def fix_modified_test_xml():
train_corpus = Corpus('./data/gold')
i = 0
for doc in train_corpus.documents():
name=doc.name.replace('gold','Test++')
with open(name,'r') as fo:
text=fo.read()
lines=text.split('\n')
i=0
for s in doc.tokenizer.tokenize_text().sentences:
sent = s.as_pairs()
for tok,lex in sent:
while not re.match('\s*<lex',lines[i]):
i+=1
lines[i] = re.sub("begin='\d+?'",
"begin='{}'".format(lex.begin),
lines[i])
lines[i] = re.sub("start='\d+?'",
"".format(lex.begin),
lines[i])
lines[i] = re.sub("end='\d+?'",
"end='{}'".format(lex.end),
lines[i])
#print tok,',',lex.begin,',',lex.end
#print re.findall('<lex .+?>(.+?)</lex>',lines[i])
#print lines[i]
i+=1
with open(doc.name.replace('gold','gold++'),'wb') as fo:
for line in lines:
fo.write(re.sub('\r','',line)) #clean carriage returns
fo.write('\n')
def run_demo(self, verbose=0):
""" test functionality of loading data, classification and evaluation """
# load training data
train_corpus = Corpus(self.train_path)
extents = list(train_corpus.extents(self.indices_function,
self.extent_class))
# load test data
if self.test_path:
train_data = extents
test_corpus = HypotheticalCorpus(self.test_path)
test_data = list(test_corpus.extents(self.indices_function,
self.extent_class))
else:
i = int(len(extents) * self.split)
train_data = extents[:i]
test_data = extents[i:]
# verbosity functionality
if verbose >= 1:
print "data loaded"
labels = [self.label_function(x) for x in extents]
if verbose >= 2:
fd = {}
for l in labels:
fd[l] = fd.get(l, 1) + 1
print fd
# train model
clf = SKClassifier(LogisticRegression(),
self.label_function,
self.feature_functions)
clf.add_labels(set(labels))
clf.train(train_data)
if verbose >= 1:
print "model trained"
# classify
pred = clf.classify(test_data,
keys = ["{a},{b},{c}".format(a=extent.basename,
b=extent.lex[0].begin,
c=extent.lex[-1].end)
for extent in test_data])
# evaluate
if self.gold_path:
gold_corpus = Corpus(self.gold_path)
gold_data = list(gold_corpus.extents(self.indices_function,
self.extent_class))
else:
gold_data = test_data
gold_labels = dict([
("{a},{b},{c}".format(a=extent.basename,
b=extent.lex[0].begin,
c=extent.lex[-1].end),
self.label_function(extent))
for extent in gold_data])
clf.evaluate(pred, gold_labels)
def generate_test_train(self):
""" create test and training instances based on provided paths """
train_corpus = Corpus(self.train_path)
extents = list(train_corpus.extents(self.indices_function,
self.extent_class))
# load test data
if self.test_path:
train_data = extents
test_corpus = HypotheticalCorpus(self.test_path)
test_data = list(test_corpus.extents(self.indices_function,
self.extent_class))
else:
i = int(len(extents) * self.split)
train_data = extents[:i]
test_data = extents[i:]
return train_data, test_data
def evaluate_qs_o_link(self):
""" given test and gold copora paths with matching docs,
compare proposed tags with actual tags """
clf = SKClassifier(LogisticRegression(),
self.label_function,
self.feature_functions)
# hyp
test_corpus = Corpus(self.test_path)
test_data = list(test_corpus.qs_o_link_triples(self.indices_function,
self.extent_class))
test_labels = dict([
(link_key.format(doc_name=extent.basename,
trigger_start = extent.token[0]['start'],
trigger_end = extent.token[0]['end'],
from_start = extent.token[1]['start'],
from_end = extent.token[1]['end'],
to_start = extent.token[2]['start'],
to_end = extent.token[2]['end']),
self.label_function(extent))
for extent in test_data])
clf.add_labels(test_labels.values())
# ref
gold_corpus = Corpus(self.gold_path)
gold_data = list(gold_corpus.qs_o_link_triples(self.indices_function,
self.extent_class))
gold_labels = dict([
(link_key.format(doc_name=extent.basename,
trigger_start = extent.token[0]['start'],
trigger_end = extent.token[0]['end'],
from_start = extent.token[1]['start'],
from_end = extent.token[1]['end'],
to_start = extent.token[2]['start'],
to_end = extent.token[2]['end']),
self.label_function(extent))
for extent in gold_data])
clf.add_labels(gold_labels.values())
cm = clf.evaluate(test_labels, gold_labels)
# output dict for
return cm
def evaluate(self):
""" given test and gold copora paths with matching docs,
compare proposed tags with actual tags """
clf = SKClassifier(LogisticRegression(),
self.label_function,
self.feature_functions)
# hyp
test_corpus = Corpus(self.test_path)
test_data = list(test_corpus.extents(self.indices_function,
self.extent_class))
test_labels = dict([
("{a},{b},{c}".format(a=extent.basename,
b=extent.lex[0].begin,
c=extent.lex[-1].end),
self.label_function(extent))
for extent in test_data])
clf.add_labels(test_labels.values())
# ref
gold_corpus = Corpus(self.gold_path)
gold_data = list(gold_corpus.extents(self.indices_function,
self.extent_class))
gold_labels = dict([
("{a},{b},{c}".format(a=extent.basename,
b=extent.lex[0].begin,
c=extent.lex[-1].end),
self.label_function(extent))
for extent in gold_data])
for key in test_labels.keys():
if key not in gold_labels:
gold_labels[key] = 'False'
for key in gold_labels.keys():
if key not in test_labels:
test_labels[key] = 'False'
clf.add_labels(test_labels.values())
clf.add_labels(gold_labels.values())
cm = clf.evaluate(test_labels, gold_labels)
# output dict for
return cm