| 1 | /* ---------------------------------------------------------------------------- |
| 2 | |
| 3 | * GTSAM Copyright 2010, Georgia Tech Research Corporation, |
| 4 | * Atlanta, Georgia 30332-0415 |
| 5 | * All Rights Reserved |
| 6 | * Authors: Frank Dellaert, et al. (see THANKS for the full author list) |
| 7 | |
| 8 | * See LICENSE for the license information |
| 9 | |
| 10 | * -------------------------------------------------------------------------- */ |
| 11 | |
| 12 | /** |
| 13 | * @file testQPSolver.cpp |
| 14 | * @brief Test simple QP solver for a linear inequality constraint |
| 15 | * @date Apr 10, 2014 |
| 16 | * @author Duy-Nguyen Ta |
| 17 | * @author Ivan Dario Jimenez |
| 18 | */ |
| 19 | |
| 20 | #include <gtsam/config.h> |
| 21 | #if GTSAM_USE_BOOST_FEATURES |
| 22 | #include <gtsam_unstable/linear/QPSParser.h> |
| 23 | #endif |
| 24 | |
| 25 | #include <gtsam/base/Testable.h> |
| 26 | #include <gtsam/inference/Symbol.h> |
| 27 | #include <gtsam_unstable/linear/QPSolver.h> |
| 28 | |
| 29 | #include <CppUnitLite/TestHarness.h> |
| 30 | |
| 31 | using namespace std; |
| 32 | using namespace gtsam; |
| 33 | using namespace gtsam::symbol_shorthand; |
| 34 | |
| 35 | static const Vector kOne = Vector::Ones(newSize: 1), kZero = Vector::Zero(size: 1); |
| 36 | |
| 37 | /* ************************************************************************* */ |
| 38 | // Create test graph according to Forst10book_pg171Ex5 |
| 39 | QP createTestCase() { |
| 40 | QP qp; |
| 41 | |
| 42 | // Objective functions x1^2 - x1*x2 + x2^2 - 3*x1 + 5 |
| 43 | // Note the Hessian encodes: |
| 44 | // 0.5*x1'*G11*x1 + x1'*G12*x2 + 0.5*x2'*G22*x2 - x1'*g1 - x2'*g2 + |
| 45 | // 0.5*f |
| 46 | // Hence, we have G11=2, G12 = -1, g1 = +3, G22 = 2, g2 = 0, f = 10 |
| 47 | // TODO: THIS TEST MIGHT BE WRONG : the last parameter might be 5 instead of |
| 48 | // 10 because the form of the equation |
| 49 | // Should be 0.5x'Gx + gx + f : Nocedal 449 |
| 50 | qp.cost.push_back(factor: HessianFactor(X(j: 1), X(j: 2), 2.0 * I_1x1, -I_1x1, 3.0 * I_1x1, |
| 51 | 2.0 * I_1x1, Z_1x1, 10.0)); |
| 52 | |
| 53 | // Inequality constraints |
| 54 | qp.inequalities.add(args: X(j: 1), args: I_1x1, args: X(j: 2), args: I_1x1, args: 2, |
| 55 | args: 0); // x1 + x2 <= 2 --> x1 + x2 -2 <= 0, --> b=2 |
| 56 | qp.inequalities.add(args: X(j: 1), args: -I_1x1, args: 0, args: 1); // -x1 <= 0 |
| 57 | qp.inequalities.add(args: X(j: 2), args: -I_1x1, args: 0, args: 2); // -x2 <= 0 |
| 58 | qp.inequalities.add(args: X(j: 1), args: I_1x1, args: 1.5, args: 3); // x1 <= 3/2 |
| 59 | |
| 60 | return qp; |
| 61 | } |
| 62 | |
| 63 | TEST(QPSolver, TestCase) { |
| 64 | VectorValues values; |
| 65 | double x1 = 5, x2 = 7; |
| 66 | values.insert(j: X(j: 1), value: x1 * I_1x1); |
| 67 | values.insert(j: X(j: 2), value: x2 * I_1x1); |
| 68 | QP qp = createTestCase(); |
| 69 | DOUBLES_EQUAL(29, x1 * x1 - x1 * x2 + x2 * x2 - 3 * x1 + 5, 1e-9); |
| 70 | DOUBLES_EQUAL(29, qp.cost[0]->error(values), 1e-9); |
