def initialisation(self):
"""
This is the initialisation of the parameters of the gaussians
On the choice of using the Pycluster library see http://stackoverflow.com/questions/1545606/python-k-means-algorithm
"""
labels, errors, found = knn(self.data, self.number_of_gaussians)
# grouped data example
# 0th element of grouped data represents the 0th cluster and consists
# of the frames belonging to that cluster
group_data = [[] for i in xrange(self.number_of_gaussians)]
for index, label in enumerate(labels):
group_data[label].append(self.data[index].tolist())
total_weight = sum(map(lambda x: len(x), group_data))
dimension = (self.data.shape)[1]
for everyMatrix in group_data:
everyMatrix = np.array(everyMatrix)
mean_vector = mean(everyMatrix)
covariance_matrix = covariance(everyMatrix)
gaussian = MultiVariateGaussian(mean_vector, covariance_matrix,
dimension)
weight = float(len(everyMatrix)) / total_weight
self.gaussians.append({'gaussian': gaussian, 'weight': weight})
def initialisation(self):
"""
This is the initialisation of the parameters of the gaussians
On the choice of using the Pycluster library see http://stackoverflow.com/questions/1545606/python-k-means-algorithm
"""
labels, errors, found = knn(self.data, self.number_of_gaussians)
# grouped data example
# 0th element of grouped data represents the 0th cluster and consists
# of the frames belonging to that cluster
group_data = [[] for i in xrange(self.number_of_gaussians)]
for index, label in enumerate(labels):
group_data[label].append(self.data[index].tolist())
total_weight = sum(map(lambda x: len(x), group_data))
dimension = (self.data.shape)[1]
for everyMatrix in group_data:
everyMatrix = np.array(everyMatrix)
mean_vector = mean(everyMatrix)
covariance_matrix = covariance(everyMatrix)
gaussian = MultiVariateGaussian(
mean_vector, covariance_matrix, dimension)
weight = float(len(everyMatrix)) / total_weight
self.gaussians.append({'gaussian': gaussian, 'weight': weight})
def getGeneralisedSampleVariance(self):
return self.generalisedSampleVariance
def setGeneralisedSampleVariance(self):
self.generalisedSampleVariance = np.linalg.det(
self.getCovarianceMatrix())
def getPrecisionMatrix(self):
return self.precision_matrix
def setPrecisionMatrix(self):
self.precision_matrix = np.fabs(
np.linalg.inv(self.getCovarianceMatrix()))
def probabilityDensityFunction(self, data):
variation_from_mean = data - self.mean_vector
self.MahalanobisDistance = (
(variation_from_mean) * self.precision_matrix) * (variation_from_mean.T)
self.denominator = (
(2 * np.pi) ** (self.number_of_random_variables / 2)) * (self.generalisedSampleVariance ** 0.5)
return (np.exp((self.MahalanobisDistance * (-0.5))) / self.denominator).item(0)
if __name__ == '__main__':
data = test.get_test_data()
mean = mean(data)
covariance = covariance(data)
print "covariance_matrix", covariance
gmm_cluster = MultiVariateGaussian(mean, covariance, 2)
print gmm_cluster.probabilityDensityFunction(np.matrix([3.0, 4.0]))
