xref: /petsc/src/tao/leastsquares/tutorials/cs1.c (revision 40badf4fbc550ac1f60bd080eaff6de6d55b946d)
1 /* XH: todo add cs1f.F90 and asjust makefile */
2 /*
3    Include "petsctao.h" so that we can use TAO solvers.  Note that this
4    file automatically includes libraries such as:
5      petsc.h       - base PETSc routines   petscvec.h - vectors
6      petscsys.h    - system routines        petscmat.h - matrices
7      petscis.h     - index sets            petscksp.h - Krylov subspace methods
8      petscviewer.h - viewers               petscpc.h  - preconditioners
9 
10 */
11 
12 #include <petsctao.h>
13 
14 /*
15 Description:   Compressive sensing test example 1.
16                0.5*||Ax-b||^2 + lambda*||D*x||_1
17                Xiang Huang: Nov 19, 2018
18 
19 Reference:     None
20 */
21 
22 static char help[] = "Finds the least-squares solution to the under constraint linear model Ax = b, with L1-norm regularizer. \n\
23             A is a M*N real matrix (M<N), x is sparse. \n\
24             We find the sparse solution by solving 0.5*||Ax-b||^2 + lambda*||D*x||_1, where lambda (by default 1e-4) is a user specified weight.\n\
25             D is the K*N transform matrix so that D*x is sparse. By default D is identity matrix, so that D*x = x.\n";
26 /*T
27    Concepts: TAO^Solving a system of nonlinear equations, nonlinear least squares
28    Routines: TaoCreate();
29    Routines: TaoSetType();
30    Routines: TaoSetSeparableObjectiveRoutine();
31    Routines: TaoSetJacobianRoutine();
32    Routines: TaoSetSolution();
33    Routines: TaoSetFromOptions();
34    Routines: TaoSetConvergenceHistory(); TaoGetConvergenceHistory();
35    Routines: TaoSolve();
36    Routines: TaoView(); TaoDestroy();
37    Processors: 1
38 T*/
39 
40 #define M 3
41 #define N 5
42 #define K 4
43 
44 /* User-defined application context */
45 typedef struct {
46   /* Working space. linear least square:  f(x) = A*x - b */
47   PetscReal A[M][N];    /* array of coefficients */
48   PetscReal b[M];       /* array of observations */
49   PetscReal xGT[M];     /* array of ground truth object, which can be used to compare the reconstruction result */
50   PetscReal D[K][N];    /* array of coefficients for 0.5*||Ax-b||^2 + lambda*||D*x||_1 */
51   PetscReal J[M][N];    /* dense jacobian matrix array. For linear least square, J = A. For nonlinear least square, it is different from A */
52   PetscInt  idm[M];     /* Matrix row, column indices for jacobian and dictionary */
53   PetscInt  idn[N];
54   PetscInt  idk[K];
55 } AppCtx;
56 
57 /* User provided Routines */
58 PetscErrorCode InitializeUserData(AppCtx *);
59 PetscErrorCode FormStartingPoint(Vec);
60 PetscErrorCode FormDictionaryMatrix(Mat,AppCtx *);
61 PetscErrorCode EvaluateFunction(Tao,Vec,Vec,void *);
62 PetscErrorCode EvaluateJacobian(Tao,Vec,Mat,Mat,void *);
63 
64 /*--------------------------------------------------------------------*/
65 int main(int argc,char **argv)
66 {
67   PetscErrorCode ierr;               /* used to check for functions returning nonzeros */
68   Vec            x,f;               /* solution, function f(x) = A*x-b */
69   Mat            J,D;               /* Jacobian matrix, Transform matrix */
70   Tao            tao;                /* Tao solver context */
71   PetscInt       i;                  /* iteration information */
72   PetscReal      hist[100],resid[100];
73   PetscInt       lits[100];
74   AppCtx         user;               /* user-defined work context */
75 
76   ierr = PetscInitialize(&argc,&argv,(char *)0,help);if (ierr) return ierr;
77 
78   /* Allocate solution and vector function vectors */
