bop_lns.cc 24.3 KB
Newer Older
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591
// Copyright 2010-2018 Google LLC
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License.
// You may obtain a copy of the License at
//
//     http://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing, software
// distributed under the License is distributed on an "AS IS" BASIS,
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
// See the License for the specific language governing permissions and
// limitations under the License.

#include "ortools/bop/bop_lns.h"

#include <deque>
#include <string>
#include <vector>

#include "absl/memory/memory.h"
#include "google/protobuf/text_format.h"
#include "ortools/base/cleanup.h"
#include "ortools/base/commandlineflags.h"
#include "ortools/base/stl_util.h"
#include "ortools/glop/lp_solver.h"
#include "ortools/lp_data/lp_print_utils.h"
#include "ortools/sat/boolean_problem.h"
#include "ortools/sat/lp_utils.h"
#include "ortools/sat/sat_solver.h"
#include "ortools/util/bitset.h"

namespace operations_research {
namespace bop {

using ::operations_research::sat::LinearBooleanConstraint;
using ::operations_research::sat::LinearBooleanProblem;
using ::operations_research::glop::ColIndex;
using ::operations_research::glop::DenseRow;
using ::operations_research::glop::LinearProgram;
using ::operations_research::glop::LPSolver;

//------------------------------------------------------------------------------
// BopCompleteLNSOptimizer
//------------------------------------------------------------------------------

namespace {
void UseBopSolutionForSatAssignmentPreference(const BopSolution& solution,
                                              sat::SatSolver* solver) {
  for (int i = 0; i < solution.Size(); ++i) {
    solver->SetAssignmentPreference(
        sat::Literal(sat::BooleanVariable(i), solution.Value(VariableIndex(i))),
        1.0);
  }
}
}  // namespace

BopCompleteLNSOptimizer::BopCompleteLNSOptimizer(
    const std::string& name, const BopConstraintTerms& objective_terms)
    : BopOptimizerBase(name),
      state_update_stamp_(ProblemState::kInitialStampValue),
      objective_terms_(objective_terms) {}

BopCompleteLNSOptimizer::~BopCompleteLNSOptimizer() {}

BopOptimizerBase::Status BopCompleteLNSOptimizer::SynchronizeIfNeeded(
    const ProblemState& problem_state, int num_relaxed_vars) {
  if (state_update_stamp_ == problem_state.update_stamp()) {
    return BopOptimizerBase::CONTINUE;
  }
  state_update_stamp_ = problem_state.update_stamp();

  // Load the current problem to the solver.
  sat_solver_ = absl::make_unique<sat::SatSolver>();
  const BopOptimizerBase::Status status =
      LoadStateProblemToSatSolver(problem_state, sat_solver_.get());
  if (status != BopOptimizerBase::CONTINUE) return status;

  // Add the constraint that forces the solver to look for a solution
  // at a distance <= num_relaxed_vars from the current one. Note that not all
  // the terms appear in this constraint.
  //
  // TODO(user): if the current solution didn't change, there is no need to
  // re-run this optimizer if we already proved UNSAT.
  std::vector<sat::LiteralWithCoeff> cst;
  for (BopConstraintTerm term : objective_terms_) {
    if (problem_state.solution().Value(term.var_id) && term.weight < 0) {
      cst.push_back(sat::LiteralWithCoeff(
          sat::Literal(sat::BooleanVariable(term.var_id.value()), false), 1.0));
    } else if (!problem_state.solution().Value(term.var_id) &&
               term.weight > 0) {
      cst.push_back(sat::LiteralWithCoeff(
          sat::Literal(sat::BooleanVariable(term.var_id.value()), true), 1.0));
    }
  }
  sat_solver_->AddLinearConstraint(
      /*use_lower_bound=*/false, sat::Coefficient(0),
      /*use_upper_bound=*/true, sat::Coefficient(num_relaxed_vars), &cst);

  if (sat_solver_->IsModelUnsat()) return BopOptimizerBase::ABORT;

  // It sounds like a good idea to force the solver to find a similar solution
  // from the current one. On another side, this is already somewhat enforced by
  // the constraint above, so it will need more investigation.
  UseBopSolutionForSatAssignmentPreference(problem_state.solution(),
                                           sat_solver_.get());
  return BopOptimizerBase::CONTINUE;
}

bool BopCompleteLNSOptimizer::ShouldBeRun(
    const ProblemState& problem_state) const {
  return problem_state.solution().IsFeasible();
}

