def kernelkmeans(fileData, epocs ,normalized_axis = 0 ,norm='l1', robust_gamma = 0.00015):
#parameters
epocs = sio.loadmat(epocs)['epocs'][0][0]
#gamma_logscale = [1,2,3,4,5,6,7,8,9,10,-1,-2,-3,-4,-5,-6,-7,-8,-9,-10,-11,-12,-13,-14,-15,-16,-17,-18]
gamma_logscale = [-3,-4]
print np.size(gamma_logscale)
print epocs
print epocs*np.size(gamma_logscale)
clustering_accuracy = np.zeros(epocs*np.size(gamma_logscale))
calculate_purity = np.zeros(epocs*np.size(gamma_logscale))
calculate_nmi = np.zeros(epocs*np.size(gamma_logscale))
cont = 0
for gamma in gamma_logscale:
for epocs in xrange(epocs):
print epocs
labels_true,labels_pred,features= KernelKMeans.robust_kernel(fileData, normalized_axis = normalized_axis,norm=norm,gamma=2**gamma, robust_gamma = robust_gamma)
clustering_accuracy[cont] = Metrics.calculate_clusteringAccuracy(labels_true,labels_pred)
calculate_purity[cont],vector = Metrics.calculate_purity(labels_true,labels_pred)
calculate_nmi[cont] = Metrics.calculate_nmi(labels_true,labels_pred)
cont +=1
sio.savemat('resultsRobust',{'clusteringAccuracy' : clustering_accuracy,'purityvec' : calculate_purity,'nmivec' : calculate_nmi})
def kernelkmeans(fileData, epocs ,normalized_axis = 0 ,norm='l1'):
#parameters
epocs = sio.loadmat(epocs)['epocs'][0][0]
gamma_logscale = [1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,0,-1,-2,-3,-4,-5,-6,-7,-8,-9,-10,-11,-12,-13,-14,-15,-16,-17,-18]
print np.size(gamma_logscale)
print epocs
print epocs*np.size(gamma_logscale)
labels_total = []
clustering_accuracy = np.zeros(epocs*np.size(gamma_logscale))
calculate_purity = np.zeros(epocs*np.size(gamma_logscale))
calculate_nmi = np.zeros(epocs*np.size(gamma_logscale))
clustering_accuracy_mean = np.zeros(np.size(gamma_logscale))
calculate_purity_mean = np.zeros(np.size(gamma_logscale))
calculate_nmi_mean = np.zeros(np.size(gamma_logscale))
clustering_accuracy_sd = np.zeros(np.size(gamma_logscale))
calculate_purity_sd = np.zeros(np.size(gamma_logscale))
calculate_nmi_sd = np.zeros(np.size(gamma_logscale))
cont = 0
contador = 0
for j,gamma in zip(range(34),gamma_logscale):
aux=cont
for epocs in xrange(30):
print epocs
labels_true,labels_pred,features= KernelKMeans.get_kernelKMeans(fileData, normalized_axis = normalized_axis,norm=norm,gamma=2**gamma)
labels_total.append(labels_pred)
if np.size(np.unique(labels_pred)) == np.size(np.unique(labels_true)):
clustering_accuracy[cont] = Metrics.calculate_clusteringAccuracy(labels_true,labels_pred)
calculate_purity[cont],vector = Metrics.calculate_purity(labels_true,labels_pred)
calculate_nmi[cont] = Metrics.calculate_nmi(labels_true,labels_pred)
cont+=1
contador+=1
clustering_accuracy_mean[j] = np.mean(clustering_accuracy[aux:cont-1])
calculate_purity_mean[j] = np.mean(calculate_purity[aux:cont-1])
calculate_nmi_mean[j] = np.mean(calculate_nmi[aux:cont-1])
clustering_accuracy_sd[j] = np.std(clustering_accuracy[aux:cont-1])
calculate_purity_sd[j] = np.std(calculate_purity[aux:cont-1])
calculate_nmi_sd[j] = np.std(calculate_nmi[aux:cont-1])