79   CHKERRQ(VecCreateSeq(PETSC_COMM_SELF,N,&x));
80   CHKERRQ(VecCreateSeq(PETSC_COMM_SELF,M,&f));
81 
82   /* Allocate Jacobian and Dictionary matrix. */
83   CHKERRQ(MatCreateSeqDense(PETSC_COMM_SELF,M,N,NULL,&J));
84   CHKERRQ(MatCreateSeqDense(PETSC_COMM_SELF,K,N,NULL,&D)); /* XH: TODO: dense -> sparse/dense/shell etc, do it on fly  */
85 
86   for (i=0;i<M;i++) user.idm[i] = i;
87   for (i=0;i<N;i++) user.idn[i] = i;
88   for (i=0;i<K;i++) user.idk[i] = i;
89 
90   /* Create TAO solver and set desired solution method */
91   CHKERRQ(TaoCreate(PETSC_COMM_SELF,&tao));
92   CHKERRQ(TaoSetType(tao,TAOBRGN));
93 
94   /* User set application context: A, D matrice, and b vector. */
95   CHKERRQ(InitializeUserData(&user));
96 
97   /* Set initial guess */
98   CHKERRQ(FormStartingPoint(x));
99 
100   /* Fill the content of matrix D from user application Context */
101   CHKERRQ(FormDictionaryMatrix(D,&user));
102 
103   /* Bind x to tao->solution. */
104   CHKERRQ(TaoSetSolution(tao,x));
105   /* Bind D to tao->data->D */
106   CHKERRQ(TaoBRGNSetDictionaryMatrix(tao,D));
107 
108   /* Set the function and Jacobian routines. */
109   CHKERRQ(TaoSetResidualRoutine(tao,f,EvaluateFunction,(void*)&user));
110   CHKERRQ(TaoSetJacobianResidualRoutine(tao,J,J,EvaluateJacobian,(void*)&user));
111 
112   /* Check for any TAO command line arguments */
113   CHKERRQ(TaoSetFromOptions(tao));
114 
115   CHKERRQ(TaoSetConvergenceHistory(tao,hist,resid,0,lits,100,PETSC_TRUE));
116 
117   /* Perform the Solve */
118   CHKERRQ(TaoSolve(tao));
119 
120   /* XH: Debug: View the result, function and Jacobian.  */
121   CHKERRQ(PetscPrintf(PETSC_COMM_SELF, "-------- result x, residual f=A*x-b, and Jacobian=A. -------- \n"));
122   CHKERRQ(VecView(x,PETSC_VIEWER_STDOUT_SELF));
123   CHKERRQ(VecView(f,PETSC_VIEWER_STDOUT_SELF));
124   CHKERRQ(MatView(J,PETSC_VIEWER_STDOUT_SELF));
125   CHKERRQ(MatView(D,PETSC_VIEWER_STDOUT_SELF));
126 
127   /* Free TAO data structures */
128   CHKERRQ(TaoDestroy(&tao));
129 
130    /* Free PETSc data structures */
131   CHKERRQ(VecDestroy(&x));
132   CHKERRQ(VecDestroy(&f));
133   CHKERRQ(MatDestroy(&J));
134   CHKERRQ(MatDestroy(&D));
135 
136   ierr = PetscFinalize();
137   return ierr;
138 }
139 
140 /*--------------------------------------------------------------------*/
141 PetscErrorCode EvaluateFunction(Tao tao, Vec X, Vec F, void *ptr)
142 {
143   AppCtx         *user = (AppCtx *)ptr;
144   PetscInt       m,n;
145   const PetscReal *x;
146   PetscReal      *b=user->b,*f;
147 
148   PetscFunctionBegin;
149   CHKERRQ(VecGetArrayRead(X,&x));
150   CHKERRQ(VecGetArray(F,&f));
151 
152   /* Even for linear least square, we do not direct use matrix operation f = A*x - b now, just for future modification and compatibility for nonlinear least square */
153   for (m=0;m<M;m++) {
154     f[m] = -b[m];
155     for (n=0;n<N;n++) {
156       f[m] += user->A[m][n]*x[n];
157     }
158   }
159   CHKERRQ(VecRestoreArrayRead(X,&x));
160   CHKERRQ(VecRestoreArray(F,&f));
161   PetscLogFlops(2.0*M*N);
162   PetscFunctionReturn(0);
163 }
164 
165 /*------------------------------------------------------------*/
166 /* J[m][n] = df[m]/dx[n] */
167 PetscErrorCode EvaluateJacobian(Tao tao, Vec X, Mat J, Mat Jpre, void *ptr)
168 {
169   AppCtx         *user = (AppCtx *)ptr;
170   PetscInt       m,n;
171   const PetscReal *x;
172 
173   PetscFunctionBegin;
174   CHKERRQ(VecGetArrayRead(X,&x)); /* not used for linear least square, but keep for future nonlinear least square) */
175   /* XH: TODO:  For linear least square, we can just set J=A fixed once, instead of keep update it! Maybe just create a function getFixedJacobian?
