def test_initialise():
X = [[1, 2, 3], [4, 5, 6]]
M = [[0, 1, 1], [1, 0, 1]]
K = 2
seed = 0
kmeans = KMeans(X, M, K)
kmeans.initialise(seed)
mins = [4.0, 2.0, 3.0]
maxs = [4.0, 2.0, 6.0]
assert numpy.array_equal(mins, kmeans.mins)
assert numpy.array_equal(maxs, kmeans.maxs)
mask_centroids = [[1, 1, 1], [1, 1, 1]]
assert numpy.array_equal(mask_centroids, kmeans.mask_centroids)
cluster_assignments = [-1, -1]
assert numpy.array_equal(cluster_assignments, kmeans.cluster_assignments)
centroids = [[4.0, 2.0, 4.2617147424925346],
[4.0, 2.0, 4.2148024123512426]]
assert numpy.array_equal(centroids, kmeans.centroids)
distances = [0, 0]
assert numpy.array_equal(distances, kmeans.distances)
def test_assignment():
X = numpy.array([[1.0, 2.0, 3.0], [4.0, 5.0, 6.0], [7.0, 8.0, 9.0]])
M = numpy.array([[1, 1, 0], [0, 1, 0], [1, 1, 1]])
K = 2
kmeans = KMeans(X, M, K)
# Test change - new closest clusters are [0,0,1] - see test_closest_cluster
centroids = [[1.0, 3.0, 1.0], [2.0, 1.0, 3.0]]
mask_centroids = [[0, 1, 1], [1, 1, 0]]
cluster_assignments = [0, 1, 1]
kmeans.centroids = centroids
kmeans.mask_centroids = mask_centroids
kmeans.cluster_assignments = cluster_assignments
change = kmeans.assignment()
assert change == True
assert numpy.array_equal([0, 0, 1], kmeans.cluster_assignments)
assert numpy.array_equal([[0, 1], [2]], kmeans.data_point_assignments)
# Test no change
centroids = [[1.0, 3.0, 1.0], [2.0, 1.0, 3.0]]
mask_centroids = [[0, 1, 1], [1, 1, 0]]
cluster_assignments = [0, 0, 1]
kmeans.centroids = centroids
kmeans.mask_centroids = mask_centroids
kmeans.cluster_assignments = cluster_assignments
change = kmeans.assignment()
assert change == False
assert numpy.array_equal([0, 0, 1], kmeans.cluster_assignments)
assert numpy.array_equal([[0, 1], [2]], kmeans.data_point_assignments)
def test_initialise():
X = [[1,2,3],[4,5,6]]
M = [[0,1,1],[1,0,1]]
K = 2
seed = 0
kmeans = KMeans(X,M,K)
kmeans.initialise(seed)
mins = [4.0,2.0,3.0]
maxs = [4.0,2.0,6.0]
assert numpy.array_equal(mins,kmeans.mins)
assert numpy.array_equal(maxs,kmeans.maxs)
mask_centroids = [[1,1,1],[1,1,1]]
assert numpy.array_equal(mask_centroids,kmeans.mask_centroids)
cluster_assignments = [-1,-1]
assert numpy.array_equal(cluster_assignments,kmeans.cluster_assignments)
centroids = [[4.0,2.0,4.2617147424925346],[4.0,2.0,4.2148024123512426]]
assert numpy.array_equal(centroids,kmeans.centroids)
distances = [0,0]
assert numpy.array_equal(distances,kmeans.distances)
def initialise(self,init_S='random',init_FG='random'):
assert init_S in ['random','exp'], "Unknown initialisation option for S: %s. Should be 'random' or 'exp'." % init_S
assert init_FG in ['random','exp','kmeans'], "Unknown initialisation option for S: %s. Should be 'random', 'exp', or 'kmeans." % init_FG
self.S = 1./self.lambdaS
if init_S == 'random':
for k,l in itertools.product(xrange(0,self.K),xrange(0,self.L)):
self.S[k,l] = exponential_draw(self.lambdaS[k,l])
self.F, self.G = 1./self.lambdaF, 1./self.lambdaG
if init_FG == 'random':
for i,k in itertools.product(xrange(0,self.I),xrange(0,self.K)):
self.F[i,k] = exponential_draw(self.lambdaF[i,k])
for j,l in itertools.product(xrange(0,self.J),xrange(0,self.L)):
self.G[j,l] = exponential_draw(self.lambdaG[j,l])
elif init_FG == 'kmeans':
print "Initialising F using KMeans."
kmeans_F = KMeans(self.R,self.M,self.K)
kmeans_F.initialise()
kmeans_F.cluster()
self.F = kmeans_F.clustering_results + 0.2
print "Initialising G using KMeans."
