def train(self, X_train, y_train, param_set=None):
'''train the classifier'''
n_classes = len(set(y_train))
if self.class_dict_coder is not None:
# ksvd in each class dictionary
self.class_dict_coder.mmap = self.mmap
self.class_dict_coder.n_jobs = self.n_jobs
self.D = self.class_dict_coder(X_train, y_train)
self.n_class_atoms = self.class_dict_coder.n_class_atoms
else:
# every datapoint is an atom
self.n_class_atoms = (np.zeros(n_classes) +
self.n_class_samples).astype(int)
n_total_atoms = np.sum(self.n_class_atoms)
n_features = X_train.shape[0]
D = np.zeros((n_features, n_total_atoms))
for c in range(n_classes):
x_c = y_train == c
Xc = X_train[:, x_c]
n_class_samples = Xc.shape[1]
Dc = init_dictionary(Xc,
self.n_class_atoms[c],
method='data',
normalize=True)
base = c * self.n_class_atoms[c]
offset = self.n_class_atoms[c]
D[:, base:base + offset] = Dc
self.D = D
def test_init_dictionary(self):
X = np.array([
[1, 2, 3, 4, 5],
[0, 2, 1, 2, 1]
])
n_datapoints = 5
n_atoms = 3
D = init_dictionary(X, n_atoms, method='data', return_unused_data=False, normalize=False)
self.assertEqual(D.shape, (2, n_atoms))
self.assertEqual(np.sum(np.array_equal(D[:, i], X[:, j])
for i in range(n_atoms) for j in range(n_datapoints)), n_atoms)
def train(self, X_train, y_train, param_set=None):
'''train the classifier'''
n_classes = len(set(y_train))
if self.D is None:
if self.class_dict_coder is not None:
# ksvd in each class dictionary
self.class_dict_coder.mmap = self.mmap
self.class_dict_coder.n_jobs = self.n_jobs
self.D = self.class_dict_coder(X_train, y_train)
self.n_class_atoms = self.class_dict_coder.n_class_atoms
else:
# every datapoint is an atom
self.n_class_atoms = (np.zeros(n_classes) +
self.n_class_samples).astype(int)
n_total_atoms = np.sum(self.n_class_atoms)
n_features = X_train.shape[0]
D = np.zeros((n_features, n_total_atoms))
for c in range(n_classes):
x_c = y_train == c
Xc = X_train[:, x_c]
n_class_samples = Xc.shape[1]
Dc = init_dictionary(Xc,
self.n_class_atoms[c],
method='data',
normalize=True)
base = c * self.n_class_atoms[c]
offset = self.n_class_atoms[c]
D[:, base:base + offset] = Dc
self.D = D
from lyssa.classify import linear_svm
self.lsvm = linear_svm()
# if not self.features_extracted:
if self.method == "global":
self.Z_train = global_src_features(X_train,
self.D,
self.sparse_coder,
self.n_class_atoms,
n_jobs=self.n_jobs)
elif self.method == "local":
pass
# self.features_extracted = True
self.lsvm.train(self.Z_train, y_train, param_set=param_set)
def test_init_dictionary(self):
X = np.array([[1, 2, 3, 4, 5], [0, 2, 1, 2, 1]])
n_datapoints = 5
n_atoms = 3
D = init_dictionary(X,
n_atoms,
method='data',
return_unused_data=False,
normalize=False)
self.assertEqual(D.shape, (2, n_atoms))
self.assertEqual(
np.sum(
np.array_equal(D[:, i], X[:, j]) for i in range(n_atoms)
for j in range(n_datapoints)), n_atoms)