| 71 | } |
| 72 | |
| 73 | TEST(QPSolver, constraintsAux) { |
| 74 | QP qp = createTestCase(); |
| 75 | |
| 76 | QPSolver solver(qp); |
| 77 | |
| 78 | VectorValues lambdas; |
| 79 | lambdas.insert(j: 0, value: (Vector(1) << -0.5).finished()); |
| 80 | lambdas.insert(j: 1, value: kZero); |
| 81 | lambdas.insert(j: 2, value: (Vector(1) << 0.3).finished()); |
| 82 | lambdas.insert(j: 3, value: (Vector(1) << 0.1).finished()); |
| 83 | int factorIx = solver.identifyLeavingConstraint(workingSet: qp.inequalities, lambdas); |
| 84 | LONGS_EQUAL(2, factorIx); |
| 85 | |
| 86 | VectorValues lambdas2; |
| 87 | lambdas2.insert(j: 0, value: (Vector(1) << -0.5).finished()); |
| 88 | lambdas2.insert(j: 1, value: kZero); |
| 89 | lambdas2.insert(j: 2, value: (Vector(1) << -0.3).finished()); |
| 90 | lambdas2.insert(j: 3, value: (Vector(1) << -0.1).finished()); |
| 91 | int factorIx2 = solver.identifyLeavingConstraint(workingSet: qp.inequalities, lambdas: lambdas2); |
| 92 | LONGS_EQUAL(-1, factorIx2); |
| 93 | } |
| 94 | |
| 95 | /* ************************************************************************* */ |
| 96 | // Create a simple test graph with one equality constraint |
| 97 | QP createEqualityConstrainedTest() { |
| 98 | QP qp; |
| 99 | |
| 100 | // Objective functions x1^2 + x2^2 |
| 101 | // Note the Hessian encodes: |
| 102 | // 0.5*x1'*G11*x1 + x1'*G12*x2 + 0.5*x2'*G22*x2 - x1'*g1 - x2'*g2 + |
| 103 | // 0.5*f |
| 104 | // Hence, we have G11=2, G12 = 0, g1 = 0, G22 = 2, g2 = 0, f = 0 |
| 105 | qp.cost.push_back(factor: HessianFactor(X(j: 1), X(j: 2), 2.0 * I_1x1, Z_1x1, Z_1x1, |
| 106 | 2.0 * I_1x1, Z_1x1, 0.0)); |
| 107 | |
| 108 | // Equality constraints |
| 109 | // x1 + x2 = 1 --> x1 + x2 -1 = 0, hence we negate the b vector |
| 110 | Matrix A1 = I_1x1; |
| 111 | Matrix A2 = I_1x1; |
| 112 | Vector b = -kOne; |
| 113 | qp.equalities.add(args: X(j: 1), args&: A1, args: X(j: 2), args&: A2, args&: b, args: 0); |
| 114 | |
| 115 | return qp; |
| 116 | } |
| 117 | |
| 118 | TEST(QPSolver, dual) { |
| 119 | QP qp = createEqualityConstrainedTest(); |
| 120 | |
| 121 | // Initials values |
| 122 | VectorValues initialValues; |
| 123 | initialValues.insert(j: X(j: 1), value: I_1x1); |
| 124 | initialValues.insert(j: X(j: 2), value: I_1x1); |
| 125 | |
| 126 | QPSolver solver(qp); |
| 127 | |
| 128 | auto dualGraph = solver.buildDualGraph(workingSet: qp.inequalities, delta: initialValues); |
| 129 | VectorValues dual = dualGraph.optimize(); |
| 130 | VectorValues expectedDual; |
| 131 | expectedDual.insert(j: 0, value: (Vector(1) << 2.0).finished()); |
| 132 | CHECK(assert_equal(expectedDual, dual, 1e-10)); |
| 133 | } |
| 134 | |
| 135 | /* ************************************************************************* */ |
| 136 | TEST(QPSolver, indentifyActiveConstraints) { |
| 137 | QP qp = createTestCase(); |
| 138 | QPSolver solver(qp); |
| 139 | |
| 140 | VectorValues currentSolution; |
| 141 | currentSolution.insert(j: X(j: 1), value: Z_1x1); |
