def bucketize(self, minhashes):
buckets = []
try:
band_bits = self.dataset.band_bits
except AttributeError:
matrix = self.dataset
band_bits = int(math.ceil(math.log(matrix.bands, 2)))
band_mask = (2**band_bits - 1)
setattr(matrix, 'band_bits', band_bits)
setattr(matrix, 'band_mask', band_mask)
setattr(matrix, 'hash_mask',
2**(settings.max_bits - band_bits) - 1)
band_mask = self.dataset.band_mask
hash_mask = self.dataset.hash_mask
for band in xrange(self.dataset.bands):
band_hash = (band_mask & band) * (hash_mask + 1)
minhashes_in_band = [
minhashes[band * self.rows + row] for row in xrange(self.rows)
]
minhashes_into_a_string = '-'.join(
[str(mh) for mh in minhashes_in_band])
bucket = band_hash | (hash_mask & int(
hashlib.md5(minhashes_into_a_string).hexdigest(), 16))
buckets.append(DbInt.to_db(bucket))
return buckets
def get_nns(self, doc_id):
doc = self.get_doc(doc_id)
if not doc:
return []
bkts = [DbInt.fm_db(bkt) for bkt in doc.buckets]
mhs = {}
for bkt in bkts:
bkt_docs = session.execute(self.nns_select, [self.ds_key, DbInt.to_db(bkt)])
for bkt_doc in bkt_docs:
mhs[bkt_doc['doc_id']] = bkt_doc['minhashes']
del mhs[doc_id]
jac = {}
for doc_id2 in mhs.keys():
jac_min = reduce(lambda x, y: x+y, map(lambda a,b: a == b, doc.minhashes,mhs[doc_id2])) / float(len(doc.minhashes))
jac[doc_id2] = 1.0 - jac_min
if 0 == int(1000*time.time()) % 100:
logging.info('Sampling (1%%) Jaccard distance %s | %s: %6.2f', doc_id, doc_id2, jac[doc_id2])
return jac
def bucketize(self, minhashes):
buckets = []
try:
band_bits = self.dataset.band_bits
except AttributeError:
matrix = self.dataset
band_bits = int(math.ceil(math.log(matrix.bands, 2)))
band_mask = (2**band_bits - 1)
setattr(matrix, 'band_bits', band_bits)
setattr(matrix, 'band_mask', band_mask)
setattr(matrix, 'hash_mask', 2**(settings.max_bits - band_bits)-1)
band_mask = self.dataset.band_mask
hash_mask = self.dataset.hash_mask
for band in xrange(self.dataset.bands):
band_hash = (band_mask & band) * (hash_mask + 1)
minhashes_in_band = [minhashes[band*self.rows + row] for row in xrange(self.rows)]
minhashes_into_a_string = '-'.join([str(mh) for mh in minhashes_in_band])
bucket = band_hash | (hash_mask & int(hashlib.md5(minhashes_into_a_string).hexdigest(), 16))
buckets.append(DbInt.to_db(bucket))
return buckets
def get_nns(self, doc_id):
doc = self.get_doc(doc_id)
if not doc:
return []
bkts = [DbInt.fm_db(bkt) for bkt in doc.buckets]
mhs = {}
for bkt in bkts:
bkt_docs = session.execute(
self.nns_select, [self.ds_key, DbInt.to_db(bkt)])
for bkt_doc in bkt_docs:
mhs[bkt_doc['doc_id']] = bkt_doc['minhashes']
del mhs[doc_id]
jac = {}
for doc_id2 in mhs.keys():
jac_min = reduce(
lambda x, y: x + y,
map(lambda a, b: a == b, doc.minhashes, mhs[doc_id2])) / float(
len(doc.minhashes))
jac[doc_id2] = 1.0 - jac_min
if 0 == int(1000 * time.time()) % 100:
logging.info('Sampling (1%%) Jaccard distance %s | %s: %6.2f',
doc_id, doc_id2, jac[doc_id2])
return jac