def tag_sentences():
'''
input: train_sentences.json
output: train_tag_sentences.json
Tags the chemicals in each sentence using ChemicalTagger
(http://chemicaltagger.ch.cam.ac.uk/). This method communicates with
ChemicalTagger through a custom REST API running on pathway.berkeley.edu
'''
sentences = json.load(open('../data/train_sentences.json'))
bar, i = pbar(len(sentences)), 0
print 'Tagging chemicals in sentences'
bar.start()
chemicals = {}
for sid, sentence in sentences.iteritems():
chems = chemtagger.get_compounds(sid, sentence)
if chems:
chemicals[sid] = chems
i += 1
bar.update(i)
bar.finish()
json.dump(chemicals, open('../data/train_tag_sentences.json', 'wb'),
indent=2, sort_keys=True)
print 'Result dumped to ../data/train_tag_sentences.json'
def extract(sid, sentence):
reactants = []
chemicals = chemtagger.get_compounds(sid, sentence)
if chemicals is None:
return reactants
chemicals = sanitize_chemicals(chemicals)
chems = [y for x in chemicals for y in x.split()]
stemmed_sentence = ' '.join([x if x in chems else STEMMER.stem(x)
for x in sentence.split()])
tagged_sentence = ' %s ' % stemmed_sentence
for chem in sorted(chemicals, key=len, reverse=True):
tagged_sentence = tagged_sentence.replace(
' %s ' % chem, ' $%s$chem ' % chem)
tagged_sentence = tagged_sentence.strip()
grouped_sentence = group_list(tagged_sentence)
for pattern_id in expand_patterns():
for pattern in expand_patterns()[pattern_id]:
groups = pattern.findall(grouped_sentence, overlapped=True)
matches = expand_chems(groups)
for match in matches:
if match and len(match) > 1:
reactants.append((pattern_id, match))
return reactants