kmeans_G = KMeans(self.R.T,self.M.T,self.L)
kmeans_G.initialise()
kmeans_G.cluster()
self.G = kmeans_G.clustering_results + 0.2
self.tau = gamma_mode(self.alpha_s(), self.beta_s())
def test_assignment():
X = numpy.array([[1.0,2.0,3.0],[4.0,5.0,6.0],[7.0,8.0,9.0]])
M = numpy.array([[1,1,0],[0,1,0],[1,1,1]])
K = 2
kmeans = KMeans(X,M,K)
# Test change - new closest clusters are [0,0,1] - see test_closest_cluster
centroids = [[1.0,3.0,1.0],[2.0,1.0,3.0]]
mask_centroids = [[0,1,1],[1,1,0]]
cluster_assignments = [0,1,1]
kmeans.centroids = centroids
kmeans.mask_centroids = mask_centroids
kmeans.cluster_assignments = cluster_assignments
change = kmeans.assignment()
assert change == True
assert numpy.array_equal([0,0,1],kmeans.cluster_assignments)
assert numpy.array_equal([[0,1],[2]],kmeans.data_point_assignments)
# Test no change
centroids = [[1.0,3.0,1.0],[2.0,1.0,3.0]]
mask_centroids = [[0,1,1],[1,1,0]]
cluster_assignments = [0,0,1]
kmeans.centroids = centroids
kmeans.mask_centroids = mask_centroids
kmeans.cluster_assignments = cluster_assignments
change = kmeans.assignment()
assert change == False
assert numpy.array_equal([0,0,1],kmeans.cluster_assignments)
assert numpy.array_equal([[0,1],[2]],kmeans.data_point_assignments)
def test_create_matrix():
X = numpy.array([[1.0, 2.0, 3.0], [4.0, 5.0, 6.0], [7.0, 8.0, 9.0]])
M = numpy.array([[1, 1, 0], [0, 1, 0], [1, 1, 1]])
K = 2
kmeans = KMeans(X, M, K)
kmeans.cluster_assignments = numpy.array([1, 0, 1])
kmeans.create_matrix()
expected_clustering_results = [[0, 1], [1, 0], [0, 1]]
clustering_results = kmeans.clustering_results
assert numpy.array_equal(expected_clustering_results, clustering_results)
def test_create_matrix():
X = numpy.array([[1.0,2.0,3.0],[4.0,5.0,6.0],[7.0,8.0,9.0]])
M = numpy.array([[1,1,0],[0,1,0],[1,1,1]])
K = 2
kmeans = KMeans(X,M,K)
kmeans.cluster_assignments = numpy.array([1,0,1])
kmeans.create_matrix()
expected_clustering_results = [[0,1],[1,0],[0,1]]
clustering_results = kmeans.clustering_results
assert numpy.array_equal(expected_clustering_results,clustering_results)
def test_random_cluster_centroid():
X = [[1,2,3],[4,5,6]]
M = [[0,1,1],[1,0,1]]
K = 2
kmeans = KMeans(X,M,K)
kmeans.mins = [4.0,2.0,3.0]
kmeans.maxs = [4.0,2.0,6.0]
expected_centroid = [4.0,2.0,4.2617147424925346]
random.seed(0)
centroid = kmeans.random_cluster_centroid()
assert numpy.array_equal(expected_centroid,centroid)
def test_random_cluster_centroid():
X = [[1, 2, 3], [4, 5, 6]]
M = [[0, 1, 1], [1, 0, 1]]
K = 2
kmeans = KMeans(X, M, K)
kmeans.mins = [4.0, 2.0, 3.0]
kmeans.maxs = [4.0, 2.0, 6.0]
expected_centroid = [4.0, 2.0, 4.2617147424925346]
random.seed(0)
centroid = kmeans.random_cluster_centroid()
assert numpy.array_equal(expected_centroid, centroid)
def initialise(self,init_S='random',init_FG='random',tauFSG={}):
self.tauF = tauFSG['tauF'] if 'tauF' in tauFSG else numpy.ones((self.I,self.K))
self.tauS = tauFSG['tauS'] if 'tauS' in tauFSG else numpy.ones((self.K,self.L))
self.tauG = tauFSG['tauG'] if 'tauG' in tauFSG else numpy.ones((self.J,self.L))
assert init_S in ['exp','random'], "Unrecognised init option for S: %s." % init_S
self.muS = 1./self.lambdaS
if init_S == 'random':
for k,l in itertools.product(xrange(0,self.K),xrange(0,self.L)):
self.muS[k,l] = exponential_draw(self.lambdaS[k,l])
assert init_FG in ['exp','random','kmeans'], "Unrecognised init option for F,G: %s." % init_FG
self.muF, self.muG = 1./self.lambdaF, 1./self.lambdaG
if init_FG == 'random':
for i,k in itertools.product(xrange(0,self.I),xrange(0,self.K)):
self.muF[i,k] = exponential_draw(self.lambdaF[i,k])
for j,l in itertools.product(xrange(0,self.J),xrange(0,self.L)):
self.muG[j,l] = exponential_draw(self.lambdaG[j,l])
elif init_FG == 'kmeans':
print "Initialising F using KMeans."
kmeans_F = KMeans(self.R,self.M,self.K)
kmeans_F.initialise()
kmeans_F.cluster()
self.muF = kmeans_F.clustering_results #+ 0.2
print "Initialising G using KMeans."
kmeans_G = KMeans(self.R.T,self.M.T,self.L)
kmeans_G.initialise()
kmeans_G.cluster()
self.muG = kmeans_G.clustering_results #+ 0.2
# Initialise the expectations and variances
self.expF, self.varF = numpy.zeros((self.I,self.K)), numpy.zeros((self.I,self.K))
self.expS, self.varS = numpy.zeros((self.K,self.L)), numpy.zeros((self.K,self.L))
self.expG, self.varG = numpy.zeros((self.J,self.L)), numpy.zeros((self.J,self.L))
for k in range(0,self.K):
self.update_exp_F(k)
for k,l in itertools.product(xrange(0,self.K),xrange(0,self.L)):
self.update_exp_S(k,l)
for l in range(0,self.L):
self.update_exp_G(l)