BopOptimizerBase::Status BopCompleteLNSOptimizer::Optimize(
    const BopParameters& parameters, const ProblemState& problem_state,
    LearnedInfo* learned_info, TimeLimit* time_limit) {
  SCOPED_TIME_STAT(&stats_);
  CHECK(learned_info != nullptr);
  CHECK(time_limit != nullptr);
  learned_info->Clear();

  const BopOptimizerBase::Status sync_status =
      SynchronizeIfNeeded(problem_state, parameters.num_relaxed_vars());
  if (sync_status != BopOptimizerBase::CONTINUE) {
    return sync_status;
  }

  CHECK(sat_solver_ != nullptr);
  const double initial_dt = sat_solver_->deterministic_time();
  auto advance_dt = ::absl::MakeCleanup([initial_dt, this, &time_limit]() {
    time_limit->AdvanceDeterministicTime(sat_solver_->deterministic_time() -
                                         initial_dt);
  });

  // Set the parameters for this run.
  // TODO(user): Because of this, we actually loose the perfect continuity
  // between runs, and the restart policy is resetted... Fix this.
  sat::SatParameters sat_params;
  sat_params.set_max_number_of_conflicts(
      parameters.max_number_of_conflicts_in_random_lns());
  sat_params.set_max_time_in_seconds(time_limit->GetTimeLeft());
  sat_params.set_max_deterministic_time(time_limit->GetDeterministicTimeLeft());
  sat_params.set_random_seed(parameters.random_seed());

  sat_solver_->SetParameters(sat_params);
  const sat::SatSolver::Status sat_status = sat_solver_->Solve();
  if (sat_status == sat::SatSolver::FEASIBLE) {
    SatAssignmentToBopSolution(sat_solver_->Assignment(),
                               &learned_info->solution);
    return BopOptimizerBase::SOLUTION_FOUND;
  }
  if (sat_status == sat::SatSolver::LIMIT_REACHED) {
    return BopOptimizerBase::CONTINUE;
  }

  // Because of the "LNS" constraint, we can't deduce anything about the problem
  // in this case.
  return BopOptimizerBase::ABORT;
}

//------------------------------------------------------------------------------
// BopAdaptiveLNSOptimizer
//------------------------------------------------------------------------------

namespace {
// Returns false if the limit is reached while solving the LP.
bool UseLinearRelaxationForSatAssignmentPreference(
    const BopParameters& parameters, const LinearBooleanProblem& problem,
    sat::SatSolver* sat_solver, TimeLimit* time_limit) {
  // TODO(user): Re-use the lp_model and lp_solver or build a model with only
  //              needed constraints and variables.
  glop::LinearProgram lp_model;
  sat::ConvertBooleanProblemToLinearProgram(problem, &lp_model);

  // Set bounds of variables fixed by the sat_solver.
  const sat::Trail& propagation_trail = sat_solver->LiteralTrail();
  for (int trail_index = 0; trail_index < propagation_trail.Index();
       ++trail_index) {
    const sat::Literal fixed_literal = propagation_trail[trail_index];
    const glop::Fractional value = fixed_literal.IsPositive() ? 1.0 : 0.0;
    lp_model.SetVariableBounds(ColIndex(fixed_literal.Variable().value()),
                               value, value);
  }

  glop::LPSolver lp_solver;
  NestedTimeLimit nested_time_limit(time_limit, time_limit->GetTimeLeft(),
                                    parameters.lp_max_deterministic_time());
  const glop::ProblemStatus lp_status =
      lp_solver.SolveWithTimeLimit(lp_model, nested_time_limit.GetTimeLimit());

  if (lp_status != glop::ProblemStatus::OPTIMAL &&
      lp_status != glop::ProblemStatus::PRIMAL_FEASIBLE &&
      lp_status != glop::ProblemStatus::IMPRECISE) {
    // We have no useful information from the LP, we will abort this LNS.
    return false;
  }