176     For nonlinear least square, we require x to compute J, keep codes here for future nonlinear least square*/
177   for (m=0; m<M; ++m) {
178     for (n=0; n<N; ++n) {
179       user->J[m][n] = user->A[m][n];
180     }
181   }
182 
183   CHKERRQ(MatSetValues(J,M,user->idm,N,user->idn,(PetscReal *)user->J,INSERT_VALUES));
184   CHKERRQ(MatAssemblyBegin(J,MAT_FINAL_ASSEMBLY));
185   CHKERRQ(MatAssemblyEnd(J,MAT_FINAL_ASSEMBLY));
186 
187   CHKERRQ(VecRestoreArrayRead(X,&x));/* not used for linear least square, but keep for future nonlinear least square) */
188   PetscLogFlops(0);  /* 0 for linear least square, >0 for nonlinear least square */
189   PetscFunctionReturn(0);
190 }
191 
192 /* ------------------------------------------------------------ */
193 /* Currently fixed matrix, in future may be dynamic for D(x)? */
194 PetscErrorCode FormDictionaryMatrix(Mat D,AppCtx *user)
195 {
196   PetscFunctionBegin;
197   CHKERRQ(MatSetValues(D,K,user->idk,N,user->idn,(PetscReal *)user->D,INSERT_VALUES));
198   CHKERRQ(MatAssemblyBegin(D,MAT_FINAL_ASSEMBLY));
199   CHKERRQ(MatAssemblyEnd(D,MAT_FINAL_ASSEMBLY));
200 
201   PetscLogFlops(0); /* 0 for fixed dictionary matrix, >0 for varying dictionary matrix */
202   PetscFunctionReturn(0);
203 }
204 
205 /* ------------------------------------------------------------ */
206 PetscErrorCode FormStartingPoint(Vec X)
207 {
208   PetscFunctionBegin;
209   CHKERRQ(VecSet(X,0.0));
210   PetscFunctionReturn(0);
211 }
212 
213 /* ---------------------------------------------------------------------- */
214 PetscErrorCode InitializeUserData(AppCtx *user)
215 {
216   PetscReal *b=user->b; /* **A=user->A, but we don't kown the dimension of A in this way, how to fix? */
217   PetscInt  m,n,k; /* loop index for M,N,K dimension. */
218 
219   PetscFunctionBegin;
220   /* b = A*x while x = [0;0;1;0;0] here*/
221   m = 0;
222   b[m++] = 0.28;
223   b[m++] = 0.55;
224   b[m++] = 0.96;
225 
226   /* matlab generated random matrix, uniformly distributed in [0,1] with 2 digits accuracy. rng(0); A = rand(M, N); A = round(A*100)/100;
227   A = [0.81  0.91  0.28  0.96  0.96
228        0.91  0.63  0.55  0.16  0.49
229        0.13  0.10  0.96  0.97  0.80]
230   */
231   m=0; n=0; user->A[m][n++] = 0.81; user->A[m][n++] = 0.91; user->A[m][n++] = 0.28; user->A[m][n++] = 0.96; user->A[m][n++] = 0.96;
232   ++m; n=0; user->A[m][n++] = 0.91; user->A[m][n++] = 0.63; user->A[m][n++] = 0.55; user->A[m][n++] = 0.16; user->A[m][n++] = 0.49;
233   ++m; n=0; user->A[m][n++] = 0.13; user->A[m][n++] = 0.10; user->A[m][n++] = 0.96; user->A[m][n++] = 0.97; user->A[m][n++] = 0.80;
234 
235   /* initialize to 0 */
236   for (k=0; k<K; k++) {
237     for (n=0; n<N; n++) {
238       user->D[k][n] = 0.0;
239     }
240   }
241   /* Choice I: set D to identity matrix of size N*N for testing */
242   /* for (k=0; k<K; k++) user->D[k][k] = 1.0; */
243   /* Choice II: set D to Backward difference matrix of size (N-1)*N, with zero extended boundary assumption */
244   for (k=0;k<K;k++) {
245       user->D[k][k]   = -1.0;
246       user->D[k][k+1] = 1.0;
247   }
248 
249   PetscFunctionReturn(0);
250 }
251 
252 /*TEST
253 
254    build:
255       requires: !complex !single !quad !defined(PETSC_USE_64BIT_INDICES)
256 
257    test:
258       localrunfiles: cs1Data_A_b_xGT
259       args: -tao_smonitor -tao_max_it 100 -tao_type pounders -tao_gatol 1.e-6
260 
261    test:
262       suffix: 2
263       localrunfiles: cs1Data_A_b_xGT
264       args: -tao_monitor -tao_max_it 100 -tao_type brgn -tao_brgn_regularization_type l2prox -tao_brgn_regularizer_weight 1e-8 -tao_gatol 1.e-6 -tao_brgn_subsolver_tao_bnk_ksp_converged_reason
265 
266    test:
267       suffix: 3
268       localrunfiles: cs1Data_A_b_xGT
269       args: -tao_monitor -tao_max_it 100 -tao_type brgn -tao_brgn_regularization_type l1dict -tao_brgn_regularizer_weight 1e-8 -tao_brgn_l1_smooth_epsilon 1e-6 -tao_gatol 1.e-6
270 
271    test:
272       suffix: 4
273       localrunfiles: cs1Data_A_b_xGT
274       args: -tao_monitor -tao_max_it 100 -tao_type brgn -tao_brgn_regularization_type l2pure -tao_brgn_regularizer_weight 1e-8 -tao_gatol 1.e-6
275 
276    test:
277       suffix: 5
278       localrunfiles: cs1Data_A_b_xGT
279       args: -tao_monitor -tao_max_it 100 -tao_type brgn -tao_brgn_regularization_type lm -tao_gatol 1.e-6 -tao_brgn_subsolver_tao_type bnls
280 
281 TEST*/
282