# Initialise tau using the updates
self.update_tau()
#self.alpha_s, self.beta_s = self.alpha, self.beta
self.update_exp_tau()
def test_find_known_coordinate_values():
# Normal test case
X = numpy.array([[1.0,2.0,3.0],[4.0,5.0,6.0],[7.0,8.0,9.0]])
M = numpy.array([[1,1,0],[0,1,0],[1,1,1]])
K = 2
kmeans = KMeans(X,M,K)
kmeans.data_point_assignments = numpy.array([[0,1],[2]]) #points 0,1 to cluster 0, point 2 to cluster 1
expected_lists_known_coordinate_values_0 = [[1.0],[2.0,5.0],[]]
expected_lists_known_coordinate_values_1 = [[7.0],[8.0],[9.0]]
lists_known_coordinate_values_0 = kmeans.find_known_coordinate_values(0)
lists_known_coordinate_values_1 = kmeans.find_known_coordinate_values(1)
assert numpy.array_equal(expected_lists_known_coordinate_values_0,lists_known_coordinate_values_0)
assert numpy.array_equal(expected_lists_known_coordinate_values_1,lists_known_coordinate_values_1)
# Cluster without any points
X = numpy.array([[1.0,2.0,3.0],[4.0,5.0,6.0],[7.0,8.0,9.0]])
M = numpy.array([[1,1,0],[0,1,0],[1,1,1]])
K = 2
kmeans = KMeans(X,M,K)
kmeans.data_point_assignments = numpy.array([[0,1,2],[]]) #points 0,1,2 to cluster 0, none to cluster 1
expected_lists_known_coordinate_values_0 = [[1.0,7.0],[2.0,5.0,8.0],[9.0]]
expected_lists_known_coordinate_values_1 = None
lists_known_coordinate_values_0 = kmeans.find_known_coordinate_values(0)
lists_known_coordinate_values_1 = kmeans.find_known_coordinate_values(1)
assert numpy.array_equal(expected_lists_known_coordinate_values_0,lists_known_coordinate_values_0)
assert numpy.array_equal(expected_lists_known_coordinate_values_1,lists_known_coordinate_values_1)
def test_compute_MSE():
# Test case: no overlap
X = numpy.ones((1, 5))
M = numpy.ones((1, 5))
K = 1
x1 = [1.0, 2.0, 3.0, 4.0, 5.0]
x2 = [5.0, 4.5, 3.0, 2.5, 1.0]
mask1 = [0, 1, 1, 0, 0]
mask2 = [1, 0, 0, 0, 1]
kmeans = KMeans(X, M, K)
output = kmeans.compute_MSE(x1, x2, mask1, mask2)
assert output == None
# Overlap
mask1 = [1, 1, 1, 0, 1]
mask2 = [0, 1, 1, 1, 1]
expected_output = (2.5**2 + 4.0**2) / 3.0
output = kmeans.compute_MSE(x1, x2, mask1, mask2)
assert expected_output == output
def test_compute_MSE():
# Test case: no overlap
X = numpy.ones((1,5))
M = numpy.ones((1,5))
K = 1
x1 = [1.0,2.0,3.0,4.0,5.0]
x2 = [5.0,4.5,3.0,2.5,1.0]
mask1 = [0,1,1,0,0]
mask2 = [1,0,0,0,1]
kmeans = KMeans(X,M,K)
output = kmeans.compute_MSE(x1,x2,mask1,mask2)
assert output == None
# Overlap
mask1 = [1,1,1,0,1]
mask2 = [0,1,1,1,1]
expected_output = ( 2.5**2 + 4.0**2 ) / 3.0
output = kmeans.compute_MSE(x1,x2,mask1,mask2)
assert expected_output == output
def test_find_point_furthest_away():
X = numpy.array([[1.0, 2.0, 3.0], [4.0, 5.0, 6.0], [7.0, 8.0, 9.0]])
M = numpy.array([[1, 1, 0], [0, 1, 0], [1, 1, 1]])
K = 2
kmeans = KMeans(X, M, K)
# Equal distance for point 0
centroids = [[1.0, 3.0, 1.0], [2.0, 1.0, 3.0]]
mask_centroids = [[0, 1, 1], [1, 1, 0]]
kmeans.centroids = centroids
kmeans.mask_centroids = mask_centroids
kmeans.closest_cluster(X[0], 0, M[0]) # MSE = 1.0 vs 1.0
kmeans.closest_cluster(X[1], 1, M[1]) # MSE = 4.0 vs 16.0
kmeans.closest_cluster(X[2], 2, M[2]) # MSE = 44.5 vs 37.0
expected_furthest_away = 2
furthest_away = kmeans.find_point_furthest_away()
assert expected_furthest_away == furthest_away
def initialise(self,init_S='random',init_FG='random',expo_prior=1.):
assert init_S in ['ones','random','exponential'], "Unrecognised init option for S: %s." % init_S
assert init_FG in ['ones','random','exponential','kmeans'], "Unrecognised init option for F,G: %s." % init_FG
if init_S == 'ones':
self.S = numpy.ones((self.K,self.L))
elif init_S == 'random':
self.S = numpy.random.rand(self.K,self.L)
elif init_S == 'exponential':
self.S = numpy.empty((self.K,self.L))
for k,l in itertools.product(xrange(0,self.K),xrange(0,self.L)):
self.S[k,l] = exponential_draw(expo_prior)
if init_FG == 'ones':
self.F = numpy.ones((self.I,self.K))
self.G = numpy.ones((self.J,self.L))
elif init_FG == 'random':
self.F = numpy.random.rand(self.I,self.K)
self.G = numpy.random.rand(self.J,self.L)
elif init_FG == 'exponential':
self.F = numpy.empty((self.I,self.K))
self.G = numpy.empty((self.J,self.L))
for i,k in itertools.product(xrange(0,self.I),xrange(0,self.K)):
self.F[i,k] = exponential_draw(expo_prior)
for j,l in itertools.product(xrange(0,self.J),xrange(0,self.L)):
self.G[j,l] = exponential_draw(expo_prior)
elif init_FG == 'kmeans':
print "Initialising F using KMeans."
kmeans_F = KMeans(self.R,self.M,self.K)
kmeans_F.initialise()
kmeans_F.cluster()
self.F = kmeans_F.clustering_results + 0.2
print "Initialising G using KMeans."
