xref: /petsc/src/tao/leastsquares/tutorials/cs1.c (revision ebead697dbf761eb322f829370bbe90b3bd93fa3)
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 
27 #define M 3
28 #define N 5
29 #define K 4
30 
31 /* User-defined application context */
32 typedef struct {
33   /* Working space. linear least square:  f(x) = A*x - b */
34   PetscReal A[M][N];    /* array of coefficients */
35   PetscReal b[M];       /* array of observations */
36   PetscReal xGT[M];     /* array of ground truth object, which can be used to compare the reconstruction result */
37   PetscReal D[K][N];    /* array of coefficients for 0.5*||Ax-b||^2 + lambda*||D*x||_1 */
38   PetscReal J[M][N];    /* dense jacobian matrix array. For linear least square, J = A. For nonlinear least square, it is different from A */
39   PetscInt  idm[M];     /* Matrix row, column indices for jacobian and dictionary */
40   PetscInt  idn[N];
41   PetscInt  idk[K];
42 } AppCtx;
43 
44 /* User provided Routines */
45 PetscErrorCode InitializeUserData(AppCtx *);
46 PetscErrorCode FormStartingPoint(Vec);
47 PetscErrorCode FormDictionaryMatrix(Mat,AppCtx *);
48 PetscErrorCode EvaluateFunction(Tao,Vec,Vec,void *);
49 PetscErrorCode EvaluateJacobian(Tao,Vec,Mat,Mat,void *);
50 
51 /*--------------------------------------------------------------------*/
52 int main(int argc,char **argv)
53 {
54   Vec            x,f;               /* solution, function f(x) = A*x-b */
55   Mat            J,D;               /* Jacobian matrix, Transform matrix */
56   Tao            tao;                /* Tao solver context */
57   PetscInt       i;                  /* iteration information */
58   PetscReal      hist[100],resid[100];
59   PetscInt       lits[100];
60   AppCtx         user;               /* user-defined work context */
61 
62   PetscFunctionBeginUser;
63   PetscCall(PetscInitialize(&argc,&argv,(char *)0,help));
64 
65   /* Allocate solution and vector function vectors */
66   PetscCall(VecCreateSeq(PETSC_COMM_SELF,N,&x));
67   PetscCall(VecCreateSeq(PETSC_COMM_SELF,M,&f));
68 
69   /* Allocate Jacobian and Dictionary matrix. */
70   PetscCall(MatCreateSeqDense(PETSC_COMM_SELF,M,N,NULL,&J));
71   PetscCall(MatCreateSeqDense(PETSC_COMM_SELF,K,N,NULL,&D)); /* XH: TODO: dense -> sparse/dense/shell etc, do it on fly  */
72 
73   for (i=0;i<M;i++) user.idm[i] = i;
74   for (i=0;i<N;i++) user.idn[i] = i;
75   for (i=0;i<K;i++) user.idk[i] = i;
76 
77   /* Create TAO solver and set desired solution method */
78   PetscCall(TaoCreate(PETSC_COMM_SELF,&tao));
79   PetscCall(TaoSetType(tao,TAOBRGN));
80 
81   /* User set application context: A, D matrice, and b vector. */
82   PetscCall(InitializeUserData(&user));
83 
84   /* Set initial guess */
85   PetscCall(FormStartingPoint(x));
86 
87   /* Fill the content of matrix D from user application Context */
88   PetscCall(FormDictionaryMatrix(D,&user));
89 
90   /* Bind x to tao->solution. */
91   PetscCall(TaoSetSolution(tao,x));
92   /* Bind D to tao->data->D */
93   PetscCall(TaoBRGNSetDictionaryMatrix(tao,D));
94 
95   /* Set the function and Jacobian routines. */
96   PetscCall(TaoSetResidualRoutine(tao,f,EvaluateFunction,(void*)&user));
97   PetscCall(TaoSetJacobianResidualRoutine(tao,J,J,EvaluateJacobian,(void*)&user));
98 
99   /* Check for any TAO command line arguments */
100   PetscCall(TaoSetFromOptions(tao));
101 
102   PetscCall(TaoSetConvergenceHistory(tao,hist,resid,0,lits,100,PETSC_TRUE));
103 
104   /* Perform the Solve */
105   PetscCall(TaoSolve(tao));
106 
107   /* XH: Debug: View the result, function and Jacobian.  */
108   PetscCall(PetscPrintf(PETSC_COMM_SELF, "-------- result x, residual f=A*x-b, and Jacobian=A. -------- \n"));
109   PetscCall(VecView(x,PETSC_VIEWER_STDOUT_SELF));
110   PetscCall(VecView(f,PETSC_VIEWER_STDOUT_SELF));
111   PetscCall(MatView(J,PETSC_VIEWER_STDOUT_SELF));
112   PetscCall(MatView(D,PETSC_VIEWER_STDOUT_SELF));
113 
114   /* Free TAO data structures */
115   PetscCall(TaoDestroy(&tao));
116 
117    /* Free PETSc data structures */
118   PetscCall(VecDestroy(&x));
119   PetscCall(VecDestroy(&f));
120   PetscCall(MatDestroy(&J));
121   PetscCall(MatDestroy(&D));
122 
123   PetscCall(PetscFinalize());
124   return 0;
125 }
126 
127 /*--------------------------------------------------------------------*/
128 PetscErrorCode EvaluateFunction(Tao tao, Vec X, Vec F, void *ptr)
129 {
130   AppCtx         *user = (AppCtx *)ptr;
131   PetscInt       m,n;
132   const PetscReal *x;
133   PetscReal      *b=user->b,*f;
134 
135   PetscFunctionBegin;
136   PetscCall(VecGetArrayRead(X,&x));
137   PetscCall(VecGetArray(F,&f));
138 
139   /* 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 */
140   for (m=0;m<M;m++) {
141     f[m] = -b[m];
142     for (n=0;n<N;n++) {
143       f[m] += user->A[m][n]*x[n];
144     }
145   }
146   PetscCall(VecRestoreArrayRead(X,&x));
147   PetscCall(VecRestoreArray(F,&f));
148   PetscLogFlops(2.0*M*N);
149   PetscFunctionReturn(0);
150 }
151 
152 /*------------------------------------------------------------*/
153 /* J[m][n] = df[m]/dx[n] */
154 PetscErrorCode EvaluateJacobian(Tao tao, Vec X, Mat J, Mat Jpre, void *ptr)
155 {
156   AppCtx         *user = (AppCtx *)ptr;
157   PetscInt       m,n;
158   const PetscReal *x;
159 
160   PetscFunctionBegin;
161   PetscCall(VecGetArrayRead(X,&x)); /* not used for linear least square, but keep for future nonlinear least square) */
162   /* 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?
