| Line | Branch | Exec | Source |
|---|---|---|---|
| 1 | /* | ||
| 2 | * Copyright (c) 2000-2022 Inria | ||
| 3 | * All rights reserved. | ||
| 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 the ALICE Project-Team nor the names of its | ||
| 14 | * contributors may be used to endorse or promote products derived from this | ||
| 15 | * software without 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 HOLDER 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 | * Contact: Bruno Levy | ||
| 30 | * | ||
| 31 | * https://www.inria.fr/fr/bruno-levy | ||
| 32 | * | ||
| 33 | * Inria, | ||
| 34 | * Domaine de Voluceau, | ||
| 35 | * 78150 Le Chesnay - Rocquencourt | ||
| 36 | * FRANCE | ||
| 37 | * | ||
| 38 | */ | ||
| 39 | |||
| 40 | #ifdef GEOGRAM_WITH_HLBFGS | ||
| 41 | |||
| 42 | #include <geogram/numerics/lbfgs_optimizers.h> | ||
| 43 | #include <geogram/basic/command_line.h> | ||
| 44 | #include <geogram/basic/argused.h> | ||
| 45 | #include <geogram/third_party/HLBFGS/HLBFGS.h> | ||
| 46 | #include <geogram/bibliography/bibliography.h> | ||
| 47 | |||
| 48 | #include <setjmp.h> | ||
| 49 | #include <iostream> | ||
| 50 | |||
| 51 | namespace GEO { | ||
| 52 | |||
| 53 | /** | ||
| 54 | * \brief Optimizer configuration | ||
| 55 | * \details Manages global variables and callbacks for the | ||
| 56 | * communication between the Optimizer class and the HLBFGS | ||
| 57 | * library. | ||
| 58 | * \internal | ||
| 59 | * The current implementation is actually a bottleneck: optimizer | ||
| 60 | * execution config is stored in static variables which prevents multiple | ||
| 61 | * optimizers to execute in parallel. | ||
| 62 | */ | ||
| 63 | namespace OptimizerConfig { | ||
| 64 | |||
| 65 | static Optimizer::newiteration_callback newiteration_callback_ = nullptr; | ||
| 66 | static Optimizer::funcgrad_callback funcgrad_callback_ = nullptr; | ||
| 67 | static Optimizer::evalhessian_callback evalhessian_callback_ = nullptr; | ||
| 68 | static index_t N_ = 0; | ||
| 69 | |||
| 70 | /** | ||
| 71 | * \brief Initializes Optimizer configuration | ||
| 72 | * \details Sets the problem dimension and the various optimizer | ||
| 73 | * callbacks for the optimizer execution | ||
| 74 | * \param[in] N dimension of the problem | ||
| 75 | * \param[in] funcgrad_callback callback that evaluates | ||
| 76 | * the function to be minimized and its gradient | ||
| 77 | * \param[in] newiteration_callback callback that will be | ||
| 78 | * called at each iteration | ||
| 79 | * \param[in] evalhessian_callback callback that evaluates | ||
| 80 | * the Hessian of function to be minimized (second | ||
| 81 | * order derivatives) | ||
| 82 | */ | ||
| 83 | static void init( | ||
| 84 | index_t N, | ||
| 85 | Optimizer::funcgrad_callback funcgrad_callback, | ||
| 86 | Optimizer::newiteration_callback newiteration_callback, | ||
| 87 | Optimizer::evalhessian_callback evalhessian_callback | ||
| 88 | ) { | ||
| 89 | 10 | N_ = N; | |
| 90 | 10 | funcgrad_callback_ = funcgrad_callback; | |
| 91 | 10 | newiteration_callback_ = newiteration_callback; | |
| 92 | 10 | evalhessian_callback_ = evalhessian_callback; | |
| 93 | } | ||
| 94 | |||
| 95 | /** | ||
| 96 | * \brief HLBFGS callback called at each iteration. | ||
| 97 | * \param[in] iter current iteration | ||
| 98 | * \param[in] call_iter total number of evaluations | ||
| 99 | * \param[in] x value of the parameters at current iteration | ||
| 100 | * \param[in] f value of the function | ||
| 101 | * \param[in] g gradient of the function | ||
| 102 | * \param[in] gnorm norm of the gradient | ||
| 103 | */ | ||
| 104 | 310 | static void HLBFGS_newiteration_callback( | |
