w=%x[0,0]',
'u:wr=%x[1,0]',
'u:wrr=%x[2,0]',
# 'u:w2l=%x[-1,0]/%x[0,0]',
# 'u:w2r=%x[0,0]/%x[1,0]',
'*:p1=%m[0,0,"^.?"]',
'*:p2=%m[0,0,"^.?.?"]',
'*:p3=%m[0,0,"^.?.?.?"]',
'*:s1=%m[0,0,".?$"]',
'*:s2=%m[0,0,".?.?$"]',
'*:s3=%m[0,0,".?.?.?$"]',
'*:p?l=%t[-1,0,"\p"]',
'*:p?=%t[0,0,"\p"]',
'*:p?r=%t[1,0,"\p"]',
'*:p?a=%t[0,0,"^\p*$"]',
'*:n?l=%t[-1,0,"\d"]',
'*:n?=%t[0,0,"\d"]',
'*:n?r=%t[1,0,"\d"]',
'*:n?a=%t[0,0,"^\d*$"]',
])
peykare = PeykareReader(peykare_root, pos_map=pos_map)
train_sents, test_sents = train_test_split(list(islice(peykare.sents(), sents_limit)), test_size=test_size, random_state=0)
tagger.train(train_sents)
tagger.save_model(model_file)
print(tagger.evaluate(test_sents))
def train_chunker(train_file='corpora/train.conll', dev_file='corpora/dev.conll', test_file='corpora/test.conll', model_file='resources/chunker.model'):
tagger = POSTagger(model='resources/postagger.model')
chunker = Chunker(type='crf', algo='l-bfgs', compact=True, patterns=[
'*',
'u:wll=%x[-2,0]',
'u:wl=%x[-1,0]',
'u:w=%x[0,0]',
'u:wr=%x[1,0]',
'u:wrr=%x[2,0]',
'*:tll=%x[-2,1]',
'*:tl=%x[-1,1]',
'*:t=%x[0,1]',
'*:tr=%x[1,1]',
'*:trr=%x[2,1]',
])
def retag_trees(trees, sents):
for tree, sentence in zip(trees, tagger.tag_sents(map(untag, sents))):
for (n, word) in zip(tree.treepositions('leaves'), sentence):
tree[n] = word
train, test = DadeganReader(train_file), DadeganReader(test_file)
train_trees = list(train.chunked_trees())
retag_trees(train_trees, train.sents())
chunker.train(train_trees)
chunker.save_model(model_file)
test_trees = list(test.chunked_trees())
retag_trees(test_trees, test.sents())
print(chunker.evaluate(test_trees))
def train_maltparser(train_file='corpora/train.conll', dev_file='corpora/dev.conll', test_file='corpora/test.conll', model_file='langModel.mco', path_to_jar='resources/malt.jar', options_file='resources/malt-options.xml', features_file='resources/malt-features.xml', memory_min='-Xms7g', memory_max='-Xmx8g'):
lemmatizer, tagger = Lemmatizer(), POSTagger(model='resources/postagger.model')
train, test = DadeganReader(train_file), DadeganReader(test_file)
train_data = train_file +'.data'
with codecs.open(train_data, 'w', 'utf8') as output:
for tree, sentence in zip(train.trees(), tagger.tag_sents(map(untag, train.sents()))):
for i, (node, word) in enumerate(zip(list(tree.nodes.values())[1:], sentence), start=1):
node['mtag'] = word[1]
node['lemma'] = lemmatizer.lemmatize(node['word'], node['mtag'])
print(i, node['word'].replace(' ', '_'), node['lemma'].replace(' ', '_'), node['mtag'], node['mtag'], '_', node['head'], node['rel'], '_', '_', sep='\t', file=output)
print(file=output)
subprocess.Popen(['java', memory_min, memory_max, '-jar', path_to_jar, '-w', 'resources', '-c', model_file, '-i', train_data, '-f', options_file, '-F', features_file, '-m', 'learn']).wait()
# evaluation
parser = MaltParser(tagger=tagger, lemmatizer=lemmatizer, model_file=model_file)
parsed_trees = parser.parse_sents(map(untag, test.sents()))
test_data, test_results = test_file +'.data', test_file +'.results'
print('\n'.join([tree.to_conll(10) for tree in test.trees()]).strip(), file=codecs.open(test_data, 'w', 'utf8'))
print('\n'.join([tree.to_conll(10) for tree in parsed_trees]).strip(), file=codecs.open(test_results, 'w', 'utf8'))
subprocess.Popen(['java', '-jar', 'resources/MaltEval.jar', '-g', test_data, '-s', test_results]).wait()
def train_turboparser(train_file='corpora/train.conll', dev_file='corpora/dev.conll', test_file='corpora/test.conll', model_file='resources/turboparser.model'):
lemmatizer, tagger = Lemmatizer(), POSTagger(model='resources/postagger.model')
train, test = DadeganReader(train_file), DadeganReader(test_file)
train_data = train_file +'.data'
with codecs.open(train_data, 'w', 'utf8') as output:
for tree, sentence in zip(train.trees(), tagger.tag_sents(map(untag, train.sents()))):
for i, (node, word) in enumerate(zip(list(tree.nodes.values())[1:], sentence), start=1):
node['mtag'] = word[1]
node['lemma'] = lemmatizer.lemmatize(node['word'], node['mtag'])
print(i, node['word'].replace(' ', '_'), node['lemma'].replace(' ', '_'), node['mtag'], node['mtag'], '_', node['head'], node['rel'], '_', '_', sep='\t', file=output)
print(file=output)
subprocess.Popen(['./resources/TurboParser', '--train', '--file_train='+train_data, '--file_model='+model_file, '--logtostderr']).wait()
# evaluation
parser = TurboParser(tagger=tagger, lemmatizer=lemmatizer, model_file=model_file)
parsed_trees = parser.parse_sents(map(untag, test.sents()))
test_data, test_results = test_file +'.data', test_file +'.results'
print('\n'.join([tree.to_conll(10) for tree in test.trees()]).strip(), file=codecs.open(test_data, 'w', 'utf8'))
print('\n'.join([tree.to_conll(10) for tree in parsed_trees]).strip(), file=codecs.open(test_results, 'w', 'utf8'))
subprocess.Popen(['java', '-jar', 'resources/MaltEval.jar', '-g', test_data, '-s', test_results, '--pattern', '0.####', '--Metric', 'LAS;UAS']).wait()
def train_stanford_postagger(peykare_root='corpora/peykare', path_to_model='resources/persian.tagger', path_to_jar='resources/stanford-postagger.jar', properties_file='resources/stanford-postagger.props', memory_min='-Xms1g', memory_max='-Xmx6g', test_size=.1, pos_map=peykare_coarse_pos_e):
peykare = PeykareReader(peykare_root, pos_map=pos_map)
train_file = 'resources/tagger_train_data.txt'
train, test = train_test_split(list(peykare.sents()), test_size=test_size, random_state=0)
output = codecs.open(train_file, 'w', 'utf8')
for sentence in train:
print(*(map(lambda w: '/'.join(w).replace(' ', '_'), sentence)), file=output)
subprocess.Popen(['java', memory_min, memory_max, '-classpath', path_to_jar, 'edu.stanford.nlp.tagger.maxent.MaxentTagger', '-prop', properties_file, '-model', path_to_model, '-trainFile', train_file, '-tagSeparator', '/', '-search', 'owlqn2']).wait()
tagger = StanfordPOSTagger(path_to_jar=path_to_jar, path_to_model=path_to_model)
print(tagger.evaluate(test))