| 142 | currentSolution.insert(j: X(j: 2), value: Z_1x1); |
| 143 | |
| 144 | auto workingSet = |
| 145 | solver.identifyActiveConstraints(inequalities: qp.inequalities, initialValues: currentSolution); |
| 146 | |
| 147 | CHECK(!workingSet.at(0)->active()); // inactive |
| 148 | CHECK(workingSet.at(1)->active()); // active |
| 149 | CHECK(workingSet.at(2)->active()); // active |
| 150 | CHECK(!workingSet.at(3)->active()); // inactive |
| 151 | |
| 152 | VectorValues solution = solver.buildWorkingGraph(workingSet).optimize(); |
| 153 | VectorValues expected; |
| 154 | expected.insert(j: X(j: 1), value: kZero); |
| 155 | expected.insert(j: X(j: 2), value: kZero); |
| 156 | CHECK(assert_equal(expected, solution, 1e-100)); |
| 157 | } |
| 158 | |
| 159 | /* ************************************************************************* */ |
| 160 | TEST(QPSolver, iterate) { |
| 161 | QP qp = createTestCase(); |
| 162 | QPSolver solver(qp); |
| 163 | |
| 164 | VectorValues currentSolution; |
| 165 | currentSolution.insert(j: X(j: 1), value: Z_1x1); |
| 166 | currentSolution.insert(j: X(j: 2), value: Z_1x1); |
| 167 | |
| 168 | std::vector<VectorValues> expected(4), expectedDuals(4); |
| 169 | expected[0].insert(j: X(j: 1), value: kZero); |
| 170 | expected[0].insert(j: X(j: 2), value: kZero); |
| 171 | expectedDuals[0].insert(j: 1, value: (Vector(1) << 3).finished()); |
| 172 | expectedDuals[0].insert(j: 2, value: kZero); |
| 173 | |
| 174 | expected[1].insert(j: X(j: 1), value: (Vector(1) << 1.5).finished()); |
| 175 | expected[1].insert(j: X(j: 2), value: kZero); |
| 176 | expectedDuals[1].insert(j: 3, value: (Vector(1) << 1.5).finished()); |
| 177 | |
| 178 | expected[2].insert(j: X(j: 1), value: (Vector(1) << 1.5).finished()); |
| 179 | expected[2].insert(j: X(j: 2), value: (Vector(1) << 0.75).finished()); |
| 180 | |
| 181 | expected[3].insert(j: X(j: 1), value: (Vector(1) << 1.5).finished()); |
| 182 | expected[3].insert(j: X(j: 2), value: (Vector(1) << 0.5).finished()); |
| 183 | |
| 184 | auto workingSet = |
| 185 | solver.identifyActiveConstraints(inequalities: qp.inequalities, initialValues: currentSolution); |
| 186 | |
| 187 | QPSolver::State state(currentSolution, VectorValues(), workingSet, false, |
| 188 | 100); |
| 189 | |
| 190 | // int it = 0; |
| 191 | while (!state.converged) { |
| 192 | state = solver.iterate(state); |
| 193 | // These checks will fail because the expected solutions obtained from |
| 194 | // Forst10book do not follow exactly what we implemented from Nocedal06book. |
| 195 | // Specifically, we do not re-identify active constraints and |
| 196 | // do not recompute dual variables after every step!!! |
| 197 | // CHECK(assert_equal(expected[it], state.values, 1e-10)); |
| 198 | // CHECK(assert_equal(expectedDuals[it], state.duals, 1e-10)); |
| 199 | // it++; |
| 200 | } |
| 201 | |
| 202 | CHECK(assert_equal(expected[3], state.values, 1e-10)); |
| 203 | } |
| 204 | |
| 205 | /* ************************************************************************* */ |
