C++ Reference

C++ Reference: Routing

routing_lp_scheduling.h
Go to the documentation of this file.
1// Copyright 2010-2021 Google LLC
2// Licensed under the Apache License, Version 2.0 (the "License");
3// you may not use this file except in compliance with the License.
4// You may obtain a copy of the License at
5//
6// http://www.apache.org/licenses/LICENSE-2.0
7//
8// Unless required by applicable law or agreed to in writing, software
9// distributed under the License is distributed on an "AS IS" BASIS,
10// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
11// See the License for the specific language governing permissions and
12// limitations under the License.
13
14#ifndef OR_TOOLS_CONSTRAINT_SOLVER_ROUTING_LP_SCHEDULING_H_
15#define OR_TOOLS_CONSTRAINT_SOLVER_ROUTING_LP_SCHEDULING_H_
16
17#include <algorithm>
18#include <cstdint>
19#include <deque>
20#include <functional>
21#include <limits>
22#include <map>
23#include <memory>
24#include <string>
25#include <utility>
26#include <vector>
27
28#include "absl/container/flat_hash_map.h"
29#include "absl/memory/memory.h"
30#include "absl/time/time.h"
31#include "ortools/base/logging.h"
32#include "ortools/base/mathutil.h"
34#include "ortools/constraint_solver/routing_parameters.pb.h"
35#include "ortools/glop/lp_solver.h"
36#include "ortools/glop/parameters.pb.h"
37#include "ortools/lp_data/lp_data.h"
38#include "ortools/lp_data/lp_types.h"
39#include "ortools/sat/cp_model.pb.h"
40#include "ortools/sat/cp_model_solver.h"
41#include "ortools/sat/lp_utils.h"
42#include "ortools/sat/model.h"
43#include "ortools/sat/sat_parameters.pb.h"
44#include "ortools/util/saturated_arithmetic.h"
45#include "ortools/util/sorted_interval_list.h"
46
47namespace operations_research {
48
49// Classes to solve dimension cumul placement (aka scheduling) problems using
50// linear programming.
51
52// Utility class used in the core optimizer to tighten the cumul bounds as much
53// as possible based on the model precedences.
55 public:
57
58 // Tightens the cumul bounds starting from the current cumul var min/max,
59 // and propagating the precedences resulting from the next_accessor, and the
60 // dimension's precedence rules.
61 // Returns false iff the precedences are infeasible with the given routes.
62 // Otherwise, the user can call CumulMin() and CumulMax() to retrieve the new
63 // bounds of an index.
65 const std::function<int64_t(int64_t)>& next_accessor,
66 int64_t cumul_offset);
67
68 int64_t CumulMin(int index) const {
69 return propagated_bounds_[PositiveNode(index)];
70 }
71
72 int64_t CumulMax(int index) const {
73 const int64_t negated_upper_bound = propagated_bounds_[NegativeNode(index)];
74 return negated_upper_bound == std::numeric_limits<int64_t>::min()
75 ? std::numeric_limits<int64_t>::max()
76 : -negated_upper_bound;
77 }
78
79 const RoutingDimension& dimension() const { return dimension_; }
80
81 private:
82 // An arc "tail --offset--> head" represents the relation
83 // tail + offset <= head.
84 // As arcs are stored by tail, we don't store it in the struct.
85 struct ArcInfo {
86 int head;
87 int64_t offset;
88 };
89 static const int kNoParent;
90 static const int kParentToBePropagated;
91
92 // Return the node corresponding to the lower bound of the cumul of index and
93 // -index respectively.
94 int PositiveNode(int index) const { return 2 * index; }
95 int NegativeNode(int index) const { return 2 * index + 1; }
96
97 void AddNodeToQueue(int node) {
98 if (!node_in_queue_[node]) {
99 bf_queue_.push_back(node);
100 node_in_queue_[node] = true;
101 }
102 }
103
104 // Adds the relation first_index + offset <= second_index, by adding arcs
105 // first_index --offset--> second_index and
106 // -second_index --offset--> -first_index.
107 void AddArcs(int first_index, int second_index, int64_t offset);
108
109 bool InitializeArcsAndBounds(
110 const std::function<int64_t(int64_t)>& next_accessor,
111 int64_t cumul_offset);
112
113 bool UpdateCurrentLowerBoundOfNode(int node, int64_t new_lb, int64_t offset);
114
115 bool DisassembleSubtree(int source, int target);
116
117 bool CleanupAndReturnFalse() {
118 // We clean-up node_in_queue_ for future calls, and return false.
