• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
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 // When an iteration callback is specified, Ceres calls the callback
32 // after each minimizer step (if the minimizer has not converged) and
33 // passes it an IterationSummary object, defined below.
34 
35 #ifndef CERES_PUBLIC_ITERATION_CALLBACK_H_
36 #define CERES_PUBLIC_ITERATION_CALLBACK_H_
37 
38 #include "ceres/types.h"
39 
40 namespace ceres {
41 
42 // This struct describes the state of the optimizer after each
43 // iteration of the minimization.
44 struct IterationSummary {
IterationSummaryIterationSummary45   IterationSummary()
46       : iteration(0),
47         step_is_valid(false),
48         step_is_nonmonotonic(false),
49         step_is_successful(false),
50         cost(0.0),
51         cost_change(0.0),
52         gradient_max_norm(0.0),
53         step_norm(0.0),
54         eta(0.0),
55         linear_solver_iterations(0),
56         iteration_time_in_seconds(0.0),
57         step_solver_time_in_seconds(0.0),
58         cumulative_time_in_seconds(0.0) {}
59 
60   // Current iteration number.
61   int32 iteration;
62 
63   // Step was numerically valid, i.e., all values are finite and the
64   // step reduces the value of the linearized model.
65   //
66   // Note: step_is_valid is false when iteration = 0.
67   bool step_is_valid;
68 
69   // Step did not reduce the value of the objective function
70   // sufficiently, but it was accepted because of the relaxed
71   // acceptance criterion used by the non-monotonic trust region
72   // algorithm.
73   //
74   // Note: step_is_nonmonotonic is false when iteration = 0;
75   bool step_is_nonmonotonic;
76 
77   // Whether or not the minimizer accepted this step or not. If the
78   // ordinary trust region algorithm is used, this means that the
79   // relative reduction in the objective function value was greater
80   // than Solver::Options::min_relative_decrease. However, if the
81   // non-monotonic trust region algorithm is used
82   // (Solver::Options:use_nonmonotonic_steps = true), then even if the
83   // relative decrease is not sufficient, the algorithm may accept the
84   // step and the step is declared successful.
85   //
86   // Note: step_is_successful is false when iteration = 0.
87   bool step_is_successful;
88 
89   // Value of the objective function.
90   double cost;
91 
92   // Change in the value of the objective function in this
93   // iteration. This can be positive or negative.
94   double cost_change;
95 
96   // Infinity norm of the gradient vector.
97   double gradient_max_norm;
98 
99   // 2-norm of the size of the step computed by the optimization
100   // algorithm.
101   double step_norm;
102 
103   // For trust region algorithms, the ratio of the actual change in
104   // cost and the change in the cost of the linearized approximation.
105   double relative_decrease;
106 
107   // Size of the trust region at the end of the current iteration. For
108   // the Levenberg-Marquardt algorithm, the regularization parameter
109   // mu = 1.0 / trust_region_radius.
110   double trust_region_radius;
111 
112   // For the inexact step Levenberg-Marquardt algorithm, this is the
113   // relative accuracy with which the Newton(LM) step is solved. This
114   // number affects only the iterative solvers capable of solving
115   // linear systems inexactly. Factorization-based exact solvers
116   // ignore it.
117   double eta;
118 
119   // Number of iterations taken by the linear solver to solve for the
120   // Newton step.
121   int linear_solver_iterations;
122 
123   // Time (in seconds) spent inside the minimizer loop in the current
124   // iteration.
125   double iteration_time_in_seconds;
126 
127   // Time (in seconds) spent inside the trust region step solver.
128   double step_solver_time_in_seconds;
129 
130   // Time (in seconds) since the user called Solve().
131   double cumulative_time_in_seconds;
132 };
133 
134 // Interface for specifying callbacks that are executed at the end of
135 // each iteration of the Minimizer. The solver uses the return value
136 // of operator() to decide whether to continue solving or to
137 // terminate. The user can return three values.
138 //
139 // SOLVER_ABORT indicates that the callback detected an abnormal
140 // situation. The solver returns without updating the parameter blocks
141 // (unless Solver::Options::update_state_every_iteration is set
142 // true). Solver returns with Solver::Summary::termination_type set to
143 // USER_ABORT.
144 //
145 // SOLVER_TERMINATE_SUCCESSFULLY indicates that there is no need to
146 // optimize anymore (some user specified termination criterion has
147 // been met). Solver returns with Solver::Summary::termination_type
148 // set to USER_SUCCESS.
149 //
150 // SOLVER_CONTINUE indicates that the solver should continue
151 // optimizing.
152 //
153 // For example, the following Callback is used internally by Ceres to
154 // log the progress of the optimization.
155 //
156 // Callback for logging the state of the minimizer to STDERR or STDOUT
157 // depending on the user's preferences and logging level.
158 //
159 //   class LoggingCallback : public IterationCallback {
160 //    public:
161 //     explicit LoggingCallback(bool log_to_stdout)
162 //         : log_to_stdout_(log_to_stdout) {}
163 //
164 //     ~LoggingCallback() {}
165 //
166 //     CallbackReturnType operator()(const IterationSummary& summary) {
167 //       const char* kReportRowFormat =
168 //           "% 4d: f:% 8e d:% 3.2e g:% 3.2e h:% 3.2e "
169 //           "rho:% 3.2e mu:% 3.2e eta:% 3.2e li:% 3d";
170 //       string output = StringPrintf(kReportRowFormat,
171 //                                    summary.iteration,
172 //                                    summary.cost,
173 //                                    summary.cost_change,
174 //                                    summary.gradient_max_norm,
175 //                                    summary.step_norm,
176 //                                    summary.relative_decrease,
177 //                                    summary.trust_region_radius,
178 //                                    summary.eta,
179 //                                    summary.linear_solver_iterations);
180 //       if (log_to_stdout_) {
181 //         cout << output << endl;
182 //       } else {
183 //         VLOG(1) << output;
184 //       }
185 //       return SOLVER_CONTINUE;
186 //     }
187 //
188 //    private:
189 //     const bool log_to_stdout_;
190 //   };
191 //
192 class IterationCallback {
193  public:
~IterationCallback()194   virtual ~IterationCallback() {}
195   virtual CallbackReturnType operator()(const IterationSummary& summary) = 0;
196 };
197 
198 }  // namespace ceres
199 
200 #endif  // CERES_PUBLIC_ITERATION_CALLBACK_H_
201