| 105 | int iter, int call_iter, double* x, double* f, double* g, | ||
| 106 | double* gnorm | ||
| 107 | ) { | ||
| 108 | GEO::geo_argused(iter); | ||
| 109 | GEO::geo_argused(call_iter); | ||
| 110 | 310 | (* newiteration_callback_)(N_, x, * f, g, * gnorm); | |
| 111 | 310 | } | |
| 112 | |||
| 113 | /** | ||
| 114 | * \brief HLBFGS callback that evaluates the function and its | ||
| 115 | * gradient. | ||
| 116 | * \param[in] N dimension of the problem | ||
| 117 | * \param[in] x value of the parameters at current iteration | ||
| 118 | * \param[in] prev_x value of the parameters at previous iteration | ||
| 119 | * \param[out] f value of the function | ||
| 120 | * \param[out] g gradient of the function | ||
| 121 | */ | ||
| 122 | 335 | static void HLBFGS_funcgrad_callback( | |
| 123 | int N, double* x, double* prev_x, double* f, double* g | ||
| 124 | ) { | ||
| 125 | GEO::geo_argused(prev_x); | ||
| 126 | 335 | (* funcgrad_callback_)((index_t) N, x, * f, g); | |
| 127 | 335 | } | |
| 128 | |||
| 129 | /** | ||
| 130 | * \brief HLBFGS callback that evaluates the function, its gradient | ||
| 131 | * and Hessian. | ||
| 132 | * \param[in] N dimension of the problem | ||
| 133 | * \param[in] x value of the parameters at current iteration | ||
| 134 | * \param[in] prev_x value of the parameters at previous iteration | ||
| 135 | * \param[out] f value of the function | ||
| 136 | * \param[out] g gradient of the function | ||
| 137 | * \param[out] m_hessian Hessian of the function | ||
| 138 | */ | ||
| 139 | ✗ | static void HLBFGS_evalhessian_callback( | |
| 140 | int N, double* x, double* prev_x, double* f, double* g, | ||
| 141 | HESSIAN_MATRIX& m_hessian | ||
| 142 | ) { | ||
| 143 | GEO::geo_argused(prev_x); | ||
| 144 | ✗ | (* evalhessian_callback_)((index_t) N, x, * f, g, m_hessian); | |
| 145 | ✗ | } | |
| 146 | } | ||
| 147 | |||
| 148 | /************************************************************************/ | ||
| 149 | |||
| 150 | 10 | HLBFGSOptimizer::HLBFGSOptimizer() : | |
| 151 | 10 | b_m1qn3_(false), | |
| 152 | 10 | b_cg_(false) { | |
| 153 |
1/2✓ Branch 1 taken 10 times.
✗ Branch 2 not taken.
|
10 | geo_cite("WEB:HLBFGS"); |
| 154 | 10 | } | |
| 155 | |||
| 156 | 40 | HLBFGSOptimizer::~HLBFGSOptimizer() { | |
| 157 | 40 | } | |
| 158 | |||
| 159 | 10 | void HLBFGSOptimizer::optimize(double* x) { | |
| 160 |
1/6✗ Branch 0 not taken.
✓ Branch 1 taken 10 times.
✗ Branch 4 not taken.
✗ Branch 5 not taken.
✗ Branch 7 not taken.
✗ Branch 8 not taken.
|
10 | geo_assert(newiteration_callback_ != nullptr); |
| 161 |
1/6✗ Branch 0 not taken.
✓ Branch 1 taken 10 times.
✗ Branch 4 not taken.
✗ Branch 5 not taken.
✗ Branch 7 not taken.
✗ Branch 8 not taken.
|
10 | geo_assert(funcgrad_callback_ != nullptr); |
| 162 |
1/6✗ Branch 0 not taken.
✓ Branch 1 taken 10 times.
✗ Branch 4 not taken.
✗ Branch 5 not taken.
✗ Branch 7 not taken.
✗ Branch 8 not taken.
|
10 | geo_assert(n_ > 0); |
| 163 |
1/6✗ Branch 0 not taken.
✓ Branch 1 taken 10 times.
✗ Branch 4 not taken.
✗ Branch 5 not taken.
✗ Branch 7 not taken.
✗ Branch 8 not taken.
|
10 | geo_assert(x != nullptr); |
| 164 | |||
| 165 | OptimizerConfig::init( | ||
| 166 | n_, | ||
| 167 | funcgrad_callback_, | ||
| 168 | newiteration_callback_, | ||
| 169 | nullptr | ||
| 170 | ); | ||
| 171 | |||
| 172 | double parameter[20]; | ||
| 173 | int hlbfgs_info[20]; | ||
| 174 | |||
| 175 | // initialize parameters and infos | ||
| 176 | 10 | INIT_HLBFGS(parameter, hlbfgs_info); | |
| 177 |
1/2✓ Branch 0 taken 10 times.
✗ Branch 1 not taken.
|
10 | hlbfgs_info[3] = b_m1qn3_ ? 1 : 0; // determines whether we use m1qn3 |
| 178 | 10 | hlbfgs_info[4] = (int) max_iter_; // max iterations | |
| 179 |
1/2✓ Branch 0 taken 10 times.
