1 #include <petsctaolinesearch.h> 2 #include <../src/tao/matrix/lmvmmat.h> 3 #include <../src/tao/bound/impls/blmvm/blmvm.h> 4 5 /*------------------------------------------------------------*/ 6 #undef __FUNCT__ 7 #define __FUNCT__ "TaoSolve_BLMVM" 8 static PetscErrorCode TaoSolve_BLMVM(Tao tao) 9 { 10 PetscErrorCode ierr; 11 TAO_BLMVM *blmP = (TAO_BLMVM *)tao->data; 12 TaoConvergedReason reason = TAO_CONTINUE_ITERATING; 13 TaoLineSearchConvergedReason ls_status = TAOLINESEARCH_CONTINUE_ITERATING; 14 PetscReal f, fold, gdx, gnorm; 15 PetscReal stepsize = 1.0,delta; 16 17 PetscFunctionBegin; 18 /* Project initial point onto bounds */ 19 ierr = TaoComputeVariableBounds(tao);CHKERRQ(ierr); 20 ierr = VecMedian(tao->XL,tao->solution,tao->XU,tao->solution);CHKERRQ(ierr); 21 ierr = TaoLineSearchSetVariableBounds(tao->linesearch,tao->XL,tao->XU);CHKERRQ(ierr); 22 23 /* Check convergence criteria */ 24 ierr = TaoComputeObjectiveAndGradient(tao, tao->solution,&f,blmP->unprojected_gradient);CHKERRQ(ierr); 25 ierr = VecBoundGradientProjection(blmP->unprojected_gradient,tao->solution, tao->XL,tao->XU,tao->gradient);CHKERRQ(ierr); 26 27 ierr = VecNorm(tao->gradient,NORM_2,&gnorm);CHKERRQ(ierr); 28 if (PetscIsInfOrNanReal(f) || PetscIsInfOrNanReal(gnorm)) SETERRQ(PETSC_COMM_SELF,1, "User provided compute function generated Inf pr NaN"); 29 30 ierr = TaoMonitor(tao, tao->niter, f, gnorm, 0.0, stepsize, &reason);CHKERRQ(ierr); 31 if (reason != TAO_CONTINUE_ITERATING) PetscFunctionReturn(0); 32 33 /* Set initial scaling for the function */ 34 if (f != 0.0) { 35 delta = 2.0*PetscAbsScalar(f) / (gnorm*gnorm); 36 } else { 37 delta = 2.0 / (gnorm*gnorm); 38 } 39 ierr = MatLMVMSetDelta(blmP->M,delta);CHKERRQ(ierr); 40 41 /* Set counter for gradient/reset steps */ 42 blmP->grad = 0; 43 blmP->reset = 0; 44 45 /* Have not converged; continue with Newton method */ 46 while (reason == TAO_CONTINUE_ITERATING) { 47 /* Compute direction */ 48 ierr = MatLMVMUpdate(blmP->M, tao->solution, tao->gradient);CHKERRQ(ierr); 49 ierr = MatLMVMSolve(blmP->M, blmP->unprojected_gradient, tao->stepdirection);CHKERRQ(ierr); 50 ierr = VecBoundGradientProjection(tao->stepdirection,tao->solution,tao->XL,tao->XU,tao->gradient);CHKERRQ(ierr); 51 52 /* Check for success (descent direction) */ 53 ierr = VecDot(blmP->unprojected_gradient, tao->gradient, &gdx);CHKERRQ(ierr); 54 if (gdx <= 0) { 55 /* Step is not descent or solve was not successful 56 Use steepest descent direction (scaled) */ 57 ++blmP->grad; 58 59 if (f != 0.0) { 60 delta = 2.0*PetscAbsScalar(f) / (gnorm*gnorm); 61 } else { 62 delta = 2.0 / (gnorm*gnorm); 63 } 64 ierr = MatLMVMSetDelta(blmP->M,delta);CHKERRQ(ierr); 65 ierr = MatLMVMReset(blmP->M);CHKERRQ(ierr); 66 ierr = MatLMVMUpdate(blmP->M, tao->solution, blmP->unprojected_gradient);CHKERRQ(ierr); 67 ierr = MatLMVMSolve(blmP->M,blmP->unprojected_gradient, tao->stepdirection);CHKERRQ(ierr); 68 } 69 ierr = VecScale(tao->stepdirection,-1.0);CHKERRQ(ierr); 70 71 /* Perform the linesearch */ 72 fold = f; 73 ierr = VecCopy(tao->solution, blmP->Xold);CHKERRQ(ierr); 74 ierr = VecCopy(blmP->unprojected_gradient, blmP->Gold);CHKERRQ(ierr); 75 ierr = TaoLineSearchSetInitialStepLength(tao->linesearch,1.0);CHKERRQ(ierr); 76 ierr = TaoLineSearchApply(tao->linesearch, tao->solution, &f, blmP->unprojected_gradient, tao->stepdirection, &stepsize, &ls_status);CHKERRQ(ierr); 77 ierr = TaoAddLineSearchCounts(tao);CHKERRQ(ierr); 78 79 if (ls_status != TAOLINESEARCH_SUCCESS && ls_status != TAOLINESEARCH_SUCCESS_USER) { 80 /* Linesearch failed 81 Reset factors and use scaled (projected) gradient step */ 82 ++blmP->reset; 83 84 f = fold; 85 ierr = VecCopy(blmP->Xold, tao->solution);CHKERRQ(ierr); 86 ierr = VecCopy(blmP->Gold, blmP->unprojected_gradient);CHKERRQ(ierr); 87 88 if (f != 0.0) { 89 delta = 2.0* PetscAbsScalar(f) / (gnorm*gnorm); 90 } else { 91 delta = 2.0/ (gnorm*gnorm); 92 } 93 ierr = MatLMVMSetDelta(blmP->M,delta);CHKERRQ(ierr); 94 ierr = MatLMVMReset(blmP->M);CHKERRQ(ierr); 95 ierr = MatLMVMUpdate(blmP->M, tao->solution, blmP->unprojected_gradient);CHKERRQ(ierr); 96 ierr = MatLMVMSolve(blmP->M, blmP->unprojected_gradient, tao->stepdirection);CHKERRQ(ierr); 97 ierr = VecScale(tao->stepdirection, -1.0);CHKERRQ(ierr); 98 99 /* This may be incorrect; linesearch has values fo stepmax and stepmin 100 that should be reset. */ 101 ierr = TaoLineSearchSetInitialStepLength(tao->linesearch,1.0);CHKERRQ(ierr); 102 ierr = TaoLineSearchApply(tao->linesearch,tao->solution,&f, blmP->unprojected_gradient, tao->stepdirection, &stepsize, &ls_status);CHKERRQ(ierr); 103 ierr = TaoAddLineSearchCounts(tao);CHKERRQ(ierr); 104 105 if (ls_status != TAOLINESEARCH_SUCCESS && ls_status != TAOLINESEARCH_SUCCESS_USER) { 106 tao->reason = TAO_DIVERGED_LS_FAILURE; 107 break; 108 } 109 } 110 111 /* Check for converged */ 112 ierr = VecBoundGradientProjection(blmP->unprojected_gradient, tao->solution, tao->XL, tao->XU, tao->gradient);CHKERRQ(ierr); 113 ierr = VecNorm(tao->gradient, NORM_2, &gnorm);CHKERRQ(ierr); 114 115 116 if (PetscIsInfOrNanReal(f) || PetscIsInfOrNanReal(gnorm)) SETERRQ(PETSC_COMM_SELF,1, "User provided compute function generated Not-a-Number"); 117 tao->niter++; 118 ierr = TaoMonitor(tao, tao->niter, f, gnorm, 0.0, stepsize, &reason);CHKERRQ(ierr); 119 } 120 PetscFunctionReturn(0); 121 } 122 123 #undef __FUNCT__ 124 #define __FUNCT__ "TaoSetup_BLMVM" 125 static PetscErrorCode TaoSetup_BLMVM(Tao tao) 126 { 127 TAO_BLMVM *blmP = (TAO_BLMVM *)tao->data; 128 PetscInt n,N; 129 PetscErrorCode ierr; 130 131 PetscFunctionBegin; 132 /* Existence of tao->solution checked in TaoSetup() */ 133 ierr = VecDuplicate(tao->solution,&blmP->Xold);CHKERRQ(ierr); 134 ierr = VecDuplicate(tao->solution,&blmP->Gold);CHKERRQ(ierr); 135 ierr = VecDuplicate(tao->solution, &blmP->unprojected_gradient);CHKERRQ(ierr); 136 137 if (!tao->stepdirection) { 138 ierr = VecDuplicate(tao->solution, &tao->stepdirection);CHKERRQ(ierr); 139 } 140 if (!tao->gradient) { 141 ierr = VecDuplicate(tao->solution,&tao->gradient);CHKERRQ(ierr); 142 } 143 if (!tao->XL) { 144 ierr = VecDuplicate(tao->solution,&tao->XL);CHKERRQ(ierr); 145 ierr = VecSet(tao->XL,PETSC_NINFINITY);CHKERRQ(ierr); 146 } 147 if (!tao->XU) { 148 ierr = VecDuplicate(tao->solution,&tao->XU);CHKERRQ(ierr); 149 ierr = VecSet(tao->XU,PETSC_INFINITY);CHKERRQ(ierr); 150 } 151 /* Create matrix for the limited memory approximation */ 152 ierr = VecGetLocalSize(tao->solution,&n);CHKERRQ(ierr); 153 ierr = VecGetSize(tao->solution,&N);CHKERRQ(ierr); 154 ierr = MatCreateLMVM(((PetscObject)tao)->comm,n,N,&blmP->M);CHKERRQ(ierr); 155 ierr = MatLMVMAllocateVectors(blmP->M,tao->solution);CHKERRQ(ierr); 156 PetscFunctionReturn(0); 157 } 158 159 /* ---------------------------------------------------------- */ 160 #undef __FUNCT__ 161 #define __FUNCT__ "TaoDestroy_BLMVM" 162 static PetscErrorCode TaoDestroy_BLMVM(Tao tao) 163 { 164 TAO_BLMVM *blmP = (TAO_BLMVM *)tao->data; 165 PetscErrorCode ierr; 166 167 PetscFunctionBegin; 168 if (tao->setupcalled) { 169 ierr = MatDestroy(&blmP->M);CHKERRQ(ierr); 170 ierr = VecDestroy(&blmP->unprojected_gradient);CHKERRQ(ierr); 171 ierr = VecDestroy(&blmP->Xold);CHKERRQ(ierr); 172 ierr = VecDestroy(&blmP->Gold);CHKERRQ(ierr); 173 } 174 ierr = PetscFree(tao->data);CHKERRQ(ierr); 175 PetscFunctionReturn(0); 176 } 177 178 /*------------------------------------------------------------*/ 179 #undef __FUNCT__ 180 #define __FUNCT__ "TaoSetFromOptions_BLMVM" 181 static PetscErrorCode TaoSetFromOptions_BLMVM(PetscOptions* PetscOptionsObject,Tao tao) 182 { 183 PetscErrorCode ierr; 184 185 PetscFunctionBegin; 186 ierr = PetscOptionsHead(PetscOptionsObject,"Limited-memory variable-metric method for bound constrained optimization");CHKERRQ(ierr); 187 ierr = TaoLineSearchSetFromOptions(tao->linesearch);CHKERRQ(ierr); 188 ierr = PetscOptionsTail();CHKERRQ(ierr); 189 PetscFunctionReturn(0); 190 } 191 192 193 /*------------------------------------------------------------*/ 194 #undef __FUNCT__ 195 #define __FUNCT__ "TaoView_BLMVM" 196 static int TaoView_BLMVM(Tao tao, PetscViewer viewer) 197 { 198 TAO_BLMVM *lmP = (TAO_BLMVM *)tao->data; 199 PetscBool isascii; 200 PetscErrorCode ierr; 201 202 PetscFunctionBegin; 203 ierr = PetscObjectTypeCompare((PetscObject)viewer, PETSCVIEWERASCII, &isascii);CHKERRQ(ierr); 204 if (isascii) { 205 ierr = PetscViewerASCIIPushTab(viewer);CHKERRQ(ierr); 206 ierr = PetscViewerASCIIPrintf(viewer, "Gradient steps: %D\n", lmP->grad);CHKERRQ(ierr); 207 ierr = PetscViewerASCIIPopTab(viewer);CHKERRQ(ierr); 208 } 209 PetscFunctionReturn(0); 210 } 211 212 #undef __FUNCT__ 213 #define __FUNCT__ "TaoComputeDual_BLMVM" 214 static PetscErrorCode TaoComputeDual_BLMVM(Tao tao, Vec DXL, Vec DXU) 215 { 216 TAO_BLMVM *blm = (TAO_BLMVM *) tao->data; 217 PetscErrorCode ierr; 218 219 PetscFunctionBegin; 220 PetscValidHeaderSpecific(tao,TAO_CLASSID,1); 221 PetscValidHeaderSpecific(DXL,VEC_CLASSID,2); 222 PetscValidHeaderSpecific(DXU,VEC_CLASSID,3); 223 if (!tao->gradient || !blm->unprojected_gradient) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ORDER,"Dual variables don't exist yet or no longer exist.