def validate(self):
x = self.data
u = self.membership
v = self.centers
k = u.shape[0]
e1 = np.sum(np.square(u) * Euclidean()(x, np.mean(x, axis=0).reshape(
1, -1)))
ek = np.sum(np.square(u) * Euclidean()(x, v))
dk = np.max(Euclidean()(v))
return (1. / k * e1 / ek * dk) ** 2
def validate(self):
spread = np.square(Euclidean()(self.data, self.centers))
separation = np.square(self.intercluster_distances())
separation[separation == 0] = np.inf
n = self.data.shape[0]
return (1. / n) * np.sum(
np.multiply(np.square(self.membership), spread)) / np.min(
separation)
def diameters(partition):
"""Calculates the maximum pairwise distance across a partition.
Parameters
----------
partition : list
a list of arrays, each containing data belonging to a specific
cluster.
"""
return np.array([np.max(Euclidean()(p)) for p in partition])
def mean_pairwise_distances(partition):
"""Calculates the mean pairwise distance across a partition.
Parameters
----------
partition : list
a list of arrays, each containing data belonging to a specific
cluster.
"""
mpd = [np.mean(Euclidean()(p)) / 2 for p in partition]
return np.array(mpd)
def mean_distance_from_mean(partition):
"""Calculates the data's mean distance from the partition's mean.
Parameters
----------
partition : list
a list of arrays, each containing data belonging to a specific
cluster.
"""
ds = [np.mean(Euclidean()(p, np.mean(p, axis=0).reshape(1, -1))) for p
in partition]
return np.array(ds)
def cluster_scatter(self):
"""Calculates the mean distance between data and cluster centers."""
return np.array(
[np.mean(Euclidean()(p, a.reshape(1, -1)), axis=1) for p, a in
zip(self.partition, self.centers)]).flatten()
def intercluster_distances(self):
"""Calculates the euclidean distance between cluster centers."""
return Euclidean()(self.centers)