sio.savemat('G:/Dropbox/Universidad/Machine Learning/Robustes/AR/KernelKMeans/results',{'labels_total' : labels_total ,
'clusteringAccuracyVec' : clustering_accuracy,
'purityVec' : calculate_purity,
'nmiVec' : calculate_nmi ,
'clusteringAccuracyMeanVec' : clustering_accuracy_mean,
'purityMeanVec' : calculate_purity_mean,
'nmiMeanVec' : calculate_nmi_mean,
'clusteringAccuracyStdVec' : clustering_accuracy_sd ,
'purityStdVec' :calculate_purity_sd ,
'nmiStdVec':calculate_nmi_sd})
def RMNMF(path,labels_name):
labels_pred_total = sio.loadmat(path+'RMNMF/results_aux')['labels_total']
labels_true = sio.loadmat(path+labels_name)['labels']
print labels_pred_total.shape
print labels_true.shape
num = 4
epocs = 30
clustering_accuracy = np.zeros(num*epocs)
calculate_purity = np.zeros(num*epocs)
calculate_nmi = np.zeros(num*epocs)
clustering_accuracy_mean = np.zeros(num)
calculate_purity_mean = np.zeros(num)
calculate_nmi_mean = np.zeros(num)
clustering_accuracy_sd = np.zeros(num)
calculate_purity_sd = np.zeros(num)
calculate_nmi_sd = np.zeros(num)
cont = 0
contador = 0
for j in range(num):
aux = cont
for i in range(30):
labels_pred = labels_pred_total[contador,:]
if np.size(np.unique(labels_pred)) == np.size(np.unique(labels_true)):
clustering_accuracy[cont] = Metrics.calculate_clusteringAccuracy(labels_true,labels_pred)
calculate_purity[cont],vector = Metrics.calculate_purity(labels_true,labels_pred)
calculate_nmi[cont] = Metrics.calculate_nmi(labels_true,labels_pred)
cont+=1
contador+=1
clustering_accuracy_mean[j] = np.mean(clustering_accuracy[aux:cont-1])
calculate_purity_mean[j] = np.mean(calculate_purity[aux:cont-1])
calculate_nmi_mean[j] = np.mean(calculate_nmi[aux:cont-1])
clustering_accuracy_sd[j] = np.std(clustering_accuracy[aux:cont-1])
calculate_purity_sd[j] = np.std(calculate_purity[aux:cont-1])
calculate_nmi_sd[j] = np.std(calculate_nmi[aux:cont-1])
sio.savemat(path+'RMNMF/results',{'labels_total' : labels_pred_total ,
'clusteringAccuracyVec' : clustering_accuracy,
'purityVec' : calculate_purity,
'nmiVec' : calculate_nmi ,
'clusteringAccuracyMeanVec' : clustering_accuracy_mean,
'purityMeanVec' : calculate_purity_mean,
'nmiMeanVec' : calculate_nmi_mean,
'clusteringAccuracyStdVec' : clustering_accuracy_sd ,
'purityStdVec' :calculate_purity_sd ,
'nmiStdVec':calculate_nmi_sd})
def Kernelseminmfnnls(path,labels_name):
labels_pred_total = sio.loadmat(path+'Kernelseminmfnnls/results_aux')['labels_total']
labels_true = sio.loadmat(path+labels_name)['labels']
print labels_pred_total.shape
print labels_true.shape
clustering_accuracy = np.zeros(21*30)
calculate_purity = np.zeros(21*30)
calculate_nmi = np.zeros(21*30)
clustering_accuracy_mean = np.zeros(21)
calculate_purity_mean = np.zeros(21)
calculate_nmi_mean = np.zeros(21)
clustering_accuracy_sd = np.zeros(21)
calculate_purity_sd = np.zeros(21)
calculate_nmi_sd = np.zeros(21)
cont = 0
contador = 0
for j in range(21):
aux = cont
for i in range(30):
labels_pred = labels_pred_total[contador,:]
#if np.size(np.unique(labels_pred)) == np.size(np.unique(labels_true)):
