def fit(self, Y, iterations=100, verbose=False):
"""
Y: numpy data [n_features, n_samples]
k: interger: number of atoms in the dictionary
if k is None, select k = round(0.2*n_samples)
"""
self.Y = Y
del Y
Y_range = np.array([0, self.Y.shape[1]])
D_range = np.array([0, self.k])
self.D = utils.pickDfromY(self.Y, Y_range, D_range)
self.X = np.zeros((self.D.shape[1], self.Y.shape[1]))
for it in range(iterations):
# update X
lasso = optimize.Lasso(self.D, self.lambd)
lasso.fit(self.Y, Xinit=self.X)
self.X = lasso.coef_
# update D
F = np.dot(self.X, self.X.T)
E = np.dot(self.Y, self.X.T)
self.D = optimize.ODL_updateD(self.D,
E,
F,
iterations=self.updateD_iters)
if verbose:
print('iter \t%d/%d \t\t loss \t%.4f' %
(it, iterations, self.loss()))
def predict(self, Y):
E = np.zeros((self.nclass, Y.shape[1]))
for c in range(self.nclass):
Dc = self._getDc(c)
lasso = optimize.Lasso(Dc, self.lambd)
lasso.fit(Y)
Xc = lasso.solve()
R1 = Y - np.dot(Dc, Xc)
E[c, :] = 0.5 * (R1 *
R1).sum(axis=0) + self.lambd * abs(Xc).sum(axis=0)
return np.argmin(E, axis=0) + 1
def predict(self, Y):
N = Y.shape[1]
lambda_list = [self.lambd]
for lambd in lambda_list:
E = np.zeros((self.nclass, N))
for c in range(self.nclass):
# Dc in D only
Dc_ = get_block_col(self.D, c, self.D_range)
# Dc in D and D0
Dc = np.hstack((Dc_, self.D0)) if self.k0 > 0 else Dc_
lasso = optimize.Lasso(Dc, lambd=lambd)
lasso.fit(Y)
Xc = lasso.solve()
R = Y - np.dot(Dc, Xc)
E[c, :] = 0.5*np.sum(R*R, axis = 0) + \
lambd*np.sum(np.abs(Xc), axis = 0)
pred = np.argmin(E, axis=0) + 1
return pred
pass
def _updateXc(self, c):
lasso = optimize.Lasso(self._getDc(c), self.lambd)
self.X[c] = lasso.fit(self._getYc(c), Xinit=self.X[c])
self.X[c] = lasso.coef_