  // Set preferences based on the solution of the relaxation.
  for (ColIndex col(0); col < lp_solver.variable_values().size(); ++col) {
    const double value = lp_solver.variable_values()[col];
    sat_solver->SetAssignmentPreference(
        sat::Literal(sat::BooleanVariable(col.value()), round(value) == 1),
        1 - fabs(value - round(value)));
  }
  return true;
}
}  // namespace

// Note(user): We prefer to start with a really low difficulty as this works
// better for large problem, and for small ones, it will be really quickly
// increased anyway. Maybe a better appproach is to start by relaxing something
// like 10 variables instead of having a fixed percentage.
BopAdaptiveLNSOptimizer::BopAdaptiveLNSOptimizer(
    const std::string& name, bool use_lp_to_guide_sat,
    NeighborhoodGenerator* neighborhood_generator,
    sat::SatSolver* sat_propagator)
    : BopOptimizerBase(name),
      use_lp_to_guide_sat_(use_lp_to_guide_sat),
      neighborhood_generator_(neighborhood_generator),
      sat_propagator_(sat_propagator),
      adaptive_difficulty_(0.001) {
  CHECK(sat_propagator != nullptr);
}

BopAdaptiveLNSOptimizer::~BopAdaptiveLNSOptimizer() {}

bool BopAdaptiveLNSOptimizer::ShouldBeRun(
    const ProblemState& problem_state) const {
  return problem_state.solution().IsFeasible();
}

BopOptimizerBase::Status BopAdaptiveLNSOptimizer::Optimize(
    const BopParameters& parameters, const ProblemState& problem_state,
    LearnedInfo* learned_info, TimeLimit* time_limit) {
  SCOPED_TIME_STAT(&stats_);
  CHECK(learned_info != nullptr);
  CHECK(time_limit != nullptr);
  learned_info->Clear();

  // Set-up a sat_propagator_ cleanup task to catch all the exit cases.
  const double initial_dt = sat_propagator_->deterministic_time();
  auto sat_propagator_cleanup =
      ::absl::MakeCleanup([initial_dt, this, &learned_info, &time_limit]() {
        if (!sat_propagator_->IsModelUnsat()) {
          sat_propagator_->SetAssumptionLevel(0);
          sat_propagator_->RestoreSolverToAssumptionLevel();
          ExtractLearnedInfoFromSatSolver(sat_propagator_, learned_info);
        }
        time_limit->AdvanceDeterministicTime(
            sat_propagator_->deterministic_time() - initial_dt);
      });

  // For the SAT conflicts limit of each LNS, we follow a luby sequence times
  // the base number of conflicts (num_conflicts_). Note that the numbers of the
  // Luby sequence are always power of two.
  //
  // We dynamically change the size of the neighborhood depending on the
  // difficulty of the problem. There is one "target" difficulty for each
  // different numbers in the Luby sequence. Note that the initial value is
  // reused from the last run.
  BopParameters local_parameters = parameters;
  int num_tries = 0;  // TODO(user): remove? our limit is 1 by default.
  while (!time_limit->LimitReached() &&
         num_tries < local_parameters.num_random_lns_tries()) {
    // Compute the target problem difficulty and generate the neighborhood.
    adaptive_difficulty_.UpdateLuby();
    const double difficulty = adaptive_difficulty_.GetParameterValue();
    neighborhood_generator_->GenerateNeighborhood(problem_state, difficulty,
                                                  sat_propagator_);

    ++num_tries;
    VLOG(2) << num_tries << "  difficulty:" << difficulty
            << "  luby:" << adaptive_difficulty_.luby_value()
            << "  fixed:" << sat_propagator_->LiteralTrail().Index() << "/"
            << problem_state.original_problem().num_variables();

    // Special case if the difficulty is too high.
    if (!sat_propagator_->IsModelUnsat()) {
      if (sat_propagator_->CurrentDecisionLevel() == 0) {
        VLOG(2) << "Nothing fixed!";
        adaptive_difficulty_.DecreaseParameter();
        continue;
      }
    }

