z = z - 1
idx = range(len(data))
np.random.shuffle(idx)
data = data[idx]
z = z[idx]
data = (data - data.mean(axis=0))/data.std(axis=0)
sigma2 = sum([np.linalg.norm(x-y)**2
for x in data for y in data])/(len(data)**2)
sigma = np.sqrt(sigma2)
rho_exp = lambda x, y: 2-2*np.exp(-np.linalg.norm(x-y)/(2*sigma))
rho_gauss = lambda x, y: 2-2*np.exp(-np.linalg.norm(x-y)**2/(2*(sigma)**2))
G = eclust.kernel_matrix(data, rho)
#G = eclust.kernel_matrix(data, rho_gauss)
#G = eclust.kernel_matrix(data, rho_exp)
k = 3
r = []
r.append(wrapper.kmeans(k, data, run_times=5))
r.append(wrapper.gmm(k, data, run_times=5))
r.append(wrapper.spectral_clustering(k, data, G, run_times=5))
r.append(wrapper.spectral(k, data, G, run_times=5))
r.append(wrapper.kernel_kmeans(k, data, G, run_times=5, ini='random'))
#r.append(wrapper.kernel_kmeans(k, data, G, run_times=5, ini='k-means++'))
#r.append(wrapper.kernel_kmeans(k, data, G, run_times=5, ini='spectral'))
r.append(wrapper.kernel_kgroups(k,data,G,run_times=5, ini='random'))
#r.append(wrapper.kernel_kgroups(k,data,G,run_times=5, ini='k-means++'))
if __name__ == "__main__":
import data
import metric
from prettytable import PrettyTable
import sys
n = 400
d = 10
n1, n2 = np.random.multinomial(n, [1/2, 1/2])
m1 = np.zeros(d)
m2 = 0.7*np.ones(d)
s1 = s2 = np.eye(d)
X, z = data.multivariate_normal([m1, m2], [s1, s2], [n1, n2])
G = eclust.kernel_matrix(X, lambda x, y: np.linalg.norm(x-y))
W = np.eye(n)
k = 2
t = PrettyTable(["Method", "Accuracy"])
zh = kernel_kmeans(k, X, G, W, run_times=5, ini="k-means++")
a = metric.accuracy(z, zh)
t.add_row(["Kernel k-means", a])
zh = kernel_kgroups(k, X, G, W, run_times=5, ini="k-means++")
a = metric.accuracy(z, zh)
t.add_row(["Kernel k-groups", a])
zh = spectral(k, X, G, W, run_times=5)
a = metric.accuracy(z, zh)
def generate_data(D):
d = 10
m1 = np.zeros(D)
s1 = np.eye(D)
m2 = np.concatenate((0.7 * np.ones(d), np.zeros(D - d)))
s2 = np.eye(D)
n1, n2 = np.random.multinomial(total_points, [0.5, 0.5])
X, z = data.multivariate_normal([m1, m2], [s1, s2], [n1, n2])
return X, z
r = []
for _ in range(num_experiments):
for dim in dimensions:
X, z = generate_data(dim)
G = eclust.kernel_matrix(X, lambda x, y: np.linalg.norm(x - y))
zh = wrapper.kmeans(k, X)
a = metric.accuracy(z, zh)
r.append(['k-means', dim, a])
zh = wrapper.gmm(k, X)
a = metric.accuracy(z, zh)
r.append(['gmm', dim, a])
zh = wrapper.spectral_clustering(k, X, G)
a = metric.accuracy(z, zh)
r.append(['spectral clustering', dim, a])
zh = wrapper.kernel_kmeans(k, X, G)
a = metric.accuracy(z, zh)
s1 = np.array([[1, 0], [0, 20]])
s2 = np.array([[1, 0], [0, 20]])
r1 = 1
r2 = 3
eps = 0.2
r = []
for _ in range(num_experiments):
n1, n2 = np.random.multinomial(total_points, [0.5, 0.5])
#X, z = data.multivariate_normal([m1, m2], [s1, s2], [n1, n2])
X, z = data.circles([r1, r2], [eps, eps], [n1, n2])
#G = eclust.kernel_matrix(X,
# lambda x, y: 2 - 2*np.exp(-np.linalg.norm(x-y)/2/2))
G = eclust.kernel_matrix(
