def generateKaggleSubmission(tagger,outfilename):
with open(outfilename, 'w') as f:
writer = csv.DictWriter(f, fieldnames=['Id', 'Prediction'])
writer.writeheader()
alltags = set()
for i,(words, tags) in enumerate(preproc.conllSeqGenerator(TRAIN_FILE)):
for tag in tags:
alltags.add(tag)
i=0
for words,_ in preproc.conllSeqGenerator(TEST_FILE):
pred_tags = tagger(words,alltags)
if isinstance(pred_tags, tuple):
pred_tags = pred_tags[0]
for tag in pred_tags:
writer.writerow({
'Id': 'test-{}'.format(i),
'Prediction':tag})
i+=1
i=0
for words,_ in preproc.conllSeqGenerator(DEV_FILE):
pred_tags = tagger(words,alltags)
if isinstance(pred_tags, tuple):
pred_tags = pred_tags[0]
for tag in pred_tags:
# print >>outfile, tag
writer.writerow({
'Id': 'dev-{}'.format(i),
'Prediction':tag})
i+=1
def evalTagger(tagger,outfilename,testfile=DEV_FILE):
alltags = set()
for i,(words, tags) in enumerate(preproc.conllSeqGenerator(TRAIN_FILE)):
for tag in tags:
alltags.add(tag)
with open(outfilename,'w') as outfile:
for words,_ in preproc.conllSeqGenerator(testfile):
pred_tags = tagger(words,alltags)
for tag in pred_tags:
print >>outfile, tag
print >>outfile, ""
return scorer.getConfusion(testfile,outfilename) #run the scorer on the prediction file
def evalTagger(tagger,outfilename,testfile=DEV_FILE):
alltags = set()
for i,(words, tags) in enumerate(preproc.conllSeqGenerator(TRAIN_FILE)):
for tag in tags:
alltags.add(tag)
with open(outfilename,'w') as outfile:
for words,_ in preproc.conllSeqGenerator(testfile):
pred_tags = tagger(words,alltags)
for tag in pred_tags:
print >>outfile, tag
print >>outfile, ""
return scorer.getConfusion(testfile,outfilename) #run the scorer on the prediction file
def test_nb_prob_mass ():
probability_masses = defaultdict(float)
allwords = set([])
for words, _ in preproc.conllSeqGenerator(TRAIN_FILE):
for word in words:
allwords.add(word)
for tag in alltags:
total_prob = sum(np.exp(weights_nb[(tag, word)]) for word in allwords)
assert_almost_equals (1.0, total_prob, places=2,
msg="UNEQUAL Expected tag %s to have total prob of 1.0, but instead has %s" %(tag, total_prob))
def evalTagger(tagger,outfilename,testfile=DEV_FILE):
"""Calculate confusion_matrix for a given tagger
Parameters:
tagger -- Function mapping (words, possible_tags) to an optimal
sequence of tags for the words
outfilename -- Filename to write tagger predictions to
testfile -- (optional) Filename containing true labels
Returns:
confusion_matrix -- dict of occurences of (true_label, pred_label)
"""
alltags = set()
for i,(words, tags) in enumerate(preproc.conllSeqGenerator(TRAIN_FILE)):
for tag in tags:
alltags.add(tag)
with open(outfilename,'w') as outfile:
for words,_ in preproc.conllSeqGenerator(testfile):
pred_tags = tagger(words,alltags)
for tag in pred_tags:
print >>outfile, tag
print >>outfile, ""
return scorer.getConfusion(testfile,outfilename) #run the scorer on the prediction file
def evalTagger(tagger, outfilename, testfile=DEV_FILE):
"""Calculate confusion_matrix for a given tagger
Parameters:
tagger -- Function mapping (words, possible_tags) to an optimal
sequence of tags for the words
outfilename -- Filename to write tagger predictions to
testfile -- (optional) Filename containing true labels
Returns:
confusion_matrix -- dict of occurences of (true_label, pred_label)
"""
alltags = set()
for i, (words, tags) in enumerate(preproc.conllSeqGenerator(TRAIN_FILE)):
for tag in tags:
alltags.add(tag)
with open(outfilename, 'w') as outfile:
for words, _ in preproc.conllSeqGenerator(testfile):
pred_tags = tagger(words, alltags)
for tag in pred_tags:
print >> outfile, tag
print >> outfile, ""
return scorer.getConfusion(
testfile, outfilename) #run the scorer on the prediction file
def test_nb_prob_mass():
probability_masses = defaultdict(float)
allwords = set([])
for words, _ in preproc.conllSeqGenerator(TRAIN_FILE):
for word in words:
allwords.add(word)
for tag in alltags:
total_prob = sum(np.exp(weights_nb[(tag, word)]) for word in allwords)
assert_almost_equals(
1.0,
total_prob,
places=2,
msg=
"UNEQUAL Expected tag %s to have total prob of 1.0, but instead has %s"
% (tag, total_prob))
