def calculate_variance(X, Xs):
kernel = RBF()
kernel.__init__()
sigma = 0.01
ks = kernel.diag(Xs)
Kt = kernel.__call__(X, X) + sigma**2
if is_invertible(Kt):
Kt = inv(Kt)
else:
Kt = pinv(Kt)
K = kernel.__call__(Xs, X)
Ktt = np.matmul(K, Kt)
Ktt = np.matmul(Ktt, K.transpose())
Ktt = Ktt.diagonal()
var = ks - Ktt.transpose()
return var
def costfun_rbf(x):
rbf = RBF()
res = np.dot(np.dot(
np.transpose(x), rbf.__call__(S, S)), x) - 2 * np.dot(
np.transpose(x), rbf.__call__(S, b)) + rbf.__call__(b, b)
return res[0][0]
# Anisotropicly distributed data
random_state = 170
X, y = datasets.make_blobs(n_samples=n_samples, random_state=random_state)
transformation = [[0.6, -0.6], [-0.4, 0.8]]
X_aniso = np.dot(X, transformation)
aniso = (X_aniso, y)
# print aniso
for datatuple in [aniso]:
r = 1
if r == 1:
run_kmeans_and_plot(datatuple, 'kernel kmeans')
elif r == 2:
run_kmeans_and_plot(datatuple, 'kmeans')
else:
# print 'test'
x = np.asarray(datatuple[0])
kernel = 'rbf'
if kernel == 'linear':
kernel_product = pairwisekernel.linear_kernel(x, x)
# print kernel_product.shape
elif kernel == 'quadratic':
kernel_product = pairwisekernel.polynomial_kernel(x, x, degree=2)
elif kernel == 'rbf':
# print 'test'
rbf_kernel = RBF()
print rbf_kernel.hyperparameters
kernel_product = rbf_kernel.__call__(x, x)
print kernel_product.shape
def get_Relative_Average_similarity(X, AverageSimilarity):
X_rand = X[np.random.randint(1, len(X), size=5000 )]
rbf = RBF(1)
M = rbf.__call__(X_rand, X)
A_temp = 1/AverageSimilarity
return np.mean(M*A_temp , axis=0)
def get_Average_similarity(X):
X_rand = X[np.random.randint(1, len(X), size=5000 )]
rbf = RBF(1)
M = rbf.__call__(X, X_rand)
AverageSimilarity = np.mean(M, axis=1)
return AverageSimilarity