kmeans_G = KMeans(self.R.T,self.M.T,self.L)
kmeans_G.initialise()
kmeans_G.cluster()
self.G = kmeans_G.clustering_results + 0.2
def test_find_point_furthest_away():
X = numpy.array([[1.0,2.0,3.0],[4.0,5.0,6.0],[7.0,8.0,9.0]])
M = numpy.array([[1,1,0],[0,1,0],[1,1,1]])
K = 2
kmeans = KMeans(X,M,K)
# Equal distance for point 0
centroids = [[1.0,3.0,1.0],[2.0,1.0,3.0]]
mask_centroids = [[0,1,1],[1,1,0]]
kmeans.centroids = centroids
kmeans.mask_centroids = mask_centroids
kmeans.closest_cluster(X[0],0,M[0]) # MSE = 1.0 vs 1.0
kmeans.closest_cluster(X[1],1,M[1]) # MSE = 4.0 vs 16.0
kmeans.closest_cluster(X[2],2,M[2]) # MSE = 44.5 vs 37.0
expected_furthest_away = 2
furthest_away = kmeans.find_point_furthest_away()
assert expected_furthest_away == furthest_away
def test_closest_cluster():
X = numpy.array([[1.0, 2.0, 3.0], [4.0, 5.0, 6.0], [7.0, 8.0, 9.0]])
M = numpy.array([[1, 1, 0], [0, 1, 0], [1, 1, 1]])
K = 2
kmeans = KMeans(X, M, K)
# Equal distance for point 0
centroids = [[1.0, 3.0, 1.0], [2.0, 1.0, 3.0]]
mask_centroids = [[0, 1, 1], [1, 1, 0]]
kmeans.centroids = centroids
kmeans.mask_centroids = mask_centroids
expected_closest_cluster_0 = 0 # MSE = 1.0 vs 1.0
expected_closest_cluster_1 = 0 # MSE = 4.0 vs 16.0
expected_closest_cluster_2 = 1 # MSE = 44.5 vs 37.0
closest_cluster_0 = kmeans.closest_cluster(X[0], 0, M[0])
closest_cluster_1 = kmeans.closest_cluster(X[1], 1, M[1])
closest_cluster_2 = kmeans.closest_cluster(X[2], 2, M[2])
assert expected_closest_cluster_0 == closest_cluster_0
assert expected_closest_cluster_1 == closest_cluster_1
assert expected_closest_cluster_2 == closest_cluster_2
# Also test whether the distances are set correctly
expected_distances = [1.0, 4.0, 37.0]
distances = kmeans.distances
assert numpy.array_equal(expected_distances, distances)
# Test when all MSEs return None (impossible but still testing behaviour)
centroids = numpy.ones((2, 3))
mask_centroids = [[0, 0, 1], [0, 0, 0]]
kmeans.centroids = centroids
kmeans.mask_centroids = mask_centroids
expected_closest_cluster = 1
closest_cluster = kmeans.closest_cluster(X[0], 0, M[0])
assert expected_closest_cluster == closest_cluster
def test_closest_cluster():
X = numpy.array([[1.0,2.0,3.0],[4.0,5.0,6.0],[7.0,8.0,9.0]])
M = numpy.array([[1,1,0],[0,1,0],[1,1,1]])
K = 2
kmeans = KMeans(X,M,K)
# Equal distance for point 0
centroids = [[1.0,3.0,1.0],[2.0,1.0,3.0]]
mask_centroids = [[0,1,1],[1,1,0]]
kmeans.centroids = centroids
kmeans.mask_centroids = mask_centroids
expected_closest_cluster_0 = 0 # MSE = 1.0 vs 1.0
expected_closest_cluster_1 = 0 # MSE = 4.0 vs 16.0
expected_closest_cluster_2 = 1 # MSE = 44.5 vs 37.0
closest_cluster_0 = kmeans.closest_cluster(X[0],0,M[0])
closest_cluster_1 = kmeans.closest_cluster(X[1],1,M[1])
closest_cluster_2 = kmeans.closest_cluster(X[2],2,M[2])
assert expected_closest_cluster_0 == closest_cluster_0
assert expected_closest_cluster_1 == closest_cluster_1
assert expected_closest_cluster_2 == closest_cluster_2
# Also test whether the distances are set correctly
expected_distances = [1.0,4.0,37.0]
distances = kmeans.distances
assert numpy.array_equal(expected_distances,distances)
# Test when all MSEs return None (impossible but still testing behaviour)
centroids = numpy.ones((2,3))
mask_centroids = [[0,0,1],[0,0,0]]
kmeans.centroids = centroids
kmeans.mask_centroids = mask_centroids
expected_closest_cluster = 1
closest_cluster = kmeans.closest_cluster(X[0],0,M[0])
assert expected_closest_cluster == closest_cluster
def test_find_known_coordinate_values():
# Normal test case
X = numpy.array([[1.0, 2.0, 3.0], [4.0, 5.0, 6.0], [7.0, 8.0, 9.0]])
M = numpy.array([[1, 1, 0], [0, 1, 0], [1, 1, 1]])
K = 2
kmeans = KMeans(X, M, K)
kmeans.data_point_assignments = numpy.array(
[[0, 1], [2]]) #points 0,1 to cluster 0, point 2 to cluster 1
expected_lists_known_coordinate_values_0 = [[1.0], [2.0, 5.0], []]
expected_lists_known_coordinate_values_1 = [[7.0], [8.0], [9.0]]
lists_known_coordinate_values_0 = kmeans.find_known_coordinate_values(0)
lists_known_coordinate_values_1 = kmeans.find_known_coordinate_values(1)
assert numpy.array_equal(expected_lists_known_coordinate_values_0,
lists_known_coordinate_values_0)
assert numpy.array_equal(expected_lists_known_coordinate_values_1,
lists_known_coordinate_values_1)
# Cluster without any points
X = numpy.array([[1.0, 2.0, 3.0], [4.0, 5.0, 6.0], [7.0, 8.0, 9.0]])
M = numpy.array([[1, 1, 0], [0, 1, 0], [1, 1, 1]])
K = 2
kmeans = KMeans(X, M, K)
kmeans.data_point_assignments = numpy.array(
[[0, 1, 2], []]) #points 0,1,2 to cluster 0, none to cluster 1
expected_lists_known_coordinate_values_0 = [[1.0, 7.0], [2.0, 5.0, 8.0],
[9.0]]
expected_lists_known_coordinate_values_1 = None
lists_known_coordinate_values_0 = kmeans.find_known_coordinate_values(0)
lists_known_coordinate_values_1 = kmeans.find_known_coordinate_values(1)
assert numpy.array_equal(expected_lists_known_coordinate_values_0,
lists_known_coordinate_values_0)
assert numpy.array_equal(expected_lists_known_coordinate_values_1,
lists_known_coordinate_values_1)
def initialise(self, init_S='random', init_FG='random', tauFSG={}):
self.tauF = tauFSG['tauF'] if 'tauF' in tauFSG else numpy.ones(
(self.I, self.K))
self.tauS = tauFSG['tauS'] if 'tauS' in tauFSG else numpy.ones(
(self.K, self.L))
self.tauG = tauFSG['tauG'] if 'tauG' in tauFSG else numpy.ones(
(self.J, self.L))
assert init_S in ['exp', 'random'
], "Unrecognised init option for S: %s." % init_S
self.muS = 1. / self.lambdaS
if init_S == 'random':
for k, l in itertools.product(xrange(0, self.K), xrange(0,
self.L)):
self.muS[k, l] = exponential_draw(self.lambdaS[k, l])
assert init_FG in ['exp', 'random', 'kmeans'
], "Unrecognised init option for F,G: %s." % init_FG
self.muF, self.muG = 1. / self.lambdaF, 1. / self.lambdaG
if init_FG == 'random':
for i, k in itertools.product(xrange(0, self.I), xrange(0,
self.K)):
self.muF[i, k] = exponential_draw(self.lambdaF[i, k])
for j, l in itertools.product(xrange(0, self.J), xrange(0,
self.L)):
self.muG[j, l] = exponential_draw(self.lambdaG[j, l])
elif init_FG == 'kmeans':
print "Initialising F using KMeans."