| 206 | TEST(QPSolver, optimizeForst10book_pg171Ex5) { |
| 207 | QP qp = createTestCase(); |
| 208 | QPSolver solver(qp); |
| 209 | VectorValues initialValues; |
| 210 | initialValues.insert(j: X(j: 1), value: Z_1x1); |
| 211 | initialValues.insert(j: X(j: 2), value: Z_1x1); |
| 212 | VectorValues solution = solver.optimize(initialValues).first; |
| 213 | VectorValues expected; |
| 214 | expected.insert(j: X(j: 1), value: (Vector(1) << 1.5).finished()); |
| 215 | expected.insert(j: X(j: 2), value: (Vector(1) << 0.5).finished()); |
| 216 | CHECK(assert_equal(expected, solution, 1e-100)); |
| 217 | } |
| 218 | |
| 219 | /* ************************************************************************* */ |
| 220 | #if GTSAM_USE_BOOST_FEATURES |
| 221 | pair<QP, QP> testParser(QPSParser parser) { |
| 222 | QP exampleqp = parser.Parse(); |
| 223 | QP expected; |
| 224 | Key X1(Symbol('X', 1)), X2(Symbol('X', 2)); |
| 225 | // min f(x,y) = 4 + 1.5x -y + 0.58x^2 + 2xy + 2yx + 10y^2 |
| 226 | expected.cost.push_back(factor: HessianFactor(X1, X2, 8.0 * I_1x1, 2.0 * I_1x1, |
| 227 | -1.5 * kOne, 10.0 * I_1x1, 2.0 * kOne, |
| 228 | 8.0)); |
| 229 | |
| 230 | expected.inequalities.add(args&: X1, args: -2.0 * I_1x1, args&: X2, args: -I_1x1, args: -2, |
| 231 | args: 0); // 2x + y >= 2 |
| 232 | expected.inequalities.add(args&: X1, args: -I_1x1, args&: X2, args: 2.0 * I_1x1, args: 6, args: 1); // -x + 2y <= 6 |
| 233 | expected.inequalities.add(args&: X1, args: I_1x1, args: 20, args: 4); // x<= 20 |
| 234 | expected.inequalities.add(args&: X1, args: -I_1x1, args: 0, args: 2); // x >= 0 |
| 235 | expected.inequalities.add(args&: X2, args: -I_1x1, args: 0, args: 3); // y > = 0 |
| 236 | return {expected, exampleqp}; |
| 237 | } |
| 238 | |
| 239 | TEST(QPSolver, ParserSyntaticTest) { |
| 240 | auto result = testParser(parser: QPSParser("QPExample.QPS" )); |
| 241 | CHECK(assert_equal(result.first.cost, result.second.cost, 1e-7)); |
| 242 | CHECK(assert_equal(result.first.inequalities, result.second.inequalities, |
| 243 | 1e-7)); |
| 244 | CHECK(assert_equal(result.first.equalities, result.second.equalities, 1e-7)); |
| 245 | } |
| 246 | |
| 247 | TEST(QPSolver, ParserSemanticTest) { |
| 248 | auto result = testParser(parser: QPSParser("QPExample.QPS" )); |
| 249 | VectorValues expected = QPSolver(result.first).optimize().first; |
| 250 | VectorValues actual = QPSolver(result.second).optimize().first; |
| 251 | CHECK(assert_equal(actual, expected, 1e-7)); |
| 252 | } |
| 253 | |
| 254 | TEST(QPSolver, QPExampleTest) { |
| 255 | QP problem = QPSParser("QPExample.QPS" ).Parse(); |
| 256 | auto solver = QPSolver(problem); |
| 257 | VectorValues actual = solver.optimize().first; |
| 258 | VectorValues expected; |
| 259 | expected.insert(j: Symbol('X', 1), value: 0.7625 * I_1x1); |
| 260 | expected.insert(j: Symbol('X', 2), value: 0.4750 * I_1x1); |
| 261 | double error_expected = problem.cost.error(x: expected); |
