def fill(self):
zf = zipfile.ZipFile(self.path)
if not zf.namelist() == ['test.txt', 'train.txt', 'vocab.txt']:
raise Exception(
'Collection arch must be contain only this files: test.txt, train.txt, vocab.txt'
' but find %s' % zf.namelist())
for text_id, doc in grouper(zf.open('train.txt'), 2):
self.documents_train.append(
map(int,
doc.decode('utf-8').strip().split()))
for text_id, doc in grouper(zf.open('test.txt'), 2):
self.documents_test.append(
map(int,
doc.decode('utf-8').strip().split()))
self.id_to_words = dict(
filter(
len,
imap(split,
zf.open('vocab.txt').read().decode('utf8').split('\n'))))
self.id_to_words = dict(
zip(map(int, self.id_to_words.keys()), self.id_to_words.values()))
self.words_to_id = dict(
imap(lambda x: (x[1], x[0]), self.id_to_words.iteritems()))
self.num_wrd = len(self.id_to_words)
print 'Read %s num doc train = %s test = %s num wrd %s' % \
(self.path, len(self.documents_train), len(self.documents_test), self.num_wrd)
return self
def fill(self):
zf = zipfile.ZipFile(self.path)
if not zf.namelist() == ['test.txt', 'train.txt', 'vocab.txt']:
raise Exception('Collection arch must be contain only this files: test.txt, train.txt, vocab.txt')
for text_id, doc in grouper(zf.open('train.txt'), 2):
self.documents_train.append(map(int, doc.decode('utf-8').strip().split()))
for text_id, doc in grouper(zf.open('test.txt'), 2):
self.documents_test.append(map(int, doc.decode('utf-8').strip().split()))
self.id_to_words = dict(filter(len, imap(split, zf.open('vocab.txt').read().decode('utf8').split('\n'))))
self.id_to_words = dict(zip(map(int, self.id_to_words.keys()), self.id_to_words.values()))
self.words_to_id = dict(imap(lambda x: (x[1], x[0]), self.id_to_words.iteritems()))
self.num_wrd = len(self.id_to_words)
print 'Read %s num doc train = %s test = %s num wrd %s' % \
(self.path, len(self.documents_train), len(self.documents_test), self.num_wrd)
return self
def train(self, collection):
bigrams = dict()
collocation_measure = lambda coloc: counts_neighbors[
coloc] * 1.0 / counts_windows[coloc]
for i, document in enumerate(collection.documents_train +
collection.documents_test):
groups_words = grouper([wrd for wrd in document], 5)
counts_neighbors = defaultdict(lambda: 0)
counts_windows = defaultdict(lambda: 0)
for group in groups_words:
count_neighbor = Counter(ngrams(group, 2))
count_window = Counter(all_pairs(group))
dicts_sum(counts_neighbors, count_neighbor)
dicts_sum(counts_windows, count_window)
buf = []
for item in counts_neighbors:
if collocation_measure(item) > self.sigma and counts_neighbors[
item] > self.min_occur:
if item[0] != item[1]:
buf.append((item, (collocation_measure(item),
counts_neighbors[item])))
bigrams[i] = dict(buf)
collection.bigrams = []
for i in xrange(
len(collection.documents_train + collection.documents_test)):
collection.bigrams += bigrams[i].keys()
collection.bigrams = set(collection.bigrams)
documents = [
wrd for document in collection.documents_train +
collection.documents_test for wrd in document
]
bigrams = filter(lambda bigr: bigr in collection.bigrams,
ngrams(documents, 2))
collection.bigrams = dict(
sorted(Counter(bigrams).items(), key=itemgetter(1),
reverse=True)[:self.top])
max_v = max(collection.words_to_id.values()) + 1
for bigram in collection.bigrams.keys():
collection.words_to_id[bigram] = max_v
collection.id_to_words[max_v] = collection.id_to_words[
bigram[0]] + '_' + collection.id_to_words[bigram[1]]
max_v += 1