kmeans_F = KMeans(self.R, self.M, self.K)
kmeans_F.initialise()
kmeans_F.cluster()
self.muF = kmeans_F.clustering_results #+ 0.2
print "Initialising G using KMeans."
kmeans_G = KMeans(self.R.T, self.M.T, self.L)
kmeans_G.initialise()
kmeans_G.cluster()
self.muG = kmeans_G.clustering_results #+ 0.2
# Initialise the expectations and variances
self.expF, self.varF = numpy.zeros((self.I, self.K)), numpy.zeros(
(self.I, self.K))
self.expS, self.varS = numpy.zeros((self.K, self.L)), numpy.zeros(
(self.K, self.L))
self.expG, self.varG = numpy.zeros((self.J, self.L)), numpy.zeros(
(self.J, self.L))
for k in range(0, self.K):
self.update_exp_F(k)
for k, l in itertools.product(xrange(0, self.K), xrange(0, self.L)):
self.update_exp_S(k, l)
for l in range(0, self.L):
self.update_exp_G(l)
# Initialise tau using the updates
self.update_tau()
#self.alpha_s, self.beta_s = self.alpha, self.beta
self.update_exp_tau()
def initialise(self, init_S='random', init_FG='random'):
assert init_S in [
'random', 'exp'
], "Unknown initialisation option for S: %s. Should be 'random' or 'exp'." % init_S
assert init_FG in [
'random', 'exp', 'kmeans'
], "Unknown initialisation option for S: %s. Should be 'random', 'exp', or 'kmeans." % init_FG
self.S = 1. / self.lambdaS
if init_S == 'random':
for k, l in itertools.product(xrange(0, self.K), xrange(0,
self.L)):
self.S[k, l] = exponential_draw(self.lambdaS[k, l])
self.F, self.G = 1. / self.lambdaF, 1. / self.lambdaG
if init_FG == 'random':
for i, k in itertools.product(xrange(0, self.I), xrange(0,
self.K)):
self.F[i, k] = exponential_draw(self.lambdaF[i, k])
for j, l in itertools.product(xrange(0, self.J), xrange(0,
self.L)):
self.G[j, l] = exponential_draw(self.lambdaG[j, l])
elif init_FG == 'kmeans':
print "Initialising F using KMeans."
kmeans_F = KMeans(self.R, self.M, self.K)
kmeans_F.initialise()
kmeans_F.cluster()
self.F = kmeans_F.clustering_results + 0.2
print "Initialising G using KMeans."
kmeans_G = KMeans(self.R.T, self.M.T, self.L)
kmeans_G.initialise()
kmeans_G.cluster()
self.G = kmeans_G.clustering_results + 0.2
self.tau = gamma_mode(self.alpha_s(), self.beta_s())
def cluster(R,M,K):
kmeans = KMeans(R,M,K)
kmeans.initialise()
kmeans.cluster()
return kmeans.clustering_results
def test_init():
# Test getting an exception when X and M are different sizes, X is not a 2D array, and K <= 0
X1 = numpy.ones(3)
M = numpy.ones((2, 3))
K = 0
with pytest.raises(AssertionError) as error:
KMeans(X1, M, K)
assert str(
error.value
) == "Input matrix X is not a two-dimensional array, but instead 1-dimensional."
X2 = numpy.ones((4, 3, 2))
with pytest.raises(AssertionError) as error:
KMeans(X2, M, K)
assert str(
error.value
) == "Input matrix X is not a two-dimensional array, but instead 3-dimensional."
X3 = numpy.ones((3, 2))
with pytest.raises(AssertionError) as error:
KMeans(X3, M, K)
assert str(
error.value
) == "Input matrix X is not of the same size as the indicator matrix M: (3, 2) and (2, 3) respectively."
X4 = numpy.ones((2, 3))
K1 = 0
with pytest.raises(AssertionError) as error:
KMeans(X4, M, K1)
assert str(error.value) == "K should be greater than 0."
# Test getting an exception if a row or column is entirely unknown
X = numpy.ones((2, 3))
M1 = [[1, 1, 1], [0, 0, 0]]
M2 = [[1, 1, 0], [1, 0, 0]]
K = 1
with pytest.raises(AssertionError) as error:
KMeans(X, M1, K)
assert str(error.value) == "Fully unobserved row in X, row 1."
with pytest.raises(AssertionError) as error:
KMeans(X, M2, K)
assert str(error.value) == "Fully unobserved column in X, column 2."