| 262 | double error_actual = problem.cost.error(x: actual); |
| 263 | CHECK(assert_equal(expected, actual, 1e-7)) |
| 264 | CHECK(assert_equal(error_expected, error_actual)) |
| 265 | } |
| 266 | |
| 267 | TEST(QPSolver, HS21) { |
| 268 | QP problem = QPSParser("HS21.QPS" ).Parse(); |
| 269 | VectorValues expected; |
| 270 | expected.insert(j: Symbol('X', 1), value: 2.0 * I_1x1); |
| 271 | expected.insert(j: Symbol('X', 2), value: 0.0 * I_1x1); |
| 272 | VectorValues actual = QPSolver(problem).optimize().first; |
| 273 | double error_actual = problem.cost.error(x: actual); |
| 274 | CHECK(assert_equal(-99.9599999, error_actual, 1e-7)) |
| 275 | CHECK(assert_equal(expected, actual)) |
| 276 | } |
| 277 | |
| 278 | TEST(QPSolver, HS35) { |
| 279 | QP problem = QPSParser("HS35.QPS" ).Parse(); |
| 280 | VectorValues actual = QPSolver(problem).optimize().first; |
| 281 | double error_actual = problem.cost.error(x: actual); |
| 282 | CHECK(assert_equal(1.11111111e-01, error_actual, 1e-7)) |
| 283 | } |
| 284 | |
| 285 | TEST(QPSolver, HS35MOD) { |
| 286 | QP problem = QPSParser("HS35MOD.QPS" ).Parse(); |
| 287 | VectorValues actual = QPSolver(problem).optimize().first; |
| 288 | double error_actual = problem.cost.error(x: actual); |
| 289 | CHECK(assert_equal(2.50000001e-01, error_actual, 1e-7)) |
| 290 | } |
| 291 | |
| 292 | TEST(QPSolver, HS51) { |
| 293 | QP problem = QPSParser("HS51.QPS" ).Parse(); |
| 294 | VectorValues actual = QPSolver(problem).optimize().first; |
| 295 | double error_actual = problem.cost.error(x: actual); |
| 296 | CHECK(assert_equal(8.88178420e-16, error_actual, 1e-7)) |
| 297 | } |
| 298 | |
| 299 | TEST(QPSolver, HS52) { |
| 300 | QP problem = QPSParser("HS52.QPS" ).Parse(); |
| 301 | VectorValues actual = QPSolver(problem).optimize().first; |
| 302 | double error_actual = problem.cost.error(x: actual); |
| 303 | CHECK(assert_equal(5.32664756, error_actual, 1e-7)) |
| 304 | } |
| 305 | |
| 306 | TEST(QPSolver, HS268) { // This test needs an extra order of magnitude of |
| 307 | // tolerance than the rest |
| 308 | QP problem = QPSParser("HS268.QPS" ).Parse(); |
| 309 | VectorValues actual = QPSolver(problem).optimize().first; |
| 310 | double error_actual = problem.cost.error(x: actual); |
| 311 | CHECK(assert_equal(5.73107049e-07, error_actual, 1e-6)) |
| 312 | } |
| 313 | |
| 314 | TEST(QPSolver, QPTEST) { // REQUIRES Jacobian Fix |
| 315 | QP problem = QPSParser("QPTEST.QPS" ).Parse(); |
| 316 | VectorValues actual = QPSolver(problem).optimize().first; |
| 317 | double error_actual = problem.cost.error(x: actual); |
| 318 | CHECK(assert_equal(0.437187500e01, error_actual, 1e-7)) |
| 319 | } |
| 320 | #endif |
| 321 | |
| 322 | /* ************************************************************************* */ |
| 323 | // Create Matlab's test graph as in |
| 324 | // http://www.mathworks.com/help/optim/ug/quadprog.html |
| 325 | QP createTestMatlabQPEx() { |
| 326 | QP qp; |
| 327 | |
| 328 | // Objective functions 0.5*x1^2 + x2^2 - x1*x2 - 2*x1 -6*x2 |
