def gaussian_kernel_kmeans():
print os.path.dirname(os.path.realpath(__file__))
os.chdir(os.path.dirname(os.path.realpath(__file__)))
data, labels = dataset_factory.dataset_factory(
'../../../../dataset/jaffe.mat',
options={
'data': 'data',
'labels': 'labels'
})
print data
k = np.unique(labels).size
data = normalize_by_range(data, axis=0)
vect_to_prove = [2**x for x in np.arange(-20, 25)]
vector = generate_logarithm_vector_kernel(data,
vect_to_prove,
percentage=0.5)
dt = [('key', 'S100'), ('value', 'S100')]
arr = np.zeros((10, ), dtype=dt)
arr[0]['value'] = str(k)
arr[0]['key'] = 'k'
arr[1]['value'] = "iter: " + str(3000)
arr[1]['key'] = 'termination_criterion'
arr[2]['value'] = 'kmeans'
arr[2]['key'] = 'initialization'
arr[3]['value'] = '15'
arr[3]['key'] = 'epocs'
arr[4]['value'] = '1'
arr[4]['key'] = 'clustering_accuracy'
arr[5]['value'] = '1'
arr[5]['key'] = 'purity'
arr[6]['value'] = 'rbf'
arr[6]['key'] = 'kernel'
arr[7]['value'] = '1'
arr[7]['key'] = 'param'
arr[8]['value'] = str(vector)
arr[8]['key'] = 'vect'
def options(vect):
option = {
arr[0]['key']: int(arr[0]['value']),
arr[1]['key']: {
'iter': 3000
},
arr[2]['key']: arr[2]['value'],
arr[3]['key']: int(arr[3]['value']),
arr[4]['key']: int(arr[4]['value']),
arr[5]['key']: int(arr[5]['value']),
arr[6]['key']: arr[6]['value'],
arr[7]['key']: int(arr[7]['value']),
arr[8]['key']: vector
}
for i in vect:
#define sigma
option['param'] = i
yield option
option = {
arr[0]['key']: int(arr[0]['value']),
arr[1]['key']: {
'iter': 3000
},
arr[2]['key']: arr[2]['value'],
arr[3]['key']: int(arr[3]['value']),
arr[4]['key']: int(arr[4]['value']),
arr[5]['key']: int(arr[5]['value']),
arr[6]['key']: arr[6]['value'],
arr[7]['key']: int(arr[7]['value']),
arr[8]['key']: vector
}
kernel_kmeans_experiment = KernelKMeansExperiment(data, option)
best_results_performance, results_performance = tunning_parameter_unsupervised_technique\
(data, labels, kernel_kmeans_experiment, options(vector))
import time
## dd/mm/yyyy format
date = (time.strftime("%d%m%Y_%H-%M_"))
np.save(
str(date) + 'linear_cnmf', {
'results_performance': results_performance,
'best_results_performance': best_results_performance,
'options': arr
})
def gaussian_cnmf():
print os.path.dirname(os.path.realpath(__file__))
os.chdir(os.path.dirname(os.path.realpath(__file__)))
data, labels = dataset_factory.dataset_factory('../../../../dataset/balance_scale.mat',
options={'data': 'data', 'labels': 'labels'})
print data
k = np.unique(labels).size
data = normalize_by_range(data, axis=0)
vect_to_prove = [2**x for x in np.arange(-20, 25)]
vector = generate_logarithm_vector_kernel(data, vect_to_prove, percentage=0.5)
dt = [('key', 'S100'), ('value', 'S100')]
arr = np.zeros((10,), dtype=dt)
arr[0]['value'] = str(k)
arr[0]['key'] = 'k'
arr[1]['value'] = "iter: " + str(3000)
arr[1]['key'] = 'termination_criterion'
arr[2]['value'] = 'random'
arr[2]['key'] = 'initialization'
arr[3]['value'] = '15'
arr[3]['key'] = 'epocs'
arr[4]['value'] = '1'
arr[4]['key'] = 'clustering_accuracy'
arr[5]['value'] = '1'
arr[5]['key'] = 'purity'
arr[6]['value'] = 'rbf'