# Test completely observed case
X = numpy.ones((2, 3))
M = numpy.ones((2, 3))
omega_rows = [[0, 1, 2], [0, 1, 2]]
omega_columns = [[0, 1], [0, 1], [0, 1]]
kmeans = KMeans(X, M, K)
assert numpy.array_equal(omega_rows, kmeans.omega_rows)
assert numpy.array_equal(omega_columns, kmeans.omega_columns)
assert kmeans.no_points == 2
assert kmeans.no_coordinates == 3
# Test partially observed case
M = [[1, 0, 1], [0, 1, 1]]
omega_rows = [[0, 2], [1, 2]]
omega_columns = [[0], [1], [0, 1]]
kmeans = KMeans(X, M, K)
assert numpy.array_equal(omega_rows, kmeans.omega_rows)
assert numpy.array_equal(omega_columns, kmeans.omega_columns)
def test_cluster():
### No missing values case.
# Points 1,2 will first go to cluster 2, and point 3 to cluster 1.
# Then point 1 will switch to cluster 1.
X = [[2, 5], [7, 5], [2, 3]]
M = numpy.ones((3, 2))
K = 2
kmeans = KMeans(X, M, K)
kmeans.centroids = [[2.0, 2.0], [4.0, 5.0]]
kmeans.mask_centroids = numpy.ones((2, 2))
kmeans.cluster_assignments = [-1, -1, -1]
expected_centroids = [[2.0, 4.0], [7.0, 5.0]]
expected_cluster_assignments = [0, 1, 0]
expected_data_point_assignments = [[0, 2], [1]]
expected_clustering_results = [[1, 0], [0, 1], [1, 0]]
kmeans.cluster()
assert numpy.array_equal(expected_centroids, kmeans.centroids)
assert numpy.array_equal(expected_cluster_assignments,
kmeans.cluster_assignments)
assert numpy.array_equal(expected_data_point_assignments,
kmeans.data_point_assignments)
assert numpy.array_equal(expected_clustering_results,
kmeans.clustering_results)
### Missing values case.
# Points 2,3,4 will first go to cluster 2, and point 1 to cluster 1.
# Then point 2 will switch to cluster 1.
X = [[2, 5], [3, -1], [10, 1], [-1, 2]]
M = [[1, 1], [1, 0], [1, 1], [0, 1]]
K = 2
kmeans = KMeans(X, M, K)
kmeans.centroids = [[2.0, 7.0], [3.0, 2.0]]
kmeans.mask_centroids = numpy.ones((2, 2))
kmeans.cluster_assignments = [-1, -1, -1, -1]
expected_centroids = [[2.5, 5.0], [10.0, 1.5]]
expected_cluster_assignments = [0, 0, 1, 1]
expected_data_point_assignments = [[0, 1], [2, 3]]
expected_clustering_results = [[1, 0], [1, 0], [0, 1], [0, 1]]
kmeans.cluster()
assert numpy.array_equal(expected_centroids, kmeans.centroids)
assert numpy.array_equal(expected_cluster_assignments,
kmeans.cluster_assignments)
assert numpy.array_equal(expected_data_point_assignments,
kmeans.data_point_assignments)
assert numpy.array_equal(expected_clustering_results,
kmeans.clustering_results)
### Cluster with 0 coordinate.
# Cluster 1 gets points 1 and 2, cluster 2 gets 3 and 4.
X = [[2, 5], [3, -1], [-1, 1], [-1, 2]]
M = [[1, 1], [1, 0], [0, 1], [0, 1]]
K = 2
kmeans = KMeans(X, M, K)
kmeans.centroids = [[2.0, 7.0], [4.0, 4.0]]
kmeans.mask_centroids = numpy.ones((2, 2))
kmeans.cluster_assignments = [-1, -1, -1, -1]
expected_centroids = [[2.5, 5.0], [0, 1.5]]
expected_cluster_assignments = [0, 0, 1, 1]
expected_data_point_assignments = [[0, 1], [2, 3]]
expected_clustering_results = [[1, 0], [1, 0], [0, 1], [0, 1]]
kmeans.cluster()
assert numpy.array_equal(expected_centroids, kmeans.centroids)
assert numpy.array_equal(expected_cluster_assignments,
kmeans.cluster_assignments)
assert numpy.array_equal(expected_data_point_assignments,
kmeans.data_point_assignments)
assert numpy.array_equal(expected_clustering_results,
kmeans.clustering_results)
def test_update():
# Normal case
X = numpy.array([[1.0, 2.0, 3.0], [4.0, 5.0, 6.0], [7.0, 8.0, 9.0]])
M = numpy.array([[1, 1, 0], [0, 1, 0], [1, 1, 1]])
K = 2
kmeans = KMeans(X, M, K)
kmeans.data_point_assignments = numpy.array(
[[0, 1], [2]]) #points 0,1 to cluster 0, point 2 to cluster 1
kmeans.centroids = [[0.0, 0.0, 0.0], [0.0, 0.0, 0.0]]
kmeans.mask_centroids = [[0.0, 0.0, 0.0], [0.0, 0.0, 0.0]]
new_centroids = [[1.0, 3.5, 0], [7.0, 8.0, 9.0]]
new_mask_centroids = [[1, 1, 0], [1, 1, 1]]
kmeans.update()
assert numpy.array_equal(new_centroids, kmeans.centroids)
assert numpy.array_equal(new_mask_centroids, kmeans.mask_centroids)
# Case when one cluster has no points assigned to it - we then randomly re-initialise that cluster
X = numpy.array([[1.0, 2.0, 3.0], [4.0, 5.0, 6.0], [7.0, 8.0, 9.0]])
M = numpy.array([[1, 1, 0], [0, 1, 0], [1, 1, 1]])
K = 2
kmeans = KMeans(X, M, K, 'random')
kmeans.data_point_assignments = numpy.array(
[[0, 1, 2], []]) #points 0,1,2 to cluster 0, none to cluster 1
kmeans.centroids = [[0.0, 0.0, 0.0], [0.0, 0.0, 0.0]]
kmeans.mask_centroids = [[0.0, 0.0, 0.0], [0.0, 0.0, 0.0]]
kmeans.mins = [1.0, 2.0, 9.0]
kmeans.maxs = [7.0, 8.0, 9.0]
new_centroids = [[4.0, 5.0, 9.0],
[6.066531109150288, 6.547726417641815, 9.0]]
new_mask_centroids = [[1, 1, 1], [1, 1, 1]]
random.seed(0)
kmeans.update()
assert numpy.array_equal(new_centroids, kmeans.centroids)
assert numpy.array_equal(new_mask_centroids, kmeans.mask_centroids)
