def fitnessFunction(individual, algorithm, n_labels, ground_truth, m1, m2, m3):
w1 = individual[0]
w2 = individual[1]
# w3 = individual[2]
w3 = 0
corr = mf.calculateCorrelationMatrix(m1, m2, m3, w1, w2, w3)
if algorithm == 'complete':
agglomerative = AgglomerativeClustering(affinity='precomputed', n_clusters=n_labels, linkage='complete').fit(corr)
labels = agglomerative.labels_
elif algorithm == 'average':
agglomerative = AgglomerativeClustering(affinity='precomputed', n_clusters=n_labels, linkage='average').fit(corr)
labels = agglomerative.labels_
elif algorithm == 'kmedoids':
_, clusters = km.kMedoids(corr, n_labels, 100)
labels = km.sortLabels(clusters)
#fitness = metrics.homogeneity_score(labels, ground_truth)
#fitness = metrics.adjusted_rand_score(labels, ground_truth)
#fitness = sum(individual)
fitness = metrics.calinski_harabaz_score(corr, labels)
corr = None
m1 = None
m2 = None
m3 = None
return fitness
def fitness(indv):
w1, w2, w3 = indv.solution
corr = mf.calculateCorrelationMatrix(matrix1, matrix2, matrix3, w1, w2, w3)
if algorithm == 'complete':
agglomerative = AgglomerativeClustering(affinity='precomputed', n_clusters=n_labels, linkage='complete').fit(corr)
labels = agglomerative.labels_
elif algorithm == 'average':
agglomerative = AgglomerativeClustering(affinity='precomputed', n_clusters=n_labels, linkage='average').fit(corr)
labels = agglomerative.labels_
elif algorithm == 'kmedoids':
_, clusters = km.kMedoids(corr, n_labels, 100)
labels = km.sortLabels(clusters)
metrics = ce.clusterEvaluation(corr, labels, ground_truth)
writer.write(str(current_iteration)+': '+str(w1)+' '+str(w2)+' '+str(w3)+' '.join(str(x) for x in metrics)+'\n')
def fitness(indv):
w1, w2, w3 = indv.solution
corr = mf.calculateCorrelationMatrix(matrix1, matrix2, matrix3, w1, w2, w3)
if algorithm == 'complete':
agglomerative = AgglomerativeClustering(affinity='precomputed',
n_clusters=n_labels,
linkage='complete').fit(corr)
labels = agglomerative.labels_
elif algorithm == 'average':
agglomerative = AgglomerativeClustering(affinity='precomputed',
n_clusters=n_labels,
linkage='average').fit(corr)
labels = agglomerative.labels_
elif algorithm == 'kmedoids':
_, clusters = km.kMedoids(corr, n_labels, 100)
labels = km.sortLabels(clusters)
metrics = ce.clusterEvaluation(corr, labels, ground_truth)
return float(metrics[0]) * 100
matrix2 = mf.calculateDistances(matrix2)
matrix3 = mf.calculateDistances(matrix3)
for w1 in np.arange(0.05, 1.05, 0.05):
w2 = 0
w3 = 1 - w1
corr = mf.calculateCorrelationMatrix(matrix1, matrix2, matrix3, w1,
w2, w3)
# Hierarchical
for link in ['complete', 'average']:
agglomerative = AgglomerativeClustering(
affinity='precomputed',
n_clusters=n_labels,
linkage='complete').fit(corr)
labels = agglomerative.labels_
metrics = ce.clusterEvaluation(corr, labels, ground_truth)
print(metrics)
ce.saveResultsWithWeights(measure1, measure2, w1,
'hierarchical_' + link, sample,
metrics)
# K-Medoids
medoids, clusters = km.kMedoids(corr, n_labels, 100)
labels = km.sortLabels(clusters)
metrics = ce.clusterEvaluation(corr, labels, ground_truth)
ce.saveResultsWithWeights(measure1, measure2, w1, 'kmedoids',
sample, metrics)
##################################
# USAGE ##########################
##################################
# readBin('/path/to/testcase', removeEquivalents=1)
# k_means_plusplus(#ofClusters, distanceMeasure)
# k_medoids(#ofClusters, distanceMeasure)
# distanceMeasure:
# 1 - Hamming
# 2 - Jaccard
# 3 - Euclidian (rounded)
kMean = KMeansPP.KMeansPP()
kMean.readBin('../test_result/MSTP_bin_result.txt')
kMean.k_means_plusplus(10, 2)
print(kMean.clusters)
reduced = greedy_pick(kMean.centroids)
mut_score(reduced, kMean.dataset)
kMed = KMedoids.KMedoids()
kMed.readBin('../test_result/MSTP_bin_result.txt')
kMed.k_medoids(9, 2)
print(kMed.clusters)
reduced = greedy_pick(kMed.medoids)
mut_score(reduced, kMed.dataset)
mean = 1.79197547637771
std_dev = 0.669382812243833
#X = (X - (mean/std_dev))
X = mf.minMaxScale(X)
X = mf.calculateDistances(X)
domains = rs.loadDomainListFromFile(sample_for_domains)
# read existing labels
n_labels = scop.getUniqueClassifications(sample_for_domains)
ground_truth = scop.getDomainLabels(domains)
ground_truth = map(int, ground_truth)
ground_truth = list(map(int, ground_truth))
X = np.asmatrix(X)
M, C = kmedoids.kMedoids(X, n_labels, 100)
print('medoids:')
for point_idx in M:
print(X[point_idx])
print('')
print('clustering result:')
for label in C:
for point_idx in C[label]:
print('label {0}: {1}'.format(label, X[point_idx]))
#X = (X - (mean/std_dev))
X = mf.minMaxScale(X)
X = mf.calculateDistances(X)
domains = rs.loadDomainListFromFile(sample_for_domains)
# read existing labels
n_labels = scop.getUniqueClassifications(sample_for_domains)
ground_truth = scop.getDomainLabels(domains)
ground_truth = map(int, ground_truth)
ground_truth = list(map(int, ground_truth))
X = np.asmatrix(X)
medoids, clusters = kmedoids.kMedoids(X, n_labels, 100)
print('clustering result:')
for label in clusters:
for point_idx in clusters[label]:
print('label {0}, {1}: {2}'.format(label, ground_truth[point_idx],
X[point_idx]))
#cl.clusterEvaluationNoLabels(X, clusters)
index_list = []
new_dict = {}
for label in clusters:
for point_idx in clusters[label]:
new_dict[point_idx] = label