def __init__(self, source_path):
'''
Create the Neighborhood, for finding nearest neighbors.
Args:
source_path (string): path to a bcolz database with three carray
columns: 'id', 'vector' and 'norm'
'''
self.source_path = source_path
# open bcolz datastores
self.vectors = bvec.carray(rootdir=source_path + "/vector")
self.norms = bvec.carray(rootdir=source_path + "/norm")
self.source_table = bcolz.ctable(rootdir=source_path)
#print("Created similarity object from BCOLZ files: source {0}; target: {1}".format(source_path, target_path))
# create similarity object
self.similarity = sim.Similarity(self.vectors, self.norms)
# create domain <-> index maps
# dictionary taking ids to indeces (source)
self.id_index_map = self._create_id_index_map(self.source_table)
self.index_id_map = self._create_index_id_map(self.source_table)
def empty_like(self, shape=None, chunklen=None, cparams=None, rootdir=None):
'''
Create an empty bvec.carray container matching this one, with optional
modifications.
'''
p_dtype = self.dtype
if shape == None:
shape = self.shape
if cparams == None:
cparams = self.cparams
if(len(shape) == 1):
result_template = np.ndarray(shape=(0), dtype=p_dtype)
return bvec.carray(result_template, expectedlen=shape[0], chunklen=chunklen, cparams=cparams, rootdir=rootdir)
elif(len(self.shape) == 2):
result_template = np.ndarray((0, shape[1]), dtype=p_dtype)
return bvec.carray(result_template, expectedlen=shape[0], chunklen=chunklen, cparams=cparams, rootdir=rootdir)
else:
raise ValueError("Can't create a carray like that. Only one and two dimensions supported.")