# Case when we use the 'singleton' option for empty clusters - reassign point furthest away to the cluster
# Points 0 and 1 go to cluster 0, 2 to cluster 1 and none to cluster 2.
# Point 2 is furthest away, so gets reassigned to cluster 2 - making
# cluster 1 empty. Then point 1 is furthest away and gets reassigned to cluster 1
X = numpy.array([[1.0, 2.0, 3.0], [4.0, 5.0, 6.0], [7.0, 8.0, 9.0]])
M = numpy.array([[1, 1, 0], [0, 1, 0], [1, 1, 1]])
K = 3
kmeans = KMeans(X, M, K, resolve_empty='singleton')
kmeans.data_point_assignments = numpy.array(
[[0, 1], [2],
[]]) #points 0,1 to cluster 0, 2 to cluster 1, none to cluster 2
kmeans.cluster_assignments = [0, 0, 1]
kmeans.centroids = [[1.0, 2.0, 3.0], [15.0, 16.0, 17.0],
[500.0, 500.0, 500.0]]
kmeans.mask_centroids = [[1, 1, 0], [1, 1, 1], [1, 1, 1]]
kmeans.distances = numpy.array([
kmeans.compute_MSE(kmeans.X[0], kmeans.centroids[0], M[0],
kmeans.mask_centroids[0]),
kmeans.compute_MSE(kmeans.X[1], kmeans.centroids[0], M[1],
kmeans.mask_centroids[0]),
kmeans.compute_MSE(kmeans.X[2], kmeans.centroids[1], M[2],
kmeans.mask_centroids[1])
])
kmeans.mins = [1.0, 2.0, 9.0]
kmeans.maxs = [7.0, 8.0, 9.0]
new_centroids = [[1.0, 2.0, 0], [4.0, 5.0, 6.0], [7.0, 8.0, 9.0]]
new_data_point_assignments = [[0], [1], [2]]
new_distances = [0, 0, 0]
kmeans.update()
assert new_data_point_assignments == list(kmeans.data_point_assignments)
assert numpy.array_equal(new_distances, kmeans.distances)
assert numpy.array_equal(new_centroids, kmeans.centroids)
def test_update():
# Normal case
X = numpy.array([[1.0,2.0,3.0],[4.0,5.0,6.0],[7.0,8.0,9.0]])
M = numpy.array([[1,1,0],[0,1,0],[1,1,1]])
K = 2
kmeans = KMeans(X,M,K)
kmeans.data_point_assignments = numpy.array([[0,1],[2]]) #points 0,1 to cluster 0, point 2 to cluster 1
kmeans.centroids = [[0.0,0.0,0.0],[0.0,0.0,0.0]]
kmeans.mask_centroids = [[0.0,0.0,0.0],[0.0,0.0,0.0]]
new_centroids = [[1.0,3.5,0],[7.0,8.0,9.0]]
new_mask_centroids = [[1,1,0],[1,1,1]]
kmeans.update()
assert numpy.array_equal(new_centroids,kmeans.centroids)
assert numpy.array_equal(new_mask_centroids,kmeans.mask_centroids)
# Case when one cluster has no points assigned to it - we then randomly re-initialise that cluster
X = numpy.array([[1.0,2.0,3.0],[4.0,5.0,6.0],[7.0,8.0,9.0]])
M = numpy.array([[1,1,0],[0,1,0],[1,1,1]])
K = 2
kmeans = KMeans(X,M,K,'random')
kmeans.data_point_assignments = numpy.array([[0,1,2],[]]) #points 0,1,2 to cluster 0, none to cluster 1
kmeans.centroids = [[0.0,0.0,0.0],[0.0,0.0,0.0]]
kmeans.mask_centroids = [[0.0,0.0,0.0],[0.0,0.0,0.0]]
kmeans.mins = [1.0,2.0,9.0]
kmeans.maxs = [7.0,8.0,9.0]
new_centroids = [[4.0,5.0,9.0],[6.066531109150288,6.547726417641815,9.0]]
new_mask_centroids = [[1,1,1],[1,1,1]]
random.seed(0)
kmeans.update()
assert numpy.array_equal(new_centroids,kmeans.centroids)
assert numpy.array_equal(new_mask_centroids,kmeans.mask_centroids)
# Case when we use the 'singleton' option for empty clusters - reassign point furthest away to the cluster
# Points 0 and 1 go to cluster 0, 2 to cluster 1 and none to cluster 2.
# Point 2 is furthest away, so gets reassigned to cluster 2 - making
# cluster 1 empty. Then point 1 is furthest away and gets reassigned to cluster 1
X = numpy.array([[1.0,2.0,3.0],[4.0,5.0,6.0],[7.0,8.0,9.0]])
M = numpy.array([[1,1,0],[0,1,0],[1,1,1]])
K = 3
kmeans = KMeans(X,M,K,resolve_empty='singleton')
kmeans.data_point_assignments = numpy.array([[0,1],[2],[]]) #points 0,1 to cluster 0, 2 to cluster 1, none to cluster 2
kmeans.cluster_assignments = [0,0,1]
kmeans.centroids = [[1.0,2.0,3.0],[15.0,16.0,17.0],[500.0,500.0,500.0]]
kmeans.mask_centroids = [[1,1,0],[1,1,1],[1,1,1]]
kmeans.distances = numpy.array([
kmeans.compute_MSE(kmeans.X[0],kmeans.centroids[0],M[0],kmeans.mask_centroids[0]),
kmeans.compute_MSE(kmeans.X[1],kmeans.centroids[0],M[1],kmeans.mask_centroids[0]),
kmeans.compute_MSE(kmeans.X[2],kmeans.centroids[1],M[2],kmeans.mask_centroids[1])
])
kmeans.mins = [1.0,2.0,9.0]
kmeans.maxs = [7.0,8.0,9.0]
new_centroids = [[1.0,2.0,0],[4.0,5.0,6.0],[7.0,8.0,9.0]]
new_data_point_assignments = [[0],[1],[2]]
new_distances = [0,0,0]
kmeans.update()
assert new_data_point_assignments == list(kmeans.data_point_assignments)
assert numpy.array_equal(new_distances,kmeans.distances)
assert numpy.array_equal(new_centroids,kmeans.centroids)
def test_cluster():
