1 // Ceres Solver - A fast non-linear least squares minimizer
2 // Copyright 2015 Google Inc. All rights reserved.
3 // http://ceres-solver.org/
5 // Redistribution and use in source and binary forms, with or without
6 // modification, are permitted provided that the following conditions are met:
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.
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.
29 // Author: sameeragarwal@google.com (Sameer Agarwal)
31 #include "ceres/levenberg_marquardt_strategy.h"
35 #include "ceres/array_utils.h"
36 #include "ceres/internal/eigen.h"
37 #include "ceres/linear_least_squares_problems.h"
38 #include "ceres/linear_solver.h"
39 #include "ceres/sparse_matrix.h"
40 #include "ceres/trust_region_strategy.h"
41 #include "ceres/types.h"
42 #include "glog/logging.h"
47 LevenbergMarquardtStrategy::LevenbergMarquardtStrategy(
48 const TrustRegionStrategy::Options& options)
49 : linear_solver_(options.linear_solver),
50 radius_(options.initial_radius),
51 max_radius_(options.max_radius),
52 min_diagonal_(options.min_lm_diagonal),
53 max_diagonal_(options.max_lm_diagonal),
54 decrease_factor_(2.0),
55 reuse_diagonal_(false) {
56 CHECK_NOTNULL(linear_solver_);
57 CHECK_GT(min_diagonal_, 0.0);
58 CHECK_LE(min_diagonal_, max_diagonal_);
59 CHECK_GT(max_radius_, 0.0);
62 LevenbergMarquardtStrategy::~LevenbergMarquardtStrategy() {
65 TrustRegionStrategy::Summary LevenbergMarquardtStrategy::ComputeStep(
66 const TrustRegionStrategy::PerSolveOptions& per_solve_options,
67 SparseMatrix* jacobian,
68 const double* residuals,
70 CHECK_NOTNULL(jacobian);
71 CHECK_NOTNULL(residuals);
74 const int num_parameters = jacobian->num_cols();
75 if (!reuse_diagonal_) {
76 if (diagonal_.rows() != num_parameters) {
77 diagonal_.resize(num_parameters, 1);
80 jacobian->SquaredColumnNorm(diagonal_.data());
81 for (int i = 0; i < num_parameters; ++i) {
82 diagonal_[i] = std::min(std::max(diagonal_[i], min_diagonal_),
87 lm_diagonal_ = (diagonal_ / radius_).array().sqrt();
89 LinearSolver::PerSolveOptions solve_options;
90 solve_options.D = lm_diagonal_.data();
91 solve_options.q_tolerance = per_solve_options.eta;
92 // Disable r_tolerance checking. Since we only care about
93 // termination via the q_tolerance. As Nash and Sofer show,
94 // r_tolerance based termination is essentially useless in
95 // Truncated Newton methods.
96 solve_options.r_tolerance = -1.0;
98 // Invalidate the output array lm_step, so that we can detect if
99 // the linear solver generated numerical garbage. This is known
100 // to happen for the DENSE_QR and then DENSE_SCHUR solver when
101 // the Jacobin is severly rank deficient and mu is too small.
102 InvalidateArray(num_parameters, step);
104 // Instead of solving Jx = -r, solve Jy = r.
105 // Then x can be found as x = -y, but the inputs jacobian and residuals
106 // do not need to be modified.
107 LinearSolver::Summary linear_solver_summary =
108 linear_solver_->Solve(jacobian, residuals, solve_options, step);
110 if (linear_solver_summary.termination_type == LINEAR_SOLVER_FATAL_ERROR) {
111 LOG(WARNING) << "Linear solver fatal error: "
112 << linear_solver_summary.message;
113 } else if (linear_solver_summary.termination_type == LINEAR_SOLVER_FAILURE) {
114 LOG(WARNING) << "Linear solver failure. Failed to compute a step: "
115 << linear_solver_summary.message;
116 } else if (!IsArrayValid(num_parameters, step)) {
117 LOG(WARNING) << "Linear solver failure. Failed to compute a finite step.";
118 linear_solver_summary.termination_type = LINEAR_SOLVER_FAILURE;
120 VectorRef(step, num_parameters) *= -1.0;
122 reuse_diagonal_ = true;
124 if (per_solve_options.dump_format_type == CONSOLE ||
125 (per_solve_options.dump_format_type != CONSOLE &&
126 !per_solve_options.dump_filename_base.empty())) {
127 if (!DumpLinearLeastSquaresProblem(per_solve_options.dump_filename_base,
128 per_solve_options.dump_format_type,
134 LOG(ERROR) << "Unable to dump trust region problem."
135 << " Filename base: " << per_solve_options.dump_filename_base;
140 TrustRegionStrategy::Summary summary;
141 summary.residual_norm = linear_solver_summary.residual_norm;
142 summary.num_iterations = linear_solver_summary.num_iterations;
143 summary.termination_type = linear_solver_summary.termination_type;
147 void LevenbergMarquardtStrategy::StepAccepted(double step_quality) {
148 CHECK_GT(step_quality, 0.0);
149 radius_ = radius_ / std::max(1.0 / 3.0,
150 1.0 - pow(2.0 * step_quality - 1.0, 3));
151 radius_ = std::min(max_radius_, radius_);
152 decrease_factor_ = 2.0;
153 reuse_diagonal_ = false;
156 void LevenbergMarquardtStrategy::StepRejected(double step_quality) {
157 radius_ = radius_ / decrease_factor_;
158 decrease_factor_ *= 2.0;
159 reuse_diagonal_ = true;
162 double LevenbergMarquardtStrategy::Radius() const {
166 } // namespace internal