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 (at) google.com (Sameer Agarwal) 30 // keir (at) google.com (Keir Mierle) 31 32 #ifndef CERES_INTERNAL_EVALUATOR_H_ 33 #define CERES_INTERNAL_EVALUATOR_H_ 34 35 #include <string> 36 #include <vector> 37 #include "ceres/internal/port.h" 38 #include "ceres/types.h" 39 40 namespace ceres { 41 42 struct CRSMatrix; 43 44 namespace internal { 45 46 class Program; 47 class SparseMatrix; 48 49 // The Evaluator interface offers a way to interact with a least squares cost 50 // function that is useful for an optimizer that wants to minimize the least 51 // squares objective. This insulates the optimizer from issues like Jacobian 52 // storage, parameterization, etc. 53 class Evaluator { 54 public: 55 virtual ~Evaluator(); 56 57 struct Options { 58 Options() 59 : num_threads(1), 60 num_eliminate_blocks(-1), 61 linear_solver_type(DENSE_QR) {} 62 63 int num_threads; 64 int num_eliminate_blocks; 65 LinearSolverType linear_solver_type; 66 }; 67 68 static Evaluator* Create(const Options& options, 69 Program* program, 70 string* error); 71 72 73 // This is used for computing the cost, residual and Jacobian for 74 // returning to the user. For actually solving the optimization 75 // problem, the optimization algorithm uses the ProgramEvaluator 76 // objects directly. 77 // 78 // The residual, gradients and jacobian pointers can be NULL, in 79 // which case they will not be evaluated. cost cannot be NULL. 80 // 81 // The parallelism of the evaluator is controlled by num_threads; it 82 // should be at least 1. 83 // 84 // Note: That this function does not take a parameter vector as 85 // input. The parameter blocks are evaluated on the values contained 86 // in the arrays pointed to by their user_state pointers. 87 // 88 // Also worth noting is that this function mutates program by 89 // calling Program::SetParameterOffsetsAndIndex() on it so that an 90 // evaluator object can be constructed. 91 static bool Evaluate(Program* program, 92 int num_threads, 93 double* cost, 94 vector<double>* residuals, 95 vector<double>* gradient, 96 CRSMatrix* jacobian); 97 98 // Build and return a sparse matrix for storing and working with the Jacobian 99 // of the objective function. The jacobian has dimensions 100 // NumEffectiveParameters() by NumParameters(), and is typically extremely 101 // sparse. Since the sparsity pattern of the Jacobian remains constant over 102 // the lifetime of the optimization problem, this method is used to 103 // instantiate a SparseMatrix object with the appropriate sparsity structure 104 // (which can be an expensive operation) and then reused by the optimization 105 // algorithm and the various linear solvers. 106 // 107 // It is expected that the classes implementing this interface will be aware 108 // of their client's requirements for the kind of sparse matrix storage and 109 // layout that is needed for an efficient implementation. For example 110 // CompressedRowOptimizationProblem creates a compressed row representation of 111 // the jacobian for use with CHOLMOD, where as BlockOptimizationProblem 112 // creates a BlockSparseMatrix representation of the jacobian for use in the 113 // Schur complement based methods. 114 virtual SparseMatrix* CreateJacobian() const = 0; 115 116 // Evaluate the cost function for the given state. Returns the cost, 117 // residuals, and jacobian in the corresponding arguments. Both residuals and 118 // jacobian are optional; to avoid computing them, pass NULL. 119 // 120 // If non-NULL, the Jacobian must have a suitable sparsity pattern; only the 121 // values array of the jacobian is modified. 122 // 123 // state is an array of size NumParameters(), cost is a pointer to a single 124 // double, and residuals is an array of doubles of size NumResiduals(). 125 virtual bool Evaluate(const double* state, 126 double* cost, 127 double* residuals, 128 double* gradient, 129 SparseMatrix* jacobian) = 0; 130 131 // Make a change delta (of size NumEffectiveParameters()) to state (of size 132 // NumParameters()) and store the result in state_plus_delta. 133 // 134 // In the case that there are no parameterizations used, this is equivalent to 135 // 136 // state_plus_delta[i] = state[i] + delta[i] ; 137 // 138 // however, the mapping is more complicated in the case of parameterizations 139 // like quaternions. This is the same as the "Plus()" operation in 140 // local_parameterization.h, but operating over the entire state vector for a 141 // problem. 142 virtual bool Plus(const double* state, 143 const double* delta, 144 double* state_plus_delta) const = 0; 145 146 // The number of parameters in the optimization problem. 147 virtual int NumParameters() const = 0; 148 149 // This is the effective number of parameters that the optimizer may adjust. 150 // This applies when there are parameterizations on some of the parameters. 151 virtual int NumEffectiveParameters() const = 0; 152 153 // The number of residuals in the optimization problem. 154 virtual int NumResiduals() const = 0; 155 }; 156 157 } // namespace internal 158 } // namespace ceres 159 160 #endif // CERES_INTERNAL_EVALUATOR_H_ 161