### No missing values case.
# Points 1,2 will first go to cluster 2, and point 3 to cluster 1.
# Then point 1 will switch to cluster 1.
X = [[2,5],[7,5],[2,3]]
M = numpy.ones((3,2))
K = 2
kmeans = KMeans(X,M,K)
kmeans.centroids = [[2.0,2.0],[4.0,5.0]]
kmeans.mask_centroids = numpy.ones((2,2))
kmeans.cluster_assignments = [-1,-1,-1]
expected_centroids = [[2.0,4.0],[7.0,5.0]]
expected_cluster_assignments = [0,1,0]
expected_data_point_assignments = [[0,2],[1]]
expected_clustering_results = [[1,0],[0,1],[1,0]]
kmeans.cluster()
assert numpy.array_equal(expected_centroids,kmeans.centroids)
assert numpy.array_equal(expected_cluster_assignments,kmeans.cluster_assignments)
assert numpy.array_equal(expected_data_point_assignments,kmeans.data_point_assignments)
assert numpy.array_equal(expected_clustering_results,kmeans.clustering_results)
### Missing values case.
# Points 2,3,4 will first go to cluster 2, and point 1 to cluster 1.
# Then point 2 will switch to cluster 1.
X = [[2,5],[3,-1],[10,1],[-1,2]]
M = [[1,1],[1,0],[1,1],[0,1]]
K = 2
kmeans = KMeans(X,M,K)
kmeans.centroids = [[2.0,7.0],[3.0,2.0]]
kmeans.mask_centroids = numpy.ones((2,2))
kmeans.cluster_assignments = [-1,-1,-1,-1]
expected_centroids = [[2.5,5.0],[10.0,1.5]]
expected_cluster_assignments = [0,0,1,1]
expected_data_point_assignments = [[0,1],[2,3]]
expected_clustering_results = [[1,0],[1,0],[0,1],[0,1]]
kmeans.cluster()
assert numpy.array_equal(expected_centroids,kmeans.centroids)
assert numpy.array_equal(expected_cluster_assignments,kmeans.cluster_assignments)
assert numpy.array_equal(expected_data_point_assignments,kmeans.data_point_assignments)
assert numpy.array_equal(expected_clustering_results,kmeans.clustering_results)
### Cluster with 0 coordinate.
# Cluster 1 gets points 1 and 2, cluster 2 gets 3 and 4.
X = [[2,5],[3,-1],[-1,1],[-1,2]]
M = [[1,1],[1,0],[0,1],[0,1]]
K = 2
kmeans = KMeans(X,M,K)
kmeans.centroids = [[2.0,7.0],[4.0,4.0]]
kmeans.mask_centroids = numpy.ones((2,2))
kmeans.cluster_assignments = [-1,-1,-1,-1]
expected_centroids = [[2.5,5.0],[0,1.5]]
expected_cluster_assignments = [0,0,1,1]
expected_data_point_assignments = [[0,1],[2,3]]
expected_clustering_results = [[1,0],[1,0],[0,1],[0,1]]
kmeans.cluster()
assert numpy.array_equal(expected_centroids,kmeans.centroids)
assert numpy.array_equal(expected_cluster_assignments,kmeans.cluster_assignments)
assert numpy.array_equal(expected_data_point_assignments,kmeans.data_point_assignments)
assert numpy.array_equal(expected_clustering_results,kmeans.clustering_results)
def initialise(self, init_S='random', init_FG='random', expo_prior=1.):
assert init_S in ['ones', 'random', 'exponential'
], "Unrecognised init option for S: %s." % init_S
assert init_FG in ['ones', 'random', 'exponential', 'kmeans'
], "Unrecognised init option for F,G: %s." % init_FG
if init_S == 'ones':
self.S = numpy.ones((self.K, self.L))
elif init_S == 'random':
self.S = numpy.random.rand(self.K, self.L)
elif init_S == 'exponential':
self.S = numpy.empty((self.K, self.L))
for k, l in itertools.product(xrange(0, self.K), xrange(0,
self.L)):
self.S[k, l] = exponential_draw(expo_prior)
if init_FG == 'ones':
self.F = numpy.ones((self.I, self.K))
self.G = numpy.ones((self.J, self.L))
elif init_FG == 'random':
self.F = numpy.random.rand(self.I, self.K)
self.G = numpy.random.rand(self.J, self.L)
elif init_FG == 'exponential':
self.F = numpy.empty((self.I, self.K))
self.G = numpy.empty((self.J, self.L))
for i, k in itertools.product(xrange(0, self.I), xrange(0,
self.K)):
self.F[i, k] = exponential_draw(expo_prior)
for j, l in itertools.product(xrange(0, self.J), xrange(0,
self.L)):
self.G[j, l] = exponential_draw(expo_prior)
elif init_FG == 'kmeans':
print "Initialising F using KMeans."
kmeans_F = KMeans(self.R, self.M, self.K)
kmeans_F.initialise()
kmeans_F.cluster()
self.F = kmeans_F.clustering_results + 0.2
print "Initialising G using KMeans."
kmeans_G = KMeans(self.R.T, self.M.T, self.L)
kmeans_G.initialise()
kmeans_G.cluster()
self.G = kmeans_G.clustering_results + 0.2
