def make_links(train_path, test_path, out_path,
index_filter, tag_class, link_name, id_prefix):
# make olink
test_data = HypotheticalCorpus(test_path)
test_data = list(test_data.extents(index_filter,
tag_class))
#parse into XML tags
id_number = 0
curr_doc = test_data[0].document
doc_name = curr_doc.basename
for extent in test_data:
if doc_name != extent.document.basename:
curr_doc.save_xml(os.path.join(out_path, doc_name))
id_number = 0
curr_doc = extent.document
doc_name = curr_doc.basename
tag = {'name': link_name,
'id': '{}{}'.format(id_prefix, id_number),
'trigger': extent.tag['id'],
'fromID': extent.tag['id'],
'fromText': extent.tag['text'],
'toID': extent.tag['id'],
'toText': extent.tag['text'],
}
extent.document.insert_tag(tag)
id_number += 1
curr_doc.save_xml(os.path.join(out_path, doc_name))
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 generate_tags(train_path, test_path, clean_path, out_path):
# make outpath
if not os.path.exists(out_path):
os.makedirs(out_path)
# clean data to write tags to
clean_corpus = HypotheticalCorpus(clean_path)
clean_data = list(clean_corpus.documents())
for type_name, type_fields in tag_types.iteritems():
# generate labels
demo = TypesClassifier(type_name, train_path, test_path, gold_path = '')
pred, test_data = demo.generate_labels()
# labels -> tagged docs
id_number = 0
i = 0
curr_doc = clean_data[0]
doc_name = curr_doc.basename
for extent in test_data:
if doc_name != extent.document.basename:
curr_doc.save_xml(os.path.join(out_path, doc_name))
id_number = 0
i += 1
curr_doc = clean_data[i]
doc_name = curr_doc.basename
offsets = "{a},{b},{c}".format(a=extent.basename,
b=extent.lex[0].begin,
c=extent.lex[-1].end)
if pred[offsets] == 'True':
tag = {'name': type_name,
'start': extent.tag['start'],
'end': extent.tag['end'],
'text': extent.tag['text'],
'id': '{}{}'.format(type_keys[type_name][0], id_number),
}
tag.update(type_fields)
curr_doc.insert_tag(tag)
id_number += 1
# switch to established files
curr_doc.save_xml(os.path.join(out_path, doc_name))
clean_corpus = HypotheticalCorpus(out_path)
clean_data = list(clean_corpus.documents())
