def gram(self, X, Xt=None):
"""
Returns the dense Gram matrix evaluated over the datapoints.
:param X: n-by-d data matrix
:param Xt: optional t-by-d test matrix
Returns:
-------
n-by-n Gram matrix over X (if Xt is not provided)
t-by-n Gram matrix between Xt and X if X is provided
"""
# TODO the test, and this function, should work for all matrix types.
if X.shape[1] != self._d:
raise ValueError("X must have vectors of dimension d")
nu = self._nu
l = self._l
if Xt is None:
D = euclidean(X, X)
else:
if Xt.shape[1] != self._d:
raise ValueError("Xt must have vectors of dimension d")
D = euclidean(X, Xt);
Y = scipy.sqrt(2.0 * nu * D) / l
K = 2.0 ** (1 - nu) / scipy.special.gamma(nu) * Y ** nu * scipy.special.kv(nu, Y)
return scipy.real(K)
def gram(self, X, Xt=None):
"""
Returns the dense Gram matrix evaluated over the datapoints.
:param X: n-by-d data matrix
:param Xt: optional t-by-d test matrix
Returns:
-------
n-by-n Gram matrix over X (if Xt is not provided)
t-by-n Gram matrix between Xt and X if X is provided
"""
# TODO the test, and this function, should work for all matrix types.
if X.shape[1] != self._d:
raise ValueError("X must have vectors of dimension d")
sigma = self._sigma
if Xt is None:
K = numpy.exp(-euclidean(X, X)/(2*sigma**2))
else:
if Xt.shape[1] != self._d:
raise ValueError("Xt must have vectors of dimension d")
K = numpy.exp(-euclidean(X, Xt)/(2*sigma**2))
return K
def __init__(self, ListOfObjects, algorithm=None, eps=800):
"""Parameters:
ListOfObjects
algorithm
"""
# first the centroids: A_i
if algorithm == "optics":
self.list_centroids, SetOfClusters = centroids(ListOfObjects,
algorithm,
eps)
else:
self.list_centroids, SetOfClusters = centroids(ListOfObjects)
# second the S value or internal dispertion of each centroid
self.S = []
sum_acumulated = 0
for key in SetOfClusters.keys():
for vector in SetOfClusters[key]:
v1 = vector.Coords
v2 = self.list_centroids[key - 1]
result = np.array((v1[0] - v2[0], v1[1] - v2[1]))
result = np.power(result, 2)
result = np.sum(result)
sum_acumulated += result
self.S.append(
np.sqrt(
np.true_divide(sum_acumulated, len(SetOfClusters[key]))))
# third the distance beetween every cluster
NumberOfClusters = len(self.list_centroids)
self.M = np.zeros([NumberOfClusters, NumberOfClusters])
for i in range(NumberOfClusters - 1):
for j in range(i + 1, NumberOfClusters):
self.M[i, j] = dist.euclidean(self.list_centroids[i],
self.list_centroids[j])
# fourth the ratio beetween internal dispersion of each centroid and
# between centroids.
self.R = np.zeros([NumberOfClusters, NumberOfClusters])
for i in range(NumberOfClusters - 1):
for j in range(i + 1, NumberOfClusters):
self.R[i, j] = np.true_divide(self.S[i] + self.S[j],
self.M[i, j])
self.R[j, i] = self.R[i, j]
# the Davies-Bouldien value.
self.value = 0
for i in range(NumberOfClusters):
self.value += np.max(self.R[i, :])
self.value = np.true_divide(self.value, NumberOfClusters)
def __init__(self, ListOfObjects, algorithm=None, eps=800):
"""Parameters:
ListOfObjects
algorithm
"""
# first the centroids: A_i
if algorithm == "optics":
self.list_centroids, SetOfClusters = centroids(
ListOfObjects, algorithm, eps)
else:
self.list_centroids, SetOfClusters = centroids(ListOfObjects)
# second the S value or internal dispertion of each centroid
self.S = []
sum_acumulated = 0
for key in SetOfClusters.keys():
for vector in SetOfClusters[key]:
v1 = vector.Coords
v2 = self.list_centroids[key - 1]
result = np.array((v1[0] - v2[0], v1[1] - v2[1]))
result = np.power(result, 2)
result = np.sum(result)
sum_acumulated += result
self.S.append(
np.sqrt(np.true_divide(sum_acumulated,
len(SetOfClusters[key]))))
# third the distance beetween every cluster
NumberOfClusters = len(self.list_centroids)
self.M = np.zeros([NumberOfClusters, NumberOfClusters])
for i in range(NumberOfClusters - 1):
for j in range(i + 1, NumberOfClusters):
self.M[i, j] = dist.euclidean(self.list_centroids[i],
self.list_centroids[j])
# fourth the ratio beetween internal dispersion of each centroid and
# between centroids.
self.R = np.zeros([NumberOfClusters, NumberOfClusters])
for i in range(NumberOfClusters - 1):
for j in range(i + 1, NumberOfClusters):
self.R[i, j] = np.true_divide(self.S[i] + self.S[j], self.M[i,
j])
self.R[j, i] = self.R[i, j]
# the Davies-Bouldien value.
self.value = 0
for i in range(NumberOfClusters):
self.value += np.max(self.R[i, :])
self.value = np.true_divide(self.value, NumberOfClusters)
def __init__(self, sentence1, sentence2): self.ham = hamming(sentence1, sentence2) self.euc = euclidean(sentence1, sentence2) self.man = manhattan(sentence1, sentence2) self.w2v = similarity(sentence1, sentence2) self.res = [self.ham, self.euc, self.man, self.w2v]
