def train(self, tagged_corpus, verbose=False):
"""
Train C{tag.Affix} using the given training data. If this
method is called multiple times, then the training data will be
combined.
@param tagged_corpus: A tagged corpus. Each item should be
a C{list} of tagged tokens, where each consists of
C{text} and a C{tag}.
@type tagged_corpus: C{list} or C{iter(list)}
"""
if self.size() != 0:
raise ValueError, 'Tagger is already trained'
token_count = hit_count = 0
fd = ConditionalFreqDist()
for sentence in tagged_corpus:
for (token, tag) in sentence:
token_count += 1
# If token is long enough
if len(token) >= self._minlength:
backoff_tag = self._backoff_tag_one(token)
if tag != backoff_tag:
# get the affix and record it
affix = self._get_affix(token)
hit_count += 1
fd[affix].inc(tag)
for affix in fd.conditions():
best_tag = fd[affix].max()
if fd[affix].count(best_tag) > self._cutoff:
self._model[affix] = best_tag
# generate stats
if verbose:
size = len(self._model)
backoff = 100 - (hit_count * 100.0)/ token_count
pruning = 100 - (size * 100.0) / len(fd.conditions())
print "[Trained Affix tagger:",
print "size=%d, backoff=%.2f%%, pruning=%.2f%%]" % (
size, backoff, pruning)
def train(self, tagged_corpus, verbose=False):
"""
Train C{tag.Unigram} using the given training data.
@param tagged_corpus: A tagged corpus. Each item should be
a C{list} of tagged tokens, where each consists of
C{text} and a C{tag}.
@type tagged_corpus: C{list} or C{iter(list)}
"""
if self.size() != 0:
raise ValueError, 'Tagger is already trained'
token_count = hit_count = 0
fd = ConditionalFreqDist()
if isinstance(tagged_corpus, list) and isinstance(tagged_corpus[0], tuple):
tagged_corpus = [tagged_corpus]
for sentence in tagged_corpus:
for (token, tag) in sentence:
token_count += 1
backoff_tag = self._backoff_tag_one(token)
if tag != backoff_tag:
hit_count += 1
fd[token].inc(tag)
for token in fd.conditions():
best_tag = fd[token].max()
if fd[token].count(best_tag) > self._cutoff:
self._model[token] = best_tag
# generate stats
if verbose:
size = len(self._model)
backoff = 100 - (hit_count * 100.0)/ token_count
pruning = 100 - (size * 100.0) / len(fd.conditions())
print "[Trained Unigram tagger:",
print "size=%d, backoff=%.2f%%, pruning=%.2f%%]" % (
size, backoff, pruning)
def train(self, tagged_corpus, verbose=False):
"""
Train this C{tagger.Ngram} using the given training data.
@param tagged_corpus: A tagged corpus. Each item should be
a C{list} of tagged tokens, where each consists of
C{text} and a C{tag}.
@type tagged_corpus: C{list} or C{iter(list)}
"""
if self.size() != 0:
raise ValueError, 'Tagger is already trained'
token_count = hit_count = 0
fd = ConditionalFreqDist()
for sentence in tagged_corpus:
self._history.clear()
for (token, tag) in sentence:
token_count += 1
history = self._history.get()
backoff_tag = self._backoff_tag_one(token, history)
if tag != backoff_tag:
hit_count += 1
fd[(history, token)].inc(tag)
self._history.enqueue(tag)
for context in fd.conditions():
best_tag = fd[context].max()
if fd[context].count(best_tag) > self._cutoff:
self._model[context] = best_tag
# generate stats
if verbose:
size = len(self._model)
backoff = 100 - (hit_count * 100.0)/ token_count
pruning = 100 - (size * 100.0) / len(fd.conditions())
print "[Trained %d-gram tagger:" % self._n,
print "size=%d, backoff=%.2f%%, pruning=%.2f%%]" % (
size, backoff, pruning)
