def two_moons_hfs_hard(path,
dataset='/data/data_2moons_hfs',
l=4,
var=1,
eps=0,
k=10,
laplacian_normalization=""):
# a skeleton function to perform HFS, needs to be completed
# load the data
in_data = scipy.io.loadmat(path + dataset)
X = in_data['X']
Y = in_data['Y'][:, 0]
# Draw at random index of labels
Y_masked = mask_labels(Y, l)
labels = hard_hfs(X,
Y_masked,
var=var,
eps=eps,
k=k,
laplacian_normalization=laplacian_normalization)
helper.plot_classification(X, Y, labels, var=var, eps=eps, k=k)
accuracy = np.mean(labels == np.squeeze(Y))
return accuracy
def two_moons_hfs_soft(path,
cl,
cu,
l=4,
var=1,
eps=0,
k=10,
laplacian_normalization=""):
# a skeleton function to perform HFS, needs to be completed
# load the data
in_data = scipy.io.loadmat(path + '/data/data_2moons_hfs')
X = in_data['X']
Y = in_data['Y'][:, 0]
# automatically infer number of labels from samples
num_samples = X.shape[0]
num_classes = len(np.unique(Y))
# Draw at random index of labels
Y_masked = mask_labels(Y, l)
labels = soft_hfs(X,
Y_masked,
cl,
cu,
var=var,
eps=eps,
k=k,
laplacian_normalization=laplacian_normalization)
helper.plot_classification(X, Y, labels, var=var, eps=eps, k=k)
accuracy = np.mean(labels == np.squeeze(Y))
return accuracy
def two_moons_hfs():
"""
TO BE COMPLETED.
HFS for two_moons data.
"""
"""
Load the data. At home, try to use the larger dataset (question 1.2).
"""
# load the data
in_data = loadmat(os.path.join('data', 'data_2moons_hfs.mat'))
# in_data = loadmat(os.path.join('data', 'data_2moons_hfs_large.mat'))
X = in_data['X']
Y = in_data['Y'].squeeze()
# automatically infer number of labels from samples
num_samples = np.size(Y, 0)
num_classes = len(np.unique(Y))
"""
Choose the experiment parameters
"""
var = 1
eps = 0.820
k = 6
laplacian_regularization = 0.8
laplacian_normalization = ''
c_l = 50
c_u = 50
# number of labeled (unmasked) nodes provided to the hfs algorithm
l = 4
# mask labels
Y_masked = mask_labels(Y, l)
"""
compute hfs solution using either soft_hfs or hard_hfs
"""
labels = hard_hfs(X, Y_masked, laplacian_regularization, var, eps, k, laplacian_normalization)
# labels = soft_hfs(X, Y_masked, c_l , c_u, laplacian_regularization, var, eps, k, laplacian_normalization)
"""
Visualize results
"""
plot_classification(X, Y, labels, var=var, eps=0, k=k)
accuracy = np.mean(labels == np.squeeze(Y))
print('The corresponding accuracy is : {} %'.format(accuracy * 100))
return accuracy
def two_moons_hfs():
"""
TO BE COMPLETED.
HFS for two_moons data.
"""
"""
Load the data. At home, try to use the larger dataset (question 1.2).
"""
# load the data
in_data = loadmat(os.path.join('data', 'data_2moons_hfs.mat'))
#in_data = loadmat(os.path.join('data', 'data_2moons_hfs_large.mat'))
X = in_data['X']
Y = in_data['Y'].squeeze()
"""
Choose the experiment parameters
"""
var = 1.0
eps = 0
k = 6
laplacian_regularization = 0.000000001
laplacian_normalization = ''
c_l = 0.96
c_u = 0.04
# number of labeled (unmasked) nodes provided to the hfs algorithm
l = 4
# mask labels
Y_masked = mask_labels(Y, l)
if np.size(np.where(Y_masked == 1)) == 4 or np.size(
np.where(Y_masked == 2)) == 4:
print(
"WARNING : the randomness was such that the four revealed labels are in the same class"
)
"""
compute hfs solution using either soft_hfs or hard_hfs
"""
#labels = hard_hfs(X, Y_masked, laplacian_regularization = laplacian_regularization, var = var, eps=eps, k=k, laplacian_normalization = laplacian_normalization)
labels = soft_hfs(X, Y_masked, c_l, c_u, laplacian_regularization, var,
eps, k, laplacian_normalization)
"""
Visualize results
"""
plot_classification(X, Y, labels, var=var, eps=0, k=k)
accuracy = np.mean(labels == np.squeeze(Y))
return accuracy
def two_moons_hfs(small_dataset=True, method='hard', plot=True):
"""
TO BE COMPLETED.
HFS for two_moons data.
"""
"""
Load the data. At home, try to use the larger dataset (question 1.2).
"""
# load the data
if small_dataset: # questions 1.1 and 1.3
in_data = loadmat(os.path.join('data', 'data_2moons_hfs.mat'))
else: # question 1.2
in_data = loadmat(os.path.join('data', 'data_2moons_hfs_large.mat'))
X = in_data['X']
Y = in_data['Y'].squeeze()
# automatically infer number of labels from samples
num_samples = np.size(Y, 0)
num_classes = len(np.unique(Y))
"""
Choose the experiment parameters
"""
var = 1
eps = 0
if small_dataset: # questions 1.1 and 1.3 parameter
k = 12
else: # question 1.2 parameter
k = 40
laplacian_regularization = 0
laplacian_normalization = "rw"
c_l = .9
c_u = .1
# number of labeled (unmasked) nodes provided to the hfs algorithm
l = 4
# mask labels
Y_masked = mask_labels(Y, l)
"""
compute hfs solution using either soft_hfs or hard_hfs
"""
if method == 'hard':
labels = hard_hfs(X, Y_masked, laplacian_regularization, var, eps, k,
laplacian_normalization)
else: # if method == 'soft'
labels = soft_hfs(X, Y_masked, c_l, c_u, laplacian_regularization, var,
eps, k, laplacian_normalization)
"""
Visualize results
"""
if plot:
if method == 'hard':
method_name = 'Hard HFS'
else:
method_name = 'Soft HFS'
plot_classification(X,
Y,
labels,
var=var,
eps=0,
k=k,
method=method_name)
accuracy = np.mean(labels == np.squeeze(Y))
return accuracy