    // Since everything is already set-up, we try the sat_propagator_ with
    // a really low conflict limit. This allow to quickly skip over UNSAT
    // cases without the costly new problem setup.
    if (!sat_propagator_->IsModelUnsat()) {
      sat::SatParameters params;
      params.set_max_number_of_conflicts(
          local_parameters.max_number_of_conflicts_for_quick_check());
      params.set_max_time_in_seconds(time_limit->GetTimeLeft());
      params.set_max_deterministic_time(time_limit->GetDeterministicTimeLeft());
      params.set_random_seed(parameters.random_seed());
      sat_propagator_->SetParameters(params);
      sat_propagator_->SetAssumptionLevel(
          sat_propagator_->CurrentDecisionLevel());

      const sat::SatSolver::Status status = sat_propagator_->Solve();
      if (status == sat::SatSolver::FEASIBLE) {
        adaptive_difficulty_.IncreaseParameter();
        SatAssignmentToBopSolution(sat_propagator_->Assignment(),
                                   &learned_info->solution);
        return BopOptimizerBase::SOLUTION_FOUND;
      } else if (status == sat::SatSolver::ASSUMPTIONS_UNSAT) {
        // Local problem is infeasible.
        adaptive_difficulty_.IncreaseParameter();
        continue;
      }
    }

    // Restore to the assumption level.
    // This is call is important since all the fixed variable in the
    // propagator_ will be used to construct the local problem below.
    // Note that calling RestoreSolverToAssumptionLevel() might actually prove
    // the infeasibility. It is important to check the UNSAT status afterward.
    if (!sat_propagator_->IsModelUnsat()) {
      sat_propagator_->RestoreSolverToAssumptionLevel();
    }

    // Check if the problem is proved UNSAT, by previous the search or the
    // RestoreSolverToAssumptionLevel() call above.
    if (sat_propagator_->IsModelUnsat()) {
      return problem_state.solution().IsFeasible()
                 ? BopOptimizerBase::OPTIMAL_SOLUTION_FOUND
                 : BopOptimizerBase::INFEASIBLE;
    }

    // Construct and Solve the LNS subproblem.
    //
    // Note that we don't use the sat_propagator_ all the way because using a
    // clean solver on a really small problem is usually a lot faster (even we
    // the time to create the subproblem) that running a long solve under
    // assumption (like we did above with a really low conflit limit).
    const int conflict_limit =
        adaptive_difficulty_.luby_value() *
        parameters.max_number_of_conflicts_in_random_lns();

    sat::SatParameters params;
    params.set_max_number_of_conflicts(conflict_limit);
    params.set_max_time_in_seconds(time_limit->GetTimeLeft());
    params.set_max_deterministic_time(time_limit->GetDeterministicTimeLeft());
    params.set_random_seed(parameters.random_seed());

    sat::SatSolver sat_solver;
    sat_solver.SetParameters(params);

    // Starts by adding the unit clauses to fix the variables.
    const LinearBooleanProblem& problem = problem_state.original_problem();
    sat_solver.SetNumVariables(problem.num_variables());
    for (int i = 0; i < sat_propagator_->LiteralTrail().Index(); ++i) {
      CHECK(sat_solver.AddUnitClause(sat_propagator_->LiteralTrail()[i]));
    }

    // Load the rest of the problem. This will automatically create the small
    // local subproblem using the already fixed variable.
    //
    // TODO(user): modify LoadStateProblemToSatSolver() so that we can call it
    // instead and don't need to over constraint the objective below. As a
    // bonus we will also have the learned binary clauses.
    if (!LoadBooleanProblem(problem, &sat_solver)) {
      // The local problem is infeasible.
      adaptive_difficulty_.IncreaseParameter();
      continue;
    }

    if (use_lp_to_guide_sat_) {
      if (!UseLinearRelaxationForSatAssignmentPreference(
              parameters, problem, &sat_solver, time_limit)) {
        return BopOptimizerBase::LIMIT_REACHED;
      }
    } else {
      UseObjectiveForSatAssignmentPreference(problem, &sat_solver);
    }

    if (!AddObjectiveUpperBound(
            problem, sat::Coefficient(problem_state.solution().GetCost()) - 1,
            &sat_solver)) {
      // The local problem is infeasible.
      adaptive_difficulty_.IncreaseParameter();
      continue;
    }

    // Solve the local problem.
    const sat::SatSolver::Status status = sat_solver.Solve();
    time_limit->AdvanceDeterministicTime(sat_solver.deterministic_time());
    if (status == sat::SatSolver::FEASIBLE) {
      // We found a solution! abort now.
      SatAssignmentToBopSolution(sat_solver.Assignment(),
                                 &learned_info->solution);
      return BopOptimizerBase::SOLUTION_FOUND;
    }