clustering_accuracy[(j)*21+i] = Metrics.calculate_clusteringAccuracy(labels_true,labels_pred)
calculate_purity[(j)*21+i],vector = Metrics.calculate_purity(labels_true,labels_pred)
calculate_nmi[(j)*21+i] = Metrics.calculate_nmi(labels_true,labels_pred)
cont+=1
contador+=1
clustering_accuracy_mean[j] = np.mean(clustering_accuracy[(j)*21:(j)*21+30])
calculate_purity_mean[j] = np.mean(calculate_purity[(j)*21:(j)*21+30])
calculate_nmi_mean[j] = np.mean(calculate_nmi[(j)*21:(j)*21+30])
clustering_accuracy_sd[j] = np.std(clustering_accuracy[(j)*21:(j)*21+30])
calculate_purity_sd[j] = np.std(calculate_purity[(j)*21:(j)*21+30])
calculate_nmi_sd[j] = np.std(calculate_nmi[(j)*21:(j)*21+30])
sio.savemat(path+'Kernelseminmfnnls/results',{'labels_total' : labels_pred_total ,
'clusteringAccuracyVec' : clustering_accuracy,
'purityVec' : calculate_purity,
'nmiVec' : calculate_nmi ,
'clusteringAccuracyMeanVec' : clustering_accuracy_mean,
'purityMeanVec' : calculate_purity_mean,
'nmiMeanVec' : calculate_nmi_mean,
'clusteringAccuracyStdVec' : clustering_accuracy_sd ,
'purityStdVec' :calculate_purity_sd ,
'nmiStdVec':calculate_nmi_sd})
def kmeans(fileData, epocs ,normalized_axis = 0 ,norm='l1'):
#parameters
epocs = sio.loadmat(epocs)['epocs'][0][0]
clustering_accuracy = np.zeros(epocs)
calculate_purity = np.zeros(epocs)
calculate_nmi = np.zeros(epocs)
for epocs in xrange(epocs):
print epocs
labels_true,labels_pred,features= KMeans.get_kMeans(fileData,normalized_axis = normalized_axis ,norm=norm)
clustering_accuracy[epocs] = Metrics.calculate_clusteringAccuracy(labels_true,labels_pred)
calculate_purity[epocs],vector = Metrics.calculate_purity(labels_true,labels_pred)
calculate_nmi[epocs] = Metrics.calculate_nmi(labels_true,labels_pred)
sio.savemat('results',{'clusteringAccuracy' : clustering_accuracy,'purityvec' : calculate_purity,'nmivec' : calculate_nmi})
def kernelkmeans(fileData,
epocs,
normalized_axis=0,
norm='l1',
robust_gamma=0.00015):
#parameters
epocs = sio.loadmat(epocs)['epocs'][0][0]
#gamma_logscale = [1,2,3,4,5,6,7,8,9,10,-1,-2,-3,-4,-5,-6,-7,-8,-9,-10,-11,-12,-13,-14,-15,-16,-17,-18]
gamma_logscale = [-3, -4]
print np.size(gamma_logscale)
print epocs
print epocs * np.size(gamma_logscale)
clustering_accuracy = np.zeros(epocs * np.size(gamma_logscale))
calculate_purity = np.zeros(epocs * np.size(gamma_logscale))
calculate_nmi = np.zeros(epocs * np.size(gamma_logscale))
cont = 0
for gamma in gamma_logscale:
for epocs in xrange(epocs):
print epocs
labels_true, labels_pred, features = KernelKMeans.robust_kernel(
fileData,
normalized_axis=normalized_axis,
norm=norm,
gamma=2**gamma,
robust_gamma=robust_gamma)
clustering_accuracy[cont] = Metrics.calculate_clusteringAccuracy(
labels_true, labels_pred)
calculate_purity[cont], vector = Metrics.calculate_purity(
labels_true, labels_pred)
calculate_nmi[cont] = Metrics.calculate_nmi(
labels_true, labels_pred)
cont += 1
sio.savemat(
'resultsRobust', {
'clusteringAccuracy': clustering_accuracy,
'purityvec': calculate_purity,
'nmivec': calculate_nmi
})
epocs = sio.loadmat('G:/Dropbox/Universidad/Machine Learning/Robustes/ATT/parameters.mat')['epocs']