| 329 | // Note the Hessian encodes: |
| 330 | // 0.5*x1'*G11*x1 + x1'*G12*x2 + 0.5*x2'*G22*x2 - x1'*g1 - x2'*g2 + |
| 331 | // 0.5*f |
| 332 | // Hence, we have G11=1, G12 = -1, g1 = +2, G22 = 2, g2 = +6, f = 0 |
| 333 | qp.cost.push_back(factor: HessianFactor(X(j: 1), X(j: 2), 1.0 * I_1x1, -I_1x1, 2.0 * I_1x1, |
| 334 | 2.0 * I_1x1, 6 * I_1x1, 1000.0)); |
| 335 | |
| 336 | // Inequality constraints |
| 337 | qp.inequalities.add(args: X(j: 1), args: I_1x1, args: X(j: 2), args: I_1x1, args: 2, args: 0); // x1 + x2 <= 2 |
| 338 | qp.inequalities.add(args: X(j: 1), args: -I_1x1, args: X(j: 2), args: 2 * I_1x1, args: 2, args: 1); //-x1 + 2*x2 <=2 |
| 339 | qp.inequalities.add(args: X(j: 1), args: 2 * I_1x1, args: X(j: 2), args: I_1x1, args: 3, args: 2); // 2*x1 + x2 <=3 |
| 340 | qp.inequalities.add(args: X(j: 1), args: -I_1x1, args: 0, args: 3); // -x1 <= 0 |
| 341 | qp.inequalities.add(args: X(j: 2), args: -I_1x1, args: 0, args: 4); // -x2 <= 0 |
| 342 | |
| 343 | return qp; |
| 344 | } |
| 345 | |
| 346 | ///* ************************************************************************* |
| 347 | ///*/ |
| 348 | TEST(QPSolver, optimizeMatlabEx) { |
| 349 | QP qp = createTestMatlabQPEx(); |
| 350 | QPSolver solver(qp); |
| 351 | VectorValues initialValues; |
| 352 | initialValues.insert(j: X(j: 1), value: Z_1x1); |
| 353 | initialValues.insert(j: X(j: 2), value: Z_1x1); |
| 354 | VectorValues solution = solver.optimize(initialValues).first; |
| 355 | VectorValues expected; |
| 356 | expected.insert(j: X(j: 1), value: (Vector(1) << 2.0 / 3.0).finished()); |
| 357 | expected.insert(j: X(j: 2), value: (Vector(1) << 4.0 / 3.0).finished()); |
| 358 | CHECK(assert_equal(expected, solution, 1e-7)); |
| 359 | } |
| 360 | |
| 361 | ///* ************************************************************************* |
| 362 | ///*/ |
| 363 | TEST(QPSolver, optimizeMatlabExNoinitials) { |
| 364 | QP qp = createTestMatlabQPEx(); |
| 365 | QPSolver solver(qp); |
| 366 | VectorValues solution = solver.optimize().first; |
| 367 | VectorValues expected; |
| 368 | expected.insert(j: X(j: 1), value: (Vector(1) << 2.0 / 3.0).finished()); |
| 369 | expected.insert(j: X(j: 2), value: (Vector(1) << 4.0 / 3.0).finished()); |
| 370 | CHECK(assert_equal(expected, solution, 1e-7)); |
| 371 | } |
| 372 | |
| 373 | /* ************************************************************************* */ |
| 374 | // Create test graph as in Nocedal06book, Ex 16.4, pg. 475 |
| 375 | QP createTestNocedal06bookEx16_4() { |
| 376 | QP qp; |
| 377 | |
| 378 | qp.cost.add(key1: X(j: 1), A1: I_1x1, b: I_1x1); |
| 379 | qp.cost.add(key1: X(j: 2), A1: I_1x1, b: 2.5 * I_1x1); |
| 380 | |
| 381 | // Inequality constraints |
| 382 | qp.inequalities.add(args: X(j: 1), args: -I_1x1, args: X(j: 2), args: 2 * I_1x1, args: 2, args: 0); |
| 383 | qp.inequalities.add(args: X(j: 1), args: I_1x1, args: X(j: 2), args: 2 * I_1x1, args: 6, args: 1); |