    // Adapt the difficulty.
    if (sat_solver.num_failures() < 0.5 * conflict_limit) {
      adaptive_difficulty_.IncreaseParameter();
    } else if (sat_solver.num_failures() > 0.95 * conflict_limit) {
      adaptive_difficulty_.DecreaseParameter();
    }
  }

  return BopOptimizerBase::CONTINUE;
}

//------------------------------------------------------------------------------
// Neighborhood generators.
//------------------------------------------------------------------------------

namespace {

std::vector<sat::Literal> ObjectiveVariablesAssignedToTheirLowCostValue(
    const ProblemState& problem_state,
    const BopConstraintTerms& objective_terms) {
  std::vector<sat::Literal> result;
  DCHECK(problem_state.solution().IsFeasible());
  for (const BopConstraintTerm& term : objective_terms) {
    if (((problem_state.solution().Value(term.var_id) && term.weight < 0) ||
         (!problem_state.solution().Value(term.var_id) && term.weight > 0))) {
      result.push_back(
          sat::Literal(sat::BooleanVariable(term.var_id.value()),
                       problem_state.solution().Value(term.var_id)));
    }
  }
  return result;
}

}  // namespace

void ObjectiveBasedNeighborhood::GenerateNeighborhood(
    const ProblemState& problem_state, double difficulty,
    sat::SatSolver* sat_propagator) {
  // Generate the set of variable we may fix and randomize their order.
  std::vector<sat::Literal> candidates =
      ObjectiveVariablesAssignedToTheirLowCostValue(problem_state,
                                                    objective_terms_);
  std::shuffle(candidates.begin(), candidates.end(), *random_);

  // We will use the sat_propagator to fix some variables as long as the number
  // of propagated variables in the solver is under our target.
  const int num_variables = sat_propagator->NumVariables();
  const int target = round((1.0 - difficulty) * num_variables);

  sat_propagator->Backtrack(0);
  for (const sat::Literal literal : candidates) {
    if (sat_propagator->LiteralTrail().Index() == target) break;
    if (sat_propagator->LiteralTrail().Index() > target) {
      // We prefer to error on the large neighborhood side, so we backtrack the
      // last enqueued literal.
      sat_propagator->Backtrack(
          std::max(0, sat_propagator->CurrentDecisionLevel() - 1));
      break;
    }
    sat_propagator->EnqueueDecisionAndBacktrackOnConflict(literal);
    if (sat_propagator->IsModelUnsat()) return;
  }
}

void ConstraintBasedNeighborhood::GenerateNeighborhood(
    const ProblemState& problem_state, double difficulty,
    sat::SatSolver* sat_propagator) {
  // Randomize the set of constraint
  const LinearBooleanProblem& problem = problem_state.original_problem();
  const int num_constraints = problem.constraints_size();
  std::vector<int> ct_ids(num_constraints, 0);
  for (int ct_id = 0; ct_id < num_constraints; ++ct_id) ct_ids[ct_id] = ct_id;
  std::shuffle(ct_ids.begin(), ct_ids.end(), *random_);

  // Mark that we want to relax all the variables of these constraints as long
  // as the number of relaxed variable is lower than our difficulty target.
  const int num_variables = sat_propagator->NumVariables();
  const int target = round(difficulty * num_variables);
  int num_relaxed = 0;
  std::vector<bool> variable_is_relaxed(problem.num_variables(), false);
  for (int i = 0; i < ct_ids.size(); ++i) {
    if (num_relaxed >= target) break;
    const LinearBooleanConstraint& constraint = problem.constraints(ct_ids[i]);

    // We exclude really large constraints since they are probably note helpful
    // in picking a nice neighborhood.
    if (constraint.literals_size() > 0.7 * num_variables) continue;

    for (int j = 0; j < constraint.literals_size(); ++j) {
      const VariableIndex var_id(constraint.literals(j) - 1);
      if (!variable_is_relaxed[var_id.value()]) {
        ++num_relaxed;
        variable_is_relaxed[var_id.value()] = true;
      }
    }
  }