data = np.load(fileData)
X = data['data']
labelsReal = data['labels']
sigma = 2**3
nClusters = 10
X = normalizeByRange.normalizeByRange(X,0)
X= np.transpose(X)
K = gaussianKernel.gaussianKernel(X,sigma=sigma)
H,labelsPred,t = kernel_k_means.KKmeans(K,nClusters,1000)
print labelsPred
print accuracy.accuracy(labelsReal,labelsPred)
print metrics.calculate_clusteringAccuracy(labelsReal,labelsPred)
model = KernelKMeans.KernelKMeans(n_clusters=nClusters,
kernel="rbf", gamma=1./sigma,
verbose=1)
print model
model.fit(np.transpose(X))
labelsPred = model.labels_
print accuracy.accuracy(labelsReal,labelsPred)
print metrics.calculate_clusteringAccuracy(labelsReal,labelsPred)
import scipy.io as sio
#parameters
#fileData = 'G:/Dropbox/Universidad/Machine Learning/Robustes/Abalone/abalone.npz'
#epocs = sio.loadmat('G:/Dropbox/Universidad/Machine Learning/Robustes/Abalone/parameters.mat')['epocs']
fileData = 'G:/Dropbox/Universidad/Machine Learning/Robustes/ATT/att.npz'
epocs = sio.loadmat('G:/Dropbox/Universidad/Machine Learning/Robustes/ATT/parameters.mat')['epocs']
n_clusters = 3
clustering_accuracy = np.zeros(epocs)
calculate_purity = np.zeros(epocs)
calculate_nmi = np.zeros(epocs)
labels_total = np.zeros((epocs,400))
for epocs in xrange(epocs):
labels_true,labels_pred,features= KMeans.get_kMeans(fileData,normalized_axis = 0 ,norm='l1')
labels_total[epocs,:] = labels_pred
clustering_accuracy[epocs] = Metrics.calculate_clusteringAccuracy(labels_true,labels_pred)
calculate_purity[epocs],vector = Metrics.calculate_purity(labels_true,labels_pred)
calculate_nmi[epocs] = Metrics.calculate_nmi(labels_true,labels_pred)
clustering_accuracy_mean = np.mean(clustering_accuracy)
calculate_purity_mean = np.mean(calculate_purity)
calculate_nmi_mean = np.mean(calculate_nmi)
sio.savemat('results',{'labels_total' : labels_total ,'clusteringAccuracyVec' : clustering_accuracy,'purityVec' : calculate_purity,'nmiVec' : calculate_nmi , 'clusteringAccuracyMeanVec' : clustering_accuracy_mean, 'purityMeanVec' : calculate_purity_mean, 'nmiMeanVec' : calculate_nmi_mean})
)['epocs']
n_clusters = 3
clustering_accuracy = np.zeros(epocs)
calculate_purity = np.zeros(epocs)
calculate_nmi = np.zeros(epocs)
labels_total = np.zeros((epocs, 400))
for epocs in xrange(epocs):
labels_true, labels_pred, features = KMeans.get_kMeans(fileData,
normalized_axis=0,
norm='l1')
labels_total[epocs, :] = labels_pred
clustering_accuracy[epocs] = Metrics.calculate_clusteringAccuracy(
labels_true, labels_pred)
calculate_purity[epocs], vector = Metrics.calculate_purity(
labels_true, labels_pred)
calculate_nmi[epocs] = Metrics.calculate_nmi(labels_true, labels_pred)
clustering_accuracy_mean = np.mean(clustering_accuracy)
calculate_purity_mean = np.mean(calculate_purity)
calculate_nmi_mean = np.mean(calculate_nmi)
sio.savemat(
'results', {
'labels_total': labels_total,
'clusteringAccuracyVec': clustering_accuracy,
'purityVec': calculate_purity,
'nmiVec': calculate_nmi,
'clusteringAccuracyMeanVec': clustering_accuracy_mean,