def score_one_clustering(X, truelabels, num_components, num_iterations):
"""Cluster a dataset and evaluate it using Conditional Entropy"""
#scipy's builtin K-means is very slow, use mpi-version instead.
#from scipy.cluster.vq import kmeans,vq
#clst,dist = kmeans(X, num_components, NUM_ITERATIONS)
#labels,dist = vq(X, clst)
clst,dist,labels = _mpi_kmeans.kmeans(X, num_components, 200, num_iterations)
print truelabels
print labels-1
return condentropy(truelabels,labels-1)
def score_one_clustering(X, truelabels, num_components, num_iterations):
"""Cluster a dataset and evaluate it using Conditional Entropy"""
#scipy's builtin K-means is very slow, use mpi-version instead.
#from scipy.cluster.vq import kmeans,vq
#clst,dist = kmeans(X, num_components, NUM_ITERATIONS)
#labels,dist = vq(X, clst)
clst, dist, labels = _mpi_kmeans.kmeans(X, num_components, 200,
num_iterations)
print truelabels
print labels - 1
return condentropy(truelabels, labels - 1)
def process(self, data):
# Perform KMeans clustering
if pynopticon.verbosity > 0:
print "Performing kmeans clustering with k=%i..." % self.numClusters
# Sample from data, randomly take self.sampleFromData percent of the vectors
samplePoints = numpy.random.permutation(range(len(data)))[0:int(round(len(data)*self.sampleFromData))]
data = numpy.array(data, dtype=c_double)
data = data[samplePoints,:]
self.codebook, self.dist, self.labels = _mpi_kmeans.kmeans(numpy.array(data, dtype=c_double), self.numClusters, self.maxiter, self.numruns)
return self.codebook