✗ Branch 1 not taken.
|
10 | hlbfgs_info[5] = |
| 180 |
2/4✓ Branch 2 taken 10 times.
✗ Branch 3 not taken.
✓ Branch 4 taken 10 times.
✗ Branch 5 not taken.
|
10 | GEO::CmdLine::get_arg_bool("debug") ? 1 : 0; // verbose |
| 181 |
1/2✓ Branch 0 taken 10 times.
✗ Branch 1 not taken.
|
10 | hlbfgs_info[10] = b_cg_ ? 1 : 0; // determines whether we use cg |
| 182 | 10 | parameter[5] = 0; // disabled | |
| 183 | 10 | parameter[6] = epsg_; | |
| 184 | |||
| 185 | 10 | HLBFGS( | |
| 186 | 10 | (int) n_, | |
| 187 | 10 | (int) m_, | |
| 188 | x, | ||
| 189 | OptimizerConfig::HLBFGS_funcgrad_callback, | ||
| 190 | nullptr, | ||
| 191 | HLBFGS_UPDATE_Hessian, | ||
| 192 | OptimizerConfig::HLBFGS_newiteration_callback, | ||
| 193 | parameter, | ||
| 194 | hlbfgs_info | ||
| 195 | ); | ||
| 196 | 10 | } | |
| 197 | |||
| 198 | /************************************************************************/ | ||
| 199 | |||
| 200 | ✗ | HLBFGS_M1QN3Optimizer::HLBFGS_M1QN3Optimizer() { | |
| 201 | set_m1qn3(true); | ||
| 202 | ✗ | } | |
| 203 | |||
| 204 | ✗ | HLBFGS_M1QN3Optimizer::~HLBFGS_M1QN3Optimizer() { | |
| 205 | ✗ | } | |
| 206 | |||
| 207 | /************************************************************************/ | ||
| 208 | |||
| 209 | ✗ | HLBFGS_CGOptimizer::HLBFGS_CGOptimizer() { | |
| 210 | set_cg(true); | ||
| 211 | ✗ | } | |
| 212 | |||
| 213 | ✗ | HLBFGS_CGOptimizer::~HLBFGS_CGOptimizer() { | |
| 214 | ✗ | } | |
| 215 | |||
| 216 | /************************************************************************/ | ||
| 217 | |||
| 218 | ✗ | HLBFGS_HessOptimizer::HLBFGS_HessOptimizer() : | |
| 219 | ✗ | T_(0) { | |
| 220 | ✗ | } | |
| 221 | |||
| 222 | ✗ | HLBFGS_HessOptimizer::~HLBFGS_HessOptimizer() { | |
| 223 | ✗ | } | |
| 224 | |||
| 225 | ✗ | void HLBFGS_HessOptimizer::optimize(double* x) { | |
| 226 | ✗ | geo_assert(newiteration_callback_ != nullptr); | |
| 227 | ✗ | geo_assert(funcgrad_callback_ != nullptr); | |
| 228 | ✗ | geo_assert(evalhessian_callback_ != nullptr); | |
| 229 | ✗ | geo_assert(n_ > 0); | |
| 230 | ✗ | geo_assert(x != nullptr); | |
| 231 | |||
| 232 | OptimizerConfig::init( | ||
| 233 | n_, | ||
| 234 | funcgrad_callback_, | ||
| 235 | newiteration_callback_, | ||
| 236 | evalhessian_callback_ | ||
| 237 | ); | ||
| 238 | |||
| 239 | double parameter[20]; | ||
| 240 | int hlbfgs_info[20]; | ||
| 241 | |||
| 242 | // initialize parameters and infos | ||
| 243 | ✗ | INIT_HLBFGS(parameter, hlbfgs_info); | |
| 244 | ✗ | hlbfgs_info[4] = (int) max_iter_; // max iterations | |
| 245 | ✗ | hlbfgs_info[6] = (int) T_; // update interval of hessian | |
| 246 | ✗ | hlbfgs_info[7] = 1; // 0: without hessian, 1: with accurate hessian | |
| 247 | |||
| 248 | ✗ | HLBFGS( | |
| 249 | ✗ | (int) n_, | |
| 250 | ✗ | (int) m_, | |
| 251 | x, | ||
| 252 | OptimizerConfig::HLBFGS_funcgrad_callback, | ||
| 253 | OptimizerConfig::HLBFGS_evalhessian_callback, | ||
| 254 | HLBFGS_UPDATE_Hessian, | ||
| 255 | OptimizerConfig::HLBFGS_newiteration_callback, | ||
| 256 | parameter, | ||
| 257 | hlbfgs_info | ||
| 258 | ); | ||
| 259 | ✗ | } | |
| 260 | } | ||
| 261 | |||
| 262 | #endif | ||
| 263 |