X, lambda x, y: 2 - 2 * np.exp(-np.linalg.norm(x - y)**2 / 2 / 1))
row = []
zh = wrapper.kmeans(k, X)
a = metric.accuracy(z, zh)
row.append(a)
zh = wrapper.gmm(k, X)
a = metric.accuracy(z, zh)
row.append(a)
zh = wrapper.spectral_clustering(k, X, G)
a = metric.accuracy(z, zh)
row.append(a)
zh = wrapper.kernel_kmeans(k, X, G, ini='random')
m1 = np.zeros(D)
s1 = 0.5*np.eye(D)
m2 = 0.5*np.concatenate((np.ones(d), np.zeros(D-d)))
s2 = np.eye(D)
n1, n2 = np.random.multinomial(n, [0.5, 0.5])
if distr_type == 'normal':
X, z = data.multivariate_normal([m1, m2], [s1, s2], [n1, n2])
elif distr_type == 'lognormal':
X, z = data.multivariate_lognormal([m1, m2], [s1, s2], [n1, n2])
return X, z
r = []
for _ in range(num_experiments):
for n in num_points:
X, z = generate_data(n)
G1 = eclust.kernel_matrix(X, lambda x, y: np.linalg.norm(x-y))
G2 = eclust.kernel_matrix(X,
lambda x, y: np.power(np.linalg.norm(x-y), 0.5))
G3 = eclust.kernel_matrix(X,
lambda x, y: 2-2*np.exp(-np.linalg.norm(x-y)/2))
zh = wrapper.kmeans(k, X)
a = metric.accuracy(z, zh)
r.append(['k-means', n, a])
zh = wrapper.gmm(k, X)
a = metric.accuracy(z, zh)
r.append(['gmm', n, a])
zh = wrapper.spectral_clustering(k, X, G3)
a = metric.accuracy(z, zh)
rho_gauss = lambda x, y: 2-2*np.exp(-np.linalg.norm(x-y)**2/(2*(1)**2))
#df = pd.read_csv('data/wdbc.data', sep=',', header=None)
#df = pd.read_csv('data/iris.data', sep=',', header=None)
#classes = {
# 'Iris-setosa': 0,
# 'Iris-versicolor': 1,
# 'Iris-virginica': 2
#}
z = np.array([classes[v] for v in df[4].values])
df = df.drop(4, axis=1)
data = df.values
data = (data - data.mean(axis=0))/data.std(axis=0)
G = eclust.kernel_matrix(data, rho_gauss)
k = 3
nt = 5
r = []
r.append(wrapper.kmeans(k, data, run_times=nt))
r.append(wrapper.gmm(k, data, run_times=nt))
r.append(wrapper.spectral_clustering(k, data, G, run_times=nt))
r.append(wrapper.spectral(k, data, G, run_times=nt))
r.append(wrapper.kernel_kmeans(k, data, G, run_times=nt, ini='spectral'))
r.append(wrapper.kernel_kgroups(k,data,G,run_times=nt, ini='spectral'))
t = PrettyTable(['Algorithm', 'Accuracy', 'A-Rand'])
algos = ['kmeans', 'GMM', 'spectral clustering', 'spectral',
'kernel k-means', 'kernel k-groups']
s1 = 0.5 * np.eye(D)
m2 = 0.5 * np.concatenate((np.ones(d), np.zeros(D - d)))
s2 = np.eye(D)
n1, n2 = np.random.multinomial(n, [0.5, 0.5])
if distr_type == 'normal':
X, z = data.multivariate_normal([m1, m2], [s1, s2], [n1, n2])
elif distr_type == 'lognormal':
X, z = data.multivariate_lognormal([m1, m2], [s1, s2], [n1, n2])
return X, z
r = []
for _ in range(num_experiments):
for n in num_points:
X, z = generate_data(n)
G1 = eclust.kernel_matrix(X, lambda x, y: np.linalg.norm(x - y))
G2 = eclust.kernel_matrix(
X, lambda x, y: np.power(np.linalg.norm(x - y), 0.5))
G3 = eclust.kernel_matrix(
X, lambda x, y: 2 - 2 * np.exp(-np.linalg.norm(x - y) / 2))
zh = wrapper.kmeans(k, X)
