def initialize_gmm_from_kmeans(self, samples):
from machine_learning import KMeans
kmeans = KMeans(n_clusters = len(self.S), init = 'Katsavounidis', verbosity = 1)
print("HMM.initializegmm_from_kmeans() begins the fit", flush = True)
kmeans.fit(numpy.vstack(samples))
print("HMM.initializegmm_from_kmeans() ends the fit", flush = True)
for k in range(len(self.S)):
self.S[k].gmm.initialize_from_centroids(kmeans.cluster_centers_[k])
def __init__(self, X):
self.num_clusters = 17
self.X_ = X
self.data_ = numpy.zeros((2, X.shape[0] + self.num_clusters))
self.data_[0, :len(X)] = X[:, 0]
self.data_[1, :len(X)] = X[:, 1]
self.kmeans = KMeans(self.num_clusters, init='random', max_iter=1)
#self.kmeans = KMeans( self.num_clusters, init='Katsavounidis', max_iter=1 )
self.Y_ = numpy.ones(len(X), dtype='int')
self.fig_, self.ax_ = pyplot.subplots()
self.ani_ = animation.FuncAnimation(self.fig_,
self.update_figure,
self.generate_data,
init_func=self.setup_plot,
interval=1000,
blit=True,
repeat=False)
self.changes_ = 0
class AnimatedClustering(object):
def __init__(self, X, clust):
self.num_clusters = clust
self.X_ = X
self.data_ = numpy.zeros((2, X.shape[0] + self.num_clusters))
self.data_[0, :len(X)] = X[:, 0]
self.data_[1, :len(X)] = X[:, 1]
#self.kmeans = KMeans( self.num_clusters, init='random', max_iter=1 )
self.kmeans = KMeans(self.num_clusters,
init='Katsavounidis',
max_iter=1)
self.Y_ = numpy.ones(len(X), dtype='int')
self.fig_, self.ax_ = pyplot.subplots()
self.ani_ = animation.FuncAnimation(self.fig_,
self.update_figure,
self.generate_data,
init_func=self.setup_plot,
interval=1000,
blit=True,
repeat=False)
self.changes_ = 0
def setup_plot(self):
self.colour_ = numpy.ones(len(self.X_) + self.num_clusters) * 3
self.sizes_ = numpy.ones(self.X_.shape[0] + self.num_clusters) * 30
self.colour_[len(self.X_):] = self.num_clusters + 1
self.sizes_[len(self.X_):] = 100
self.scat_ = self.ax_.scatter(self.data_[0, :],
self.data_[1, :],
c=self.colour_,
s=self.sizes_,
marker='o',
edgecolors='none',
animated=False)
return self.scat_,
def generate_data(self):
self.changes_ = len(self.X_)
self.kmeans.fit(self.X_)
#self.kmeans.lloyd( self.X_, num_iter=1 )
self.colour_[:len(self.X_)] = self.kmeans.predict(self.X_)
self.data_[0, len(self.X_):] = self.kmeans.cluster_centers_[:, 0]
self.data_[1, len(self.X_):] = self.kmeans.cluster_centers_[:, 1]
yield self.data_, self.colour_, self.sizes_
while self.changes_ > 0:
self.changes_ = self.kmeans.fit_iteration(self.X_)
self.Y_[:] = self.kmeans.predict(self.X_)
self.colour_[:len(self.Y_)] = self.Y_[:]
self.data_[0, len(self.X_):] = self.kmeans.cluster_centers_[:, 0]
self.data_[1, len(self.X_):] = self.kmeans.cluster_centers_[:, 1]
yield self.data_, self.colour_, self.sizes_
def update_figure(self, generated_data):
data, colour, sizes = generated_data
print("clusters = %d changes = %12d J = %20.8f %.8f" %
(self.num_clusters, self.changes_, self.kmeans.J,
self.kmeans.improvement()))
pyplot.clf()
#pyplot.set_axis_bgcolor( 'white' )
self.scat_ = self.ax_.scatter(self.data_[0, :],
self.data_[1, :],
c=colour,
s=sizes,
marker='o',
edgecolors='none',
animated=False)
#pyplot.draw()
return self.scat_,
def show(self):
pyplot.show()
N_2 = 5000
X_1 = numpy.random.rand(N_1, 2)
X_1[:, 0] *= 10
X_1[:, 1] *= 3
X_1[:, 0] += 2
X_1[:, 1] += 2
X_2 = numpy.random.rand(N_2, 2)
X_2[:, 0] *= 3
X_2[:, 1] *= 10
X_2[:, 0] += 8
X_2[:, 1] += 1
# k-means codebook generators
lloyd_1 = KMeans(n_clusters=K,
verbosity=1,
modality='Lloyd',
init='KMeans++')
lloyd_1.epsilon = 1.0e-9
lloyd_2 = KMeans(n_clusters=K,
verbosity=1,
modality='Lloyd',
init='KMeans++')
lloyd_2.epsilon = 1.0e-9
# k-means codebook estimation
lloyd_1.fit(X_1)
lloyd_2.fit(X_2)
def assignment_1(p, h):
h_2 = h**2
density_1 = numpy.exp(-0.5 * ((
import numpy
from sklearn.datasets import make_blobs
from sklearn.metrics import silhouette_score, calinski_harabasz_score, davies_bouldin_score
from matplotlib import pyplot
from machine_learning import KMeans
if __name__ == '__main__':
K = 10
K2 = 7 # set a smaller value to study the behaviour of the different algorithms when using make_blobs()
lloyd = KMeans(n_clusters=K,
verbosity=1,
modality='Lloyd',
init='KMeans++')
lloyd.epsilon = 1.0e-9
kmediods = KMeans(n_clusters=K,
verbosity=1,
modality='k-Mediods',
init='KMeans++')
lloyd.epsilon = 1.0e-8
original_kmeans = KMeans(n_clusters=K,
verbosity=1,
modality='original-k-Means')
selective_splitting = KMeans(n_clusters=K,
verbosity=1,
modality='Lloyd',
init='KMeans++')
selective_splitting.epsilon = 1.0e-9