def sparsify(distMat, cutoff, kNN, threads):
sparse_coordinates = pp_sketchlib.sparsifyDists(distMat,
distCutoff=cutoff,
kNN=kNN)
sparse_scipy = ijv_to_coo(sparse_coordinates, distMat.shape, np.float32)
# Mirror to fill in lower triangle
if cutoff > 0:
sparse_scipy = sparse_scipy + sparse_scipy.transpose()
return (sparse_scipy)
def fit(self, X, accessory):
'''Extends :func:`~ClusterFit.fit`
Gets assignments by using nearest neigbours.
Args:
X (numpy.array)
The core and accessory distances to cluster. Must be set if
preprocess is set.
accessory (bool)
Use accessory rather than core distances
Returns:
y (numpy.array)
Cluster assignments of samples in X
'''
ClusterFit.fit(self, X)
sample_size = int(round(0.5 * (1 + np.sqrt(1 + 8 * X.shape[0]))))
if (max(self.ranks) >= sample_size):
sys.stderr.write("Rank must be less than the number of samples")
sys.exit(0)
if accessory:
self.dist_col = 1
else:
self.dist_col = 0
self.nn_dists = {}
for rank in self.ranks:
row, col, data = \
pp_sketchlib.sparsifyDists(
pp_sketchlib.longToSquare(X[:, [self.dist_col]], self.threads),
0,
rank
)
data = [epsilon if d < epsilon else d for d in data]
if self.use_gpu:
self.nn_dists[rank] = cupyx.scipy.sparse.coo_matrix(
(cp.array(data), (cp.array(row), cp.array(col))),
shape=(sample_size, sample_size),
dtype=X.dtype)
else:
self.nn_dists[rank] = scipy.sparse.coo_matrix(
(data, (row, col)),
shape=(sample_size, sample_size),
dtype=X.dtype)
self.fitted = True
y = self.assign(min(self.ranks))
return y
def sparsify(distMat, cutoff, kNN, threads):
sparse_coordinates = pp_sketchlib.sparsifyDists(distMat,
distCutoff=cutoff,
kNN=kNN,
num_threads=threads)
sparse_scipy = coo_matrix((sparse_coordinates[2],
(sparse_coordinates[0], sparse_coordinates[1])),
shape=distMat.shape,
dtype=np.float32)
# Mirror to fill in lower triangle
if cutoff > 0:
sparse_scipy = sparse_scipy + sparse_scipy.transpose()
return(sparse_scipy)