1 // Ceres Solver - A fast non-linear least squares minimizer 2 // Copyright 2010, 2011, 2012 Google Inc. All rights reserved. 3 // http://code.google.com/p/ceres-solver/ 4 // 5 // Redistribution and use in source and binary forms, with or without 6 // modification, are permitted provided that the following conditions are met: 7 // 8 // * Redistributions of source code must retain the above copyright notice, 9 // this list of conditions and the following disclaimer. 10 // * Redistributions in binary form must reproduce the above copyright notice, 11 // this list of conditions and the following disclaimer in the documentation 12 // and/or other materials provided with the distribution. 13 // * Neither the name of Google Inc. nor the names of its contributors may be 14 // used to endorse or promote products derived from this software without 15 // specific prior written permission. 16 // 17 // THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" 18 // AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 19 // IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 20 // ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE 21 // LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 22 // CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 23 // SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 24 // INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 25 // CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 26 // ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 27 // POSSIBILITY OF SUCH DAMAGE. 28 // 29 // Author: sameeragarwal@google.com (Sameer Agarwal) 30 31 #ifndef CERES_INTERNAL_MINIMIZER_H_ 32 #define CERES_INTERNAL_MINIMIZER_H_ 33 34 #include <vector> 35 #include "ceres/solver.h" 36 #include "ceres/iteration_callback.h" 37 38 namespace ceres { 39 namespace internal { 40 41 class Evaluator; 42 class LinearSolver; 43 class SparseMatrix; 44 class TrustRegionStrategy; 45 46 // Interface for non-linear least squares solvers. 47 class Minimizer { 48 public: 49 // Options struct to control the behaviour of the Minimizer. Please 50 // see solver.h for detailed information about the meaning and 51 // default values of each of these parameters. 52 struct Options { OptionsOptions53 Options() { 54 Init(Solver::Options()); 55 } 56 OptionsOptions57 explicit Options(const Solver::Options& options) { 58 Init(options); 59 } 60 InitOptions61 void Init(const Solver::Options& options) { 62 num_threads = options.num_threads; 63 max_num_iterations = options.max_num_iterations; 64 max_solver_time_in_seconds = options.max_solver_time_in_seconds; 65 max_step_solver_retries = 5; 66 gradient_tolerance = options.gradient_tolerance; 67 parameter_tolerance = options.parameter_tolerance; 68 function_tolerance = options.function_tolerance; 69 min_relative_decrease = options.min_relative_decrease; 70 eta = options.eta; 71 jacobi_scaling = options.jacobi_scaling; 72 use_nonmonotonic_steps = options.use_nonmonotonic_steps; 73 max_consecutive_nonmonotonic_steps = 74 options.max_consecutive_nonmonotonic_steps; 75 lsqp_dump_directory = options.lsqp_dump_directory; 76 lsqp_iterations_to_dump = options.lsqp_iterations_to_dump; 77 lsqp_dump_format_type = options.lsqp_dump_format_type; 78 max_num_consecutive_invalid_steps = 79 options.max_num_consecutive_invalid_steps; 80 min_trust_region_radius = options.min_trust_region_radius; 81 evaluator = NULL; 82 trust_region_strategy = NULL; 83 jacobian = NULL; 84 callbacks = options.callbacks; 85 inner_iteration_minimizer = NULL; 86 } 87 88 int max_num_iterations; 89 double max_solver_time_in_seconds; 90 91 // Number of times the linear solver should be retried in case of 92 // numerical failure. The retries are done by exponentially scaling up 93 // mu at each retry. This leads to stronger and stronger 94 // regularization making the linear least squares problem better 95 // conditioned at each retry. 96 int num_threads; 97 int max_step_solver_retries; 98 double gradient_tolerance; 99 double parameter_tolerance; 100 double function_tolerance; 101 double min_relative_decrease; 102 double eta; 103 bool jacobi_scaling; 104 bool use_nonmonotonic_steps; 105 int max_consecutive_nonmonotonic_steps; 106 vector<int> lsqp_iterations_to_dump; 107 DumpFormatType lsqp_dump_format_type; 108 string lsqp_dump_directory; 109 int max_num_consecutive_invalid_steps; 110 int min_trust_region_radius; 111 112 // List of callbacks that are executed by the Minimizer at the end 113 // of each iteration. 114 // 115 // The Options struct does not own these pointers. 116 vector<IterationCallback*> callbacks; 117 118 // Object responsible for evaluating the cost, residuals and 119 // Jacobian matrix. The Options struct does not own this pointer. 120 Evaluator* evaluator; 121 122 // Object responsible for actually computing the trust region 123 // step, and sizing the trust region radius. The Options struct 124 // does not own this pointer. 125 TrustRegionStrategy* trust_region_strategy; 126 127 // Object holding the Jacobian matrix. It is assumed that the 128 // sparsity structure of the matrix has already been initialized 129 // and will remain constant for the life time of the 130 // optimization. The Options struct does not own this pointer. 131 SparseMatrix* jacobian; 132 133 Minimizer* inner_iteration_minimizer; 134 }; 135 ~Minimizer()136 virtual ~Minimizer() {} 137 138 // Note: The minimizer is expected to update the state of the 139 // parameters array every iteration. This is required for the 140 // StateUpdatingCallback to work. 141 virtual void Minimize(const Options& options, 142 double* parameters, 143 Solver::Summary* summary) = 0; 144 }; 145 146 } // namespace internal 147 } // namespace ceres 148 149 #endif // CERES_INTERNAL_MINIMIZER_H_ 150