def extract_(self, post, topic, features=[]):
if len(features) == 0: features = _featlists.keys()
else: features = filter(lambda x: x in _featlists, features)
spans = [] # span := [ span-id, [ start, stop, { "category": (major, minor), "environment_indicators": [ (word, occurrence_id) ] } ] ]
for featpair in features:
(word_pairs, phrase_pairs) = _featlists[featpair]
occ_i = 0
for word, rx in word_pairs:
position = 0
for sentence in re.split(r'\.!?', post.text):
match = rx.match(post.text)
if match:
spans.append( [ "%s-%s" % (str(post.id), len(spans)+1),
[ position + match.start(),
position + match.end(),
{ "category": featpair,
"environment_indicators":
[ (word, str(occ_i)) ] } ] ] )
position += len(sentence)
occ_i += 1
occ_i = 0
for phrase, rx in phrase_pairs:
for m in rx.finditer(post.text):
spans.append( [ "%s-%s" % (str(post.id), len(spans)+1),
[ m.start(),
m.end(),
{ "category": featpair,
"environment_indicators":
[ (m.group(0), -1) ] } ] ] )
occ_i += 1
for span in set([d[1][2]['category'] for d in spans]):
self.freq[span] += 1
self.by_topic[topic][span] += 1
if len(self.by_topic[topic].keys()) < 2: return
#self.by_side[topic][post.topic_side][span] += 1
tuples = []
for span in spans:
start, stop, name = span[1][0], span[1][1], '-'.join(span[1][2]['category'])
tuples.append((start,stop,name))
b = Boundaries()
b.initializeFromTuples(tuples)
if len(b.boundaries) == 0: return
self.by_topic[topic][POSTS_KEY] += 1
b.walk(1, max(tuples, key=operator.itemgetter(1)))
feature_vector = defaultdict(int)
for partition in b.partitions[:-1]:
unigrams = map(lambda unigram: self.stemmer.stem(unigram.lower()), re.split(r'\W', post.text[partition[0]:partition[1]]))
for unigram in unigrams:
feature_vector['{}:{}'.format(partition[2], unigram)] += 1
return feature_vector
def extract(self, post, topic, features=[]):
if len(features) == 0: features = _featlists.keys()
else: features = filter(lambda x: x in _featlists, features)
spans = [] # span := [ span-id, [ start, stop, { "category": (major, minor), "environment_indicators": [ (word, occurrence_id) ] } ] ]
for featpair in features:
(word_pairs, phrase_pairs) = _featlists[featpair]
occ_i = 0
for word, rx in word_pairs:
position = 0
for sentence in re.split(r'\.!?', post.text):
match = rx.match(post.text)
if match:
spans.append( [ "%s-%s" % (str(post.id), len(spans)+1),
[ position + match.start(),
position + match.end(),
{ "category": featpair,
"environment_indicators":
[ (word, str(occ_i)) ] } ] ] )
position += len(sentence)
occ_i += 1
occ_i = 0
for phrase, rx in phrase_pairs:
for m in rx.finditer(post.text):
spans.append( [ "%s-%s" % (str(post.id), len(spans)+1),
[ m.start(),
m.end(),
{ "category": featpair,
"environment_indicators":
[ (m.group(0), -1) ] } ] ] )
occ_i += 1
self.by_topic[topic][POSTS_KEY] += 1
environments = set()
for span in set([d[1][2]['category'] for d in spans]):
self.freq[span] += 1
self.by_topic[topic][span] += 1
self.by_side[topic][post.topic_side][span] += 1
environments.add(span[0]) #XXX TODO need to make sure that the environments we are checking for are in the top three as far as probabiltiy goes
self.environments_topic[topic].update([len(environments)])
if len(environments) < 2:
return
#print 'throwing post away. discussion: {discussion} id: {id}'.format(**{'discussion': topic, 'id': post.id})
tuples = []
for span in spans:
start, stop, name = span[1][0], span[1][1], '-'.join(span[1][2]['category'])
tuples.append((start,stop,name))
b = Boundaries()
b.initializeFromTuples(tuples)
if len(b.boundaries) == 0: return
b.walk(1, max(tuples, key=operator.itemgetter(1)))
#print 'boundaries:{boundary.boundaries}\nparititions:{boundary.partitions}'.format(boundary=b)
fv_all = defaultdict(int)
fv_collapsed = defaultdict(int)
fv_commitment = defaultdict(int)
tokens = 0
for partition in b.partitions[:-1]:
unigrams = map(lambda unigram: self.stemmer.stem(unigram.lower()), re.split(r'\W', post.text[partition[0]:partition[1]]))
tokens += len(unigrams)
for _label in set(partition[2].split()):
for unigram in unigrams:
fv_commitment['{}:{}'.format(_label, unigram)] += 1
if _label == 'none':
fv_collapsed['commitment:{unigram}'.format(unigram=unigram)] += 1
else:
fv_collapsed['non_commitment:{unigram}'.format(unigram=unigram)] += 1
for unigram in unigrams:
fv_all['unigram_{unigram}'.format(unigram=unigram)] += 1
for key in fv_all.keys():
fv_all[key] /= float(tokens)
for key in fv_commitment.keys():
fv_commitment[key] /= float(tokens)
for key in fv_collapsed.keys():
fv_collapsed[key] /= float(tokens)
fv_all[self.label] = post.topic_side
fv_commitment[self.label] = post.topic_side
fv_collapsed[self.label] = post.topic_side
return {'all': fv_all, 'collapsed': fv_collapsed, 'commitment': fv_commitment}
