def get_el_tr_sum_loss_gradient(
self, x_vals_seq: Sequence[X], supervisory_seq: Sequence[float],
gamma_lambda: float) -> Sequence[np.ndarray]:
toeplitz_mat = get_decay_toeplitz_matrix(len(x_vals_seq), gamma_lambda)
errors = self.get_func_eval_pts(x_vals_seq) - supervisory_seq
func_grad = self.get_feature_vals_pts(x_vals_seq)
return [errors.dot(toeplitz_mat.dot(func_grad))]
def get_el_tr_sum_objective_gradient(
self,
x_vals_seq: Sequence[X],
dObj_dOL: np.ndarray,
factors: np.ndarray,
gamma_lambda: float
) -> Sequence[np.ndarray]:
toep = get_decay_toeplitz_matrix(len(x_vals_seq), gamma_lambda)
features = self.get_feature_vals_pts(x_vals_seq)
return [factors.dot(toep.dot(np.diag(dObj_dOL).dot(features)))]