\n"); 224 225 ierr = VecCopy(tao->gradient,DXL);CHKERRQ(ierr); 226 ierr = VecAXPY(DXL,-1.0,blm->unprojected_gradient);CHKERRQ(ierr); 227 ierr = VecSet(DXU,0.0);CHKERRQ(ierr); 228 ierr = VecPointwiseMax(DXL,DXL,DXU);CHKERRQ(ierr); 229 230 ierr = VecCopy(blm->unprojected_gradient,DXU);CHKERRQ(ierr); 231 ierr = VecAXPY(DXU,-1.0,tao->gradient);CHKERRQ(ierr); 232 ierr = VecAXPY(DXU,1.0,DXL);CHKERRQ(ierr); 233 PetscFunctionReturn(0); 234 } 235 236 /* ---------------------------------------------------------- */ 237 /*MC 238 TAOBLMVM - Bounded limited memory variable metric is a quasi-Newton method 239 for nonlinear minimization with bound constraints. It is an extension 240 of TAOLMVM 241 242 Options Database Keys: 243 + -tao_lmm_vectors - number of vectors to use for approximation 244 . -tao_lmm_scale_type - "none","scalar","broyden" 245 . -tao_lmm_limit_type - "none","average","relative","absolute" 246 . -tao_lmm_rescale_type - "none","scalar","gl" 247 . -tao_lmm_limit_mu - mu limiting factor 248 . -tao_lmm_limit_nu - nu limiting factor 249 . -tao_lmm_delta_min - minimum delta value 250 . -tao_lmm_delta_max - maximum delta value 251 . -tao_lmm_broyden_phi - phi factor for Broyden scaling 252 . -tao_lmm_scalar_alpha - alpha factor for scalar scaling 253 . -tao_lmm_rescale_alpha - alpha factor for rescaling diagonal 254 . -tao_lmm_rescale_beta - beta factor for rescaling diagonal 255 . -tao_lmm_scalar_history - amount of history for scalar scaling 256 . -tao_lmm_rescale_history - amount of history for rescaling diagonal 257 - -tao_lmm_eps - rejection tolerance 258 259 Level: beginner 260 M*/ 261 #undef __FUNCT__ 262 #define __FUNCT__ "TaoCreate_BLMVM" 263 PETSC_EXTERN PetscErrorCode TaoCreate_BLMVM(Tao tao) 264 { 265 TAO_BLMVM *blmP; 266 const char *morethuente_type = TAOLINESEARCHMT; 267 PetscErrorCode ierr; 268 269 PetscFunctionBegin; 270 tao->ops->setup = TaoSetup_BLMVM; 271 tao->ops->solve = TaoSolve_BLMVM; 272 tao->ops->view = TaoView_BLMVM; 273 tao->ops->setfromoptions = TaoSetFromOptions_BLMVM; 274 tao->ops->destroy = TaoDestroy_BLMVM; 275 tao->ops->computedual = TaoComputeDual_BLMVM; 276 277 ierr = PetscNewLog(tao,&blmP);CHKERRQ(ierr); 278 tao->data = (void*)blmP; 279 280 /* Override default settings (unless already changed) */ 281 if (!tao->max_it_changed) tao->max_it = 2000; 282 if (!tao->max_funcs_changed) tao->max_funcs = 4000; 283 if (!tao->fatol_changed) tao->fatol = 1.0e-4; 284 if (!tao->frtol_changed) tao->frtol = 1.0e-4; 285 286 ierr = TaoLineSearchCreate(((PetscObject)tao)->comm, &tao->linesearch);CHKERRQ(ierr); 287 ierr = TaoLineSearchSetType(tao->linesearch, morethuente_type);CHKERRQ(ierr); 288 ierr = TaoLineSearchUseTaoRoutines(tao->linesearch,tao);CHKERRQ(ierr); 289 ierr = TaoLineSearchSetOptionsPrefix(tao->linesearch,tao->hdr.prefix);CHKERRQ(ierr); 290 PetscFunctionReturn(0); 291 } 292 293