163     For nonlinear least square, we require x to compute J, keep codes here for future nonlinear least square*/
164   for (m=0; m<M; ++m) {
165     for (n=0; n<N; ++n) {
166       user->J[m][n] = user->A[m][n];
167     }
168   }
169 
170   PetscCall(MatSetValues(J,M,user->idm,N,user->idn,(PetscReal *)user->J,INSERT_VALUES));
171   PetscCall(MatAssemblyBegin(J,MAT_FINAL_ASSEMBLY));
172   PetscCall(MatAssemblyEnd(J,MAT_FINAL_ASSEMBLY));
173 
174   PetscCall(VecRestoreArrayRead(X,&x));/* not used for linear least square, but keep for future nonlinear least square) */
175   PetscLogFlops(0);  /* 0 for linear least square, >0 for nonlinear least square */
176   PetscFunctionReturn(0);
177 }
178 
179 /* ------------------------------------------------------------ */
180 /* Currently fixed matrix, in future may be dynamic for D(x)? */
181 PetscErrorCode FormDictionaryMatrix(Mat D,AppCtx *user)
182 {
183   PetscFunctionBegin;
184   PetscCall(MatSetValues(D,K,user->idk,N,user->idn,(PetscReal *)user->D,INSERT_VALUES));
185   PetscCall(MatAssemblyBegin(D,MAT_FINAL_ASSEMBLY));
186   PetscCall(MatAssemblyEnd(D,MAT_FINAL_ASSEMBLY));
187 
188   PetscLogFlops(0); /* 0 for fixed dictionary matrix, >0 for varying dictionary matrix */
189   PetscFunctionReturn(0);
190 }
191 
192 /* ------------------------------------------------------------ */
193 PetscErrorCode FormStartingPoint(Vec X)
194 {
195   PetscFunctionBegin;
196   PetscCall(VecSet(X,0.0));
197   PetscFunctionReturn(0);
198 }
199 
200 /* ---------------------------------------------------------------------- */
201 PetscErrorCode InitializeUserData(AppCtx *user)
202 {
203   PetscReal *b=user->b; /* **A=user->A, but we don't kown the dimension of A in this way, how to fix? */
204   PetscInt  m,n,k; /* loop index for M,N,K dimension. */
205 
206   PetscFunctionBegin;
207   /* b = A*x while x = [0;0;1;0;0] here*/
208   m = 0;
209   b[m++] = 0.28;
210   b[m++] = 0.55;
211   b[m++] = 0.96;
212 
213   /* matlab generated random matrix, uniformly distributed in [0,1] with 2 digits accuracy. rng(0); A = rand(M, N); A = round(A*100)/100;
214   A = [0.81  0.91  0.28  0.96  0.96
215        0.91  0.63  0.55  0.16  0.49
216        0.13  0.10  0.96  0.97  0.80]
217   */
218   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;
219   ++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;
220   ++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;
221 
222   /* initialize to 0 */
223   for (k=0; k<K; k++) {
224     for (n=0; n<N; n++) {
225       user->D[k][n] = 0.0;
226     }
227   }
228   /* Choice I: set D to identity matrix of size N*N for testing */
229   /* for (k=0; k<K; k++) user->D[k][k] = 1.0; */
230   /* Choice II: set D to Backward difference matrix of size (N-1)*N, with zero extended boundary assumption */
231   for (k=0;k<K;k++) {
232       user->D[k][k]   = -1.0;
233       user->D[k][k+1] = 1.0;
234   }
235 
236   PetscFunctionReturn(0);
237 }
238 
239 /*TEST
240 
241    build:
242       requires: !complex !single !quad !defined(PETSC_USE_64BIT_INDICES)
243 
244    test:
245       localrunfiles: cs1Data_A_b_xGT
246       args: -tao_smonitor -tao_max_it 100 -tao_type pounders -tao_gatol 1.e-6
247 
248    test:
249       suffix: 2
250       localrunfiles: cs1Data_A_b_xGT
251       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
252 
253    test:
254       suffix: 3
255       localrunfiles: cs1Data_A_b_xGT
256       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
257 
258    test:
259       suffix: 4
260       localrunfiles: cs1Data_A_b_xGT
261       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
262 
263    test:
264       suffix: 5
265       localrunfiles: cs1Data_A_b_xGT
266       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
267 
268 TEST*/
269