def main_esaren():
ngram = 3
lrpad = u' '
verbose=False
corpus_mix='l'
lm = MyLM(n=ngram, verbose=verbose, lpad=lrpad, rpad=lrpad,
smoothing='Laplace', laplace_gama=0.1,
corpus_mix=corpus_mix)
from corpora.corpuslib import Train, Test, Accuracy
#train = corpuslib.Train()
train = Train()
corpus = train.get_corpus()
for item in corpus:
lm.add_doc(doc_id=item[0], doc_terms=readfile(item[1]))
#a = corpuslib.Accuracy()
#t = corpuslib.Test(root='', langid=lm, accuracy=a)
a = Accuracy()
t = Test(root='', langid=lm, accuracy=a)
t.start()
a.evaluate()
def train(self):
train = Train()
corpus = train.get_corpus()
for item in corpus:
self.lm.add_doc(doc_id=item[0], doc_terms=self._readfile(item[1]))
def train(self, root=''):
""" Trains our Language Model.
:param root: Path to training data.
"""
self.trainer = Train(root=root)
corpus = self.trainer.get_corpus()
# Show loaded Languages
#print 'Lang Set: ' + ' '.join(train.get_lang_set())
for item in corpus:
self.lm.add_doc(doc_id=item[0], doc_terms=self._readfile(item[1]))
# Save training timestamp
self.training_timestamp = self.trainer.get_last_modified()
class LangID:
""" Language Identification Class
"""
def __init__(self, unk=False):
# Shall we mark some text as unk,
# if top languages are too close?
self.unk = unk
self.min_karbasa = 0.08
# LM Parameters
ngram = 3
lrpad = u' '
verbose = False
corpus_mix = 'l'
self.lm = LM(n=ngram, verbose=verbose, lpad=lrpad, rpad=lrpad,
smoothing='Laplace', laplace_gama=0.1,
corpus_mix=corpus_mix)
self.trainer = None
self.training_timestamp = 0
@classmethod
def _readfile(cls, filename):
""" Reads a file a utf-8 file,
and retuns character tokens.
:param filename: Name of file to be read.
"""
f = codecs.open(filename, encoding='utf-8')
filedata = f.read()
f.close()
tokenz = LM.tokenize(filedata, mode='c')
#print tokenz
return tokenz
def train(self, root=''):
""" Trains our Language Model.
:param root: Path to training data.
"""
self.trainer = Train(root=root)
corpus = self.trainer.get_corpus()
# Show loaded Languages
#print 'Lang Set: ' + ' '.join(train.get_lang_set())
for item in corpus:
self.lm.add_doc(doc_id=item[0], doc_terms=self._readfile(item[1]))
# Save training timestamp
self.training_timestamp = self.trainer.get_last_modified()
def is_training_modified(self):
""" Returns `True` if training data
was modified since last training.
Returns `False` otherwise,
or if using builtin training data.
"""
last_modified = self.trainer.get_last_modified()
if last_modified > self.training_timestamp:
return True
else:
return False
def was_training_modified(self):
""" For Grammar Nazis,
just alias for is_training_modified()
"""
return self.is_training_modified()
def add_training_sample(self, text=u'', lang=''):
""" Initial step for adding new sample to training data.
You need to call `save_training_samples()` afterwards.
:param text: Sample text to be added.
:param lang: Language label for the input text.
"""
self.trainer.add(text=text, lang=lang)
def save_training_samples(self, domain='', filename=''):
""" Saves data previously added via add_training_sample().
Data saved in folder specified by Train.get_corpus_path().
:param domain: Name for domain folder.
If not set, current timestamp will be used.
:param filename: Name for file to save data in.
If not set, file.txt will be used.
Check the README file for more information about Domains.
"""
self.trainer.save(domain=domain, filename=filename)
def get_lang_set(self):
""" Returns a list of languages in training data.
"""
return self.trainer.get_lang_set()
def classify(self, text=u''):
""" Predicts the Language of a given text.
:param text: Unicode text to be classified.
"""
text = self.lm.normalize(text)
tokenz = LM.tokenize(text, mode='c')
result = self.lm.calculate(doc_terms=tokenz)
#print 'Karbasa:', self.karbasa(result)
if self.unk and self.lm.karbasa(result) < self.min_karbasa:
lang = 'unk'
else:
lang = result['calc_id']
return lang