| 384 | qp.inequalities.add(args: X(j: 1), args: I_1x1, args: X(j: 2), args: -2 * I_1x1, args: 2, args: 2); |
| 385 | qp.inequalities.add(args: X(j: 1), args: -I_1x1, args: 0.0, args: 3); |
| 386 | qp.inequalities.add(args: X(j: 2), args: -I_1x1, args: 0.0, args: 4); |
| 387 | |
| 388 | return qp; |
| 389 | } |
| 390 | |
| 391 | TEST(QPSolver, optimizeNocedal06bookEx16_4) { |
| 392 | QP qp = createTestNocedal06bookEx16_4(); |
| 393 | QPSolver solver(qp); |
| 394 | VectorValues initialValues; |
| 395 | initialValues.insert(j: X(j: 1), value: (Vector(1) << 2.0).finished()); |
| 396 | initialValues.insert(j: X(j: 2), value: Z_1x1); |
| 397 | |
| 398 | VectorValues solution = solver.optimize(initialValues).first; |
| 399 | VectorValues expected; |
| 400 | expected.insert(j: X(j: 1), value: (Vector(1) << 1.4).finished()); |
| 401 | expected.insert(j: X(j: 2), value: (Vector(1) << 1.7).finished()); |
| 402 | CHECK(assert_equal(expected, solution, 1e-7)); |
| 403 | } |
| 404 | |
| 405 | /* ************************************************************************* */ |
| 406 | TEST(QPSolver, failedSubproblem) { |
| 407 | QP qp; |
| 408 | qp.cost.add(key1: X(j: 1), A1: I_2x2, b: Z_2x1); |
| 409 | qp.cost.push_back(factor: HessianFactor(X(j: 1), Z_2x2, Z_2x1, 100.0)); |
| 410 | qp.inequalities.add(args: X(j: 1), args&: (Matrix(1, 2) << -1.0, 0.0).finished(), args: -1.0, args: 0); |
| 411 | |
| 412 | VectorValues expected; |
| 413 | expected.insert(j: X(j: 1), value: (Vector(2) << 1.0, 0.0).finished()); |
| 414 | |
| 415 | VectorValues initialValues; |
| 416 | initialValues.insert(j: X(j: 1), value: (Vector(2) << 10.0, 100.0).finished()); |
| 417 | |
| 418 | QPSolver solver(qp); |
| 419 | VectorValues solution = solver.optimize(initialValues).first; |
| 420 | |
| 421 | CHECK(assert_equal(expected, solution, 1e-7)); |
| 422 | } |
| 423 | |
| 424 | /* ************************************************************************* */ |
| 425 | TEST(QPSolver, infeasibleInitial) { |
| 426 | QP qp; |
| 427 | qp.cost.add(key1: X(j: 1), A1: I_2x2, b: Vector::Zero(size: 2)); |
| 428 | qp.cost.push_back(factor: HessianFactor(X(j: 1), Z_2x2, Vector::Zero(size: 2), 100.0)); |
| 429 | qp.inequalities.add(args: X(j: 1), args&: (Matrix(1, 2) << -1.0, 0.0).finished(), args: -1.0, args: 0); |
| 430 | |
| 431 | VectorValues expected; |
| 432 | expected.insert(j: X(j: 1), value: (Vector(2) << 1.0, 0.0).finished()); |
| 433 | |
| 434 | VectorValues initialValues; |
| 435 | initialValues.insert(j: X(j: 1), value: (Vector(2) << -10.0, 100.0).finished()); |
| 436 | |
| 437 | QPSolver solver(qp); |
| 438 | VectorValues solution; |
| 439 | CHECK_EXCEPTION(solver.optimize(initialValues), InfeasibleInitialValues); |
| 440 | } |
| 441 | |
| 442 | /* ************************************************************************* */ |
| 443 | int main() { |
| 444 | TestResult tr; |
| 445 | return TestRegistry::runAllTests(result&: tr); |
| 446 | } |
| 447 | /* ************************************************************************* */ |
| 448 | |