def _update_centroids(self, x: DNDarray, matching_centroids: DNDarray):
"""
Compute coordinates of new centroid as mean of the data points in ``x`` that are assigned to this centroid.
Parameters
----------
x : DNDarray
Input data
matching_centroids : DNDarray
Array filled with indices ``i`` indicating to which cluster ``ci`` each sample point in ``x`` is assigned
"""
new_cluster_centers = self._cluster_centers.copy()
for i in range(self.n_clusters):
# points in current cluster
selection = (matching_centroids == i).astype(ht.int64)
# accumulate points and total number of points in cluster
assigned_points = x * selection
points_in_cluster = selection.sum(axis=0, keepdim=True).clip(
1.0,
ht.iinfo(ht.int64).max)
# compute the new centroids
new_cluster_centers[i:i + 1, :] = (assigned_points /
points_in_cluster).sum(
axis=0, keepdim=True)
return new_cluster_centers
def fit(self, X):
"""
Computes the centroid of a k-means clustering.
Parameters
----------
X : ht.DNDarray, shape=(n_samples, n_features)
Training instances to cluster.
"""
# input sanitation
if not isinstance(X, ht.DNDarray):
raise ValueError(
"input needs to be a ht.DNDarray, but was {}".format(type(X)))
# initialize the clustering
self._initialize_cluster_centers(X)
self._n_iter = 0
matching_centroids = ht.zeros((X.shape[0]),
split=X.split,
device=X.device,
comm=X.comm)
X = X.expand_dims(axis=2)
new_cluster_centers = self._cluster_centers.copy()
# iteratively fit the points to the centroids
for epoch in range(self.max_iter):
# increment the iteration count
self._n_iter += 1
# determine the centroids
matching_centroids = self._fit_to_cluster(X)
# update the centroids
for i in range(self.n_clusters):
# points in current cluster
selection = (matching_centroids == i).astype(ht.int64)
# accumulate points and total number of points in cluster
assigned_points = (X * selection).sum(axis=0, keepdim=True)
points_in_cluster = selection.sum(axis=0, keepdim=True).clip(
1.0,
ht.iinfo(ht.int64).max)
# compute the new centroids
new_cluster_centers[:, :, i:i +
1] = assigned_points / points_in_cluster
# check whether centroid movement has converged
self._inertia = ((self._cluster_centers -
new_cluster_centers)**2).sum()
self._cluster_centers = new_cluster_centers.copy()
if self.tol is not None and self._inertia <= self.tol:
break
self._labels = matching_centroids.squeeze()
return self
def test_iinfo(self):
info32 = ht.iinfo(ht.int32)
self.assertEqual(info32.bits, 32)
self.assertEqual(info32.max, 2147483647)
self.assertEqual(info32.min, -2147483648)
with self.assertRaises(TypeError):
ht.iinfo(1.0)
with self.assertRaises(TypeError):
ht.iinfo(ht.float64)
with self.assertRaises(TypeError):
ht.iinfo('int16')