def TrainSenseTagger(Pcfg,CFDist):
logger.info("Training unigram tagger:")
SenseUnigramTagger = UnigramTagger(TAG='SENSE',TEXT='STEM')
#SenseUnigramTagger.train(taggedData)
SenseUnigramTagger._freqdist = invertConditionalFreqDist(CFDist)
SenseDefaultTagger = DefaultTagger('APPEAR', TAG='SENSE',TEXT='STEM')
backoff = BackoffTagger([SenseUnigramTagger,SenseDefaultTagger], TAG='SENSE',TEXT='STEM')
return backoff
def TrainSenseTagger(Pcfg, CFDist):
logger.info("Training unigram tagger:")
SenseUnigramTagger = UnigramTagger(TAG="SENSE", TEXT="STEM")
# SenseUnigramTagger.train(taggedData)
SenseUnigramTagger._freqdist = invertConditionalFreqDist(CFDist)
SenseDefaultTagger = DefaultTagger("APPEAR", TAG="SENSE", TEXT="STEM")
backoff = BackoffTagger([SenseUnigramTagger, SenseDefaultTagger], TAG="SENSE", TEXT="STEM")
return backoff
def test(
numFiles=100,
max_rules=200,
min_score=2,
ruleFile="dump.rules",
errorOutput="errors.out",
ruleOutput="rules.out",
randomize=False,
train=0.8,
trace=3,
):
NN_CD_tagger = RegexpTagger([(r"^[0-9]+(.[0-9]+)?$", "CD"), (r".*", "NN")], TAG="POS")
# train is the proportion of data used in training; the rest is reserved
# for testing.
print "Loading tagged data..."
taggedData = getWSJTokens(numFiles, randomize)
trainCutoff = int(len(taggedData) * train)
trainingData = Token(SUBTOKENS=taggedData[0:trainCutoff])
goldData = Token(SUBTOKENS=taggedData[trainCutoff:])
testingData = goldData.exclude("POS")
# Unigram tagger
print "Training unigram tagger:",
u = UnigramTagger(TAG="POS")
u.train(trainingData)
backoff = BackoffTagger([u, NN_CD_tagger], TAG="POS")
print ("[accuracy: %f]" % tagger_accuracy(backoff, [goldData]))
# Brill tagger
templates = [
SymmetricProximateTokensTemplate(ProximateTagsRule, (1, 1)),
SymmetricProximateTokensTemplate(ProximateTagsRule, (2, 2)),
SymmetricProximateTokensTemplate(ProximateTagsRule, (1, 2)),
SymmetricProximateTokensTemplate(ProximateTagsRule, (1, 3)),
# SymmetricProximateTokensTemplate(ProximateWordsRule, (1,1)),
# SymmetricProximateTokensTemplate(ProximateWordsRule, (2,2)),
# SymmetricProximateTokensTemplate(ProximateWordsRule, (1,2)),
# SymmetricProximateTokensTemplate(ProximateWordsRule, (1,3)),
ProximateTokensTemplate(ProximateTagsRule, (-1, -1), (1, 1)),
# ProximateTokensTemplate(ProximateWordsRule, (-1, -1), (1,1)),
]
# trainer = FastBrillTaggerTrainer(backoff, templates, trace, TAG='POS')
trainer = BrillTaggerTrainer(backoff, templates, trace, TAG="POS")
b = trainer.train(trainingData, max_rules, min_score)
print
print ("Brill accuracy: %f" % tagger_accuracy(b, [goldData]))
print ("\nRules: ")
printRules = file(ruleOutput, "w")
for rule in b.rules():
print (str(rule))
printRules.write(str(rule) + "\n\n")
# b.saveRules(ruleFile)
b.tag(testingData)
el = errorList(goldData, testingData)
errorFile = file(errorOutput, "w")
for e in el:
errorFile.write(e + "\n\n")
errorFile.close()
print ("Done.")
return b
def test(numFiles=100,
max_rules=200,
min_score=2,
ruleFile="dump.rules",
errorOutput="errors.out",
ruleOutput="rules.out",
randomize=False,
train=.8,
trace=3):
NN_CD_tagger = RegexpTagger([(r'^[0-9]+(.[0-9]+)?$', 'CD'), (r'.*', 'NN')],
TAG='POS')
# train is the proportion of data used in training; the rest is reserved
# for testing.
print "Loading tagged data..."
taggedData = getWSJTokens(numFiles, randomize)
trainCutoff = int(len(taggedData) * train)
trainingData = Token(SUBTOKENS=taggedData[0:trainCutoff])
goldData = Token(SUBTOKENS=taggedData[trainCutoff:])
testingData = goldData.exclude('POS')
# Unigram tagger
print "Training unigram tagger:",
u = UnigramTagger(TAG='POS')
u.train(trainingData)
backoff = BackoffTagger([u, NN_CD_tagger], TAG='POS')
print("[accuracy: %f]" % tagger_accuracy(backoff, [goldData]))
# Brill tagger
templates = [
SymmetricProximateTokensTemplate(ProximateTagsRule, (1, 1)),
SymmetricProximateTokensTemplate(ProximateTagsRule, (2, 2)),
SymmetricProximateTokensTemplate(ProximateTagsRule, (1, 2)),
SymmetricProximateTokensTemplate(ProximateTagsRule, (1, 3)),
# SymmetricProximateTokensTemplate(ProximateWordsRule, (1,1)),
# SymmetricProximateTokensTemplate(ProximateWordsRule, (2,2)),
# SymmetricProximateTokensTemplate(ProximateWordsRule, (1,2)),
# SymmetricProximateTokensTemplate(ProximateWordsRule, (1,3)),
ProximateTokensTemplate(ProximateTagsRule, (-1, -1), (1, 1)),
# ProximateTokensTemplate(ProximateWordsRule, (-1, -1), (1,1)),
]
#trainer = FastBrillTaggerTrainer(backoff, templates, trace, TAG='POS')
trainer = BrillTaggerTrainer(backoff, templates, trace, TAG='POS')
b = trainer.train(trainingData, max_rules, min_score)
print
print("Brill accuracy: %f" % tagger_accuracy(b, [goldData]))
print("\nRules: ")
printRules = file(ruleOutput, 'w')
for rule in b.rules():
print(str(rule))
printRules.write(str(rule) + "\n\n")
#b.saveRules(ruleFile)
b.tag(testingData)
el = errorList(goldData, testingData)
errorFile = file(errorOutput, 'w')
for e in el:
errorFile.write(e + "\n\n")
errorFile.close()
print("Done.")
return b