arr[6]['key'] = 'kernel'
arr[7]['value'] = '1'
arr[7]['key'] = 'param'
arr[8]['value'] = str(vector)
arr[8]['key'] = 'vect'
def options(vect):
option = {arr[0]['key']: int(arr[0]['value']),
arr[1]['key']: {'iter': 3000},
arr[2]['key']: arr[2]['value'],
arr[3]['key']: int(arr[3]['value']),
arr[4]['key']: int(arr[4]['value']),
arr[5]['key']: int(arr[5]['value']),
arr[6]['key']: arr[6]['value'],
arr[7]['key']: int(arr[7]['value']),
arr[8]['key']: vector}
for i in vect:
#define sigma
option['param'] = i
yield option
option = {arr[0]['key']: int(arr[0]['value']),
arr[1]['key']: {'iter': 3000},
arr[2]['key']: arr[2]['value'],
arr[3]['key']: int(arr[3]['value']),
arr[4]['key']: int(arr[4]['value']),
arr[5]['key']: int(arr[5]['value']),
arr[6]['key']: arr[6]['value'],
arr[7]['key']: int(arr[7]['value']),
arr[8]['key']: vector}
cnmf_experiment = CnmfExperiment(data, option)
best_results_performance, results_performance = tunning_parameter_unsupervised_technique\
(data, labels, cnmf_experiment, options(vector))
import time
## dd/mm/yyyy format
date = (time.strftime("%d%m%Y_%H-%M_"))
np.save(str(date) + 'linear_cnmf', {'results_performance': results_performance,
'best_results_performance': best_results_performance, 'options': arr})
url_dataset = '../../../../dataset/mnist_occlusion_subsample_2k.mat'
data, labels = dataset_factory.dataset_factory(
url_dataset, options={'data': 'train_data_2000', 'labels': 'labels_train_data_2000'})
contamination_data, labels = dataset_factory.dataset_factory(
url_dataset, options={'data': 'train_data_occlusion_2000', 'labels': 'labels_train_data_2000'})
import numpy as np
print data
k = np.unique(labels).size
#data, contamination_data = normalize_by_range(data, 0, contamination_data)
vect_to_prove = [2**x for x in np.arange(-20, 20)]
vector = generate_logarithm_vector_kernel(data, vect_to_prove, percentage=0.5)
dt = [('key', 'S100'), ('value', 'S100')]
arr = np.zeros((10,), dtype=dt)
arr[0]['value'] = str(k)
arr[0]['key'] = 'k'
arr[1]['value'] = "iter: " + str(3000)
arr[1]['key'] = 'termination_criterion'
arr[2]['value'] = 'random'
arr[2]['key'] = 'initialization'
arr[3]['value'] = '10'
arr[3]['key'] = 'epocs'
arr[4]['value'] = '1'
arr[4]['key'] = 'clustering_accuracy'
arr[5]['value'] = '1'
})
contamination_data, labels = dataset_factory.dataset_factory(
url_dataset,
options={
'data': 'train_data_occlusion_2000',
'labels': 'labels_train_data_2000'
})
import numpy as np
print data
k = np.unique(labels).size
#data, contamination_data = normalize_by_range(data, 0, contamination_data)
vect_to_prove = [2**x for x in np.arange(-20, 20)]
vector = generate_logarithm_vector_kernel(data, vect_to_prove, percentage=0.5)
dt = [('key', 'S100'), ('value', 'S100')]
arr = np.zeros((10, ), dtype=dt)
arr[0]['value'] = str(k)
arr[0]['key'] = 'k'
arr[1]['value'] = "iter: " + str(3000)
arr[1]['key'] = 'termination_criterion'
arr[2]['value'] = 'random'
arr[2]['key'] = 'initialization'
arr[3]['value'] = '10'
arr[3]['key'] = 'epocs'
arr[4]['value'] = '1'
arr[4]['key'] = 'clustering_accuracy'
arr[5]['value'] = '1'