  // Basic version: simply fix all the "to_fix" variable that are not relaxed.
  //
  // TODO(user): Not fixing anything that propagates a variable in
  // variable_is_relaxed may be better. It is actually a lot better in the
  // RelationGraphBasedNeighborhood. To investigate.
  sat_propagator->Backtrack(0);
  const std::vector<sat::Literal> to_fix =
      ObjectiveVariablesAssignedToTheirLowCostValue(problem_state,
                                                    objective_terms_);
  for (const sat::Literal literal : to_fix) {
    if (variable_is_relaxed[literal.Variable().value()]) continue;
    sat_propagator->EnqueueDecisionAndBacktrackOnConflict(literal);
    if (sat_propagator->IsModelUnsat()) return;
  }
}

RelationGraphBasedNeighborhood::RelationGraphBasedNeighborhood(
    const LinearBooleanProblem& problem, MTRandom* random)
    : random_(random) {
  const int num_variables = problem.num_variables();
  columns_.resize(num_variables);

  // We will ignore constraints that have more variables than this percentage of
  // the total number of variables in this neighborhood computation.
  //
  // TODO(user): Factor this out with the similar factor in
  // ConstraintBasedNeighborhood? also maybe a better approach is to order the
  // constraint, and stop the neighborhood extension without considering all of
  // them.
  const double kSizeThreshold = 0.1;
  for (int i = 0; i < problem.constraints_size(); ++i) {
    const LinearBooleanConstraint& constraint = problem.constraints(i);
    if (constraint.literals_size() > kSizeThreshold * num_variables) continue;
    for (int j = 0; j < constraint.literals_size(); ++j) {
      const sat::Literal literal(constraint.literals(j));
      columns_[VariableIndex(literal.Variable().value())].push_back(
          ConstraintIndex(i));
    }
  }
}

void RelationGraphBasedNeighborhood::GenerateNeighborhood(
    const ProblemState& problem_state, double difficulty,
    sat::SatSolver* sat_propagator) {
  // Simply walk the graph until enough variable are relaxed.
  const int num_variables = sat_propagator->NumVariables();
  const int target = round(difficulty * num_variables);
  int num_relaxed = 1;
  std::vector<bool> variable_is_relaxed(num_variables, false);
  std::deque<int> queue;

  // TODO(user): If one plan to try of lot of different LNS, maybe it will be
  // better to try to bias the distribution of "center" to be as spread as
  // possible.
  queue.push_back(random_->Uniform(num_variables));
  variable_is_relaxed[queue.back()] = true;
  while (!queue.empty() && num_relaxed < target) {
    const int var = queue.front();
    queue.pop_front();
    for (ConstraintIndex ct_index : columns_[VariableIndex(var)]) {
      const LinearBooleanConstraint& constraint =
          problem_state.original_problem().constraints(ct_index.value());
      for (int i = 0; i < constraint.literals_size(); ++i) {
        const sat::Literal literal(constraint.literals(i));
        const int next_var = literal.Variable().value();
        if (!variable_is_relaxed[next_var]) {
          ++num_relaxed;
          variable_is_relaxed[next_var] = true;
          queue.push_back(next_var);
        }
      }
    }
  }

  // Loops over all the variables in order and only fix the ones that don't
  // propagate any relaxed variables.
  DCHECK(problem_state.solution().IsFeasible());
  sat_propagator->Backtrack(0);
  for (sat::BooleanVariable var(0); var < num_variables; ++var) {
    const sat::Literal literal(
        var, problem_state.solution().Value(VariableIndex(var.value())));
    if (variable_is_relaxed[literal.Variable().value()]) continue;
    const int index =
        sat_propagator->EnqueueDecisionAndBacktrackOnConflict(literal);
    if (sat_propagator->CurrentDecisionLevel() > 0) {
      for (int i = index; i < sat_propagator->LiteralTrail().Index(); ++i) {
        if (variable_is_relaxed
                [sat_propagator->LiteralTrail()[i].Variable().value()]) {
          sat_propagator->Backtrack(sat_propagator->CurrentDecisionLevel() - 1);
        }
      }
    }
    if (sat_propagator->IsModelUnsat()) return;
  }
  VLOG(2) << "target:" << target << " relaxed:" << num_relaxed << " actual:"
          << num_variables - sat_propagator->LiteralTrail().Index();
}

}  // namespace bop
}  // namespace operations_research