a = metric.accuracy(z, zh)
r.append(['k-means', n, a])
zh = wrapper.gmm(k, X)
a = metric.accuracy(z, zh)
r.append(['gmm', n, a])
zh = wrapper.spectral_clustering(k, X, G3)
a = metric.accuracy(z, zh)
table = np.zeros((num_experiments, 5))
for i in range(num_experiments):
X, z = data.univariate_lognormal([0, -1.5], [0.3, 1.5], [100, 100])
#X, z = data.univariate_normal([0, 5], [1, 22], [15, 15])
Y = np.array([[x] for x in X])
k = 2
# 1D energy clustering
zh, cost = two_clusters1D(X)
table[i,0] = accuracy(z, zh)
# initialization
z0 = initialization.kmeanspp(k, Y, ret='labels')
Z0 = eclust.ztoZ(z0)
rho = lambda x, y: np.linalg.norm(x-y)
G = eclust.kernel_matrix(Y, rho)
z1 = initialization.spectral(k, G)
Z1 = eclust.ztoZ(z1)
# Hartigan's method
zh = eclust.energy_hartigan(k, G, Z0)
table[i,1] = accuracy(z, zh)
zh = eclust.energy_hartigan(k, G, Z1)
table[i,2] = accuracy(z, zh)
# standard k-means
km = KMeans(2)
zh = km.fit_predict(Y)
table[i, 3] = accuracy(z, zh)
idx = np.random.choice(range(len(data)), 2000)
data = data[idx]
z = z[idx]
data = (data - data.mean(axis=0)) / data.std(axis=0)
sigma2 = sum([np.linalg.norm(x - y)**2 for x in data
for y in data]) / (len(data)**2)
sigma = np.sqrt(sigma2)
rho_exp = lambda x, y: 2 - 2 * np.exp(-np.linalg.norm(x - y) / (2 * sigma))
rho_gauss = lambda x, y: 2 - 2 * np.exp(-np.linalg.norm(x - y)**2 /
(2 * (sigma)**2))
# normalize data
G = eclust.kernel_matrix(data, rho_gauss)
#G = energy.eclust.kernel_matrix(data, rho_gauss)
#G = energy.eclust.kernel_matrix(data, rho_exp)
r = []
r.append(wrapper.kmeans(3, data, run_times=5))
r.append(wrapper.gmm(3, data, run_times=5))
r.append(wrapper.spectral_clustering(3, data, G, run_times=5))
r.append(wrapper.spectral(3, data, G, run_times=5))
#r.append(wrapper.kernel_kmeans(3, data, G, run_times=5, ini='random'))
#r.append(wrapper.kernel_kmeans(3, data, G, run_times=5, ini='k-means++'))
r.append(wrapper.kernel_kmeans(3, data, G, run_times=5, ini='spectral'))
#r.append(wrapper.kernel_kgroups(3,data,G,run_times=5, ini='random'))
#r.append(wrapper.kernel_kgroups(3,data,G,run_times=5, ini='k-means++'))
r.append(wrapper.kernel_kgroups(3, data, G, run_times=5, ini='spectral'))
s1 = np.array([[1,0], [0,20]])
s2 = np.array([[1,0], [0,20]])
r1 = 1
r2 = 3
eps = 0.2
r = []
for _ in range(num_experiments):
n1, n2 = np.random.multinomial(total_points, [0.5, 0.5])
#X, z = data.multivariate_normal([m1, m2], [s1, s2], [n1, n2])
X, z = data.circles([r1, r2], [eps, eps], [n1, n2])
#G = eclust.kernel_matrix(X,
# lambda x, y: 2 - 2*np.exp(-np.linalg.norm(x-y)/2/2))
G = eclust.kernel_matrix(X,
lambda x, y: 2 - 2*np.exp(-np.linalg.norm(x-y)**2/2/1))
row = []
zh = wrapper.kmeans(k, X)
a = metric.accuracy(z, zh)
row.append(a)
zh = wrapper.gmm(k, X)
a = metric.accuracy(z, zh)
row.append(a)
zh = wrapper.spectral_clustering(k, X, G)
a = metric.accuracy(z, zh)
row.append(a)
zh = wrapper.kernel_kmeans(k, X, G, ini='random')
