xref: /petsc/src/tao/leastsquares/impls/brgn/brgn.h (revision 7d5fd1e4d9337468ad3f05b65b7facdcd2dfd2a4)
1 /*
2 Context for Bounded Regularized Gauss-Newton algorithm.
3 Extended with L1-regularizer with a linear transformation matrix D:
4 0.5*||Ax-b||^2 + lambda*||D*x||_1
5 When D is an identity matrix, we have the classic lasso, aka basis pursuit denoising in compressive sensing problem.
6 */
7 
8 #if !defined(__TAO_BRGN_H)
9 #define __TAO_BRGN_H
10 
11 #include <../src/tao/bound/impls/bnk/bnk.h>  /* BNLS, a sub-type of BNK, is used in brgn solver */
12 
13 typedef struct {
14   PetscErrorCode (*regularizerobjandgrad)(Tao,Vec,PetscReal*,Vec,void*);
15   PetscErrorCode (*regularizerhessian)(Tao,Vec,Mat,void*);
16   void           *reg_obj_ctx;
17   void           *reg_hess_ctx;
18   Mat            H,Hreg,D;  /* Hessian, Hessian for regulization part, and Dictionary matrix have size N*N, and K*N respectively. (Jacobian M*N not used here) */
19   Vec            x_old,x_work,r_work,diag,y,y_work;  /* x, r=J*x, and y=D*x have size N, M, and K respectively. */
20   Vec            damping;  /* Optional diagonal damping matrix. */
21   Tao            subsolver,parent;
22   PetscReal      lambda,epsilon,fc_old; /* lambda is regularizer weight for both L2-norm Gaussian-Newton and L1-norm, ||x||_1 is approximated with sum(sqrt(x.^2+epsilon^2)-epsilon)*/
23   PetscReal      downhill_lambda_change,uphill_lambda_change; /* With the lm regularizer lambda diag(J^T J),
24                                                                  lambda = downhill_lambda_change * lambda on steps that decrease the objective.
25                                                                  lambda = uphill_lambda_change * lambda on steps that increase the objective. */
26   PetscInt       reg_type;
27   PetscBool      mat_explicit;
28 } TAO_BRGN;
29 
30 #endif /* if !defined(__TAO_BRGN_H) */
31