119 for (int node_to_cleanup : bf_queue_) {
120 node_in_queue_[node_to_cleanup] = false;
121 }
122 bf_queue_.clear();
123 return false;
124 }
125
126 const RoutingDimension& dimension_;
127 const int64_t num_nodes_;
128
129 // TODO(user): Investigate if all arcs for a given tail can be created
130 // at the same time, in which case outgoing_arcs_ could point to an absl::Span
131 // for each tail index.
132 std::vector<std::vector<ArcInfo>> outgoing_arcs_;
133
134 std::deque<int> bf_queue_;
135 std::vector<bool> node_in_queue_;
136 std::vector<int> tree_parent_node_of_;
137 // After calling PropagateCumulBounds(), for each node index n,
138 // propagated_bounds_[2*n] and -propagated_bounds_[2*n+1] respectively contain
139 // the propagated lower and upper bounds of n's cumul variable.
140 std::vector<int64_t> propagated_bounds_;
141
142 // Vector used in DisassembleSubtree() to avoid memory reallocation.
143 std::vector<int> tmp_dfs_stack_;
144
145 // Used to store the pickup/delivery pairs encountered on the routes.
146 std::vector<std::pair<int64_t, int64_t>>
147 visited_pickup_delivery_indices_for_pair_;
148};
149
151 // An optimal solution was found respecting all constraints.
152 OPTIMAL,
153 // An optimal solution was found, however constraints which were relaxed were
154 // violated.
156 // A solution could not be found.
158};
159
161 public:
163 virtual void Clear() = 0;
164 virtual int CreateNewPositiveVariable() = 0;
165 virtual bool SetVariableBounds(int index, int64_t lower_bound,
166 int64_t upper_bound) = 0;
167 virtual void SetVariableDisjointBounds(int index,
168 const std::vector<int64_t>& starts,
169 const std::vector<int64_t>& ends) = 0;
170 virtual int64_t GetVariableLowerBound(int index) const = 0;
171 virtual int64_t GetVariableUpperBound(int index) const = 0;
172 virtual void SetObjectiveCoefficient(int index, double coefficient) = 0;
173 virtual double GetObjectiveCoefficient(int index) const = 0;
174 virtual void ClearObjective() = 0;
175 virtual int NumVariables() const = 0;
176 virtual int CreateNewConstraint(int64_t lower_bound, int64_t upper_bound) = 0;
177 virtual void SetCoefficient(int ct, int index, double coefficient) = 0;
178 virtual bool IsCPSATSolver() = 0;
179 virtual void AddObjectiveConstraint() = 0;
180 virtual void AddMaximumConstraint(int max_var, std::vector<int> vars) = 0;
181 virtual void AddProductConstraint(int product_var, std::vector<int> vars) = 0;
182 virtual void SetEnforcementLiteral(int ct, int condition) = 0;
183 virtual DimensionSchedulingStatus Solve(absl::Duration duration_limit) = 0;
184 virtual int64_t GetObjectiveValue() const = 0;
185 virtual double GetValue(int index) const = 0;
186 virtual bool SolutionIsInteger() const = 0;
187
188 // Adds a variable with bounds [lower_bound, upper_bound].
189 int AddVariable(int64_t lower_bound, int64_t upper_bound) {
190 CHECK_LE(lower_bound, upper_bound);
191 const int variable = CreateNewPositiveVariable();
192 SetVariableBounds(variable, lower_bound, upper_bound);
193 return variable;
194 }
195 // Adds a linear constraint, enforcing
196 // lower_bound <= sum variable * coeff <= upper_bound,
197 // and returns the identifier of that constraint.
199 int64_t lower_bound, int64_t upper_bound,
200 const std::vector<std::pair<int, double>>& variable_coeffs) {
201 CHECK_LE(lower_bound, upper_bound);
202 const int ct = CreateNewConstraint(lower_bound, upper_bound);
203 for (const auto& variable_coeff : variable_coeffs) {
204 SetCoefficient(ct, variable_coeff.first, variable_coeff.second);
205 }
206 return ct;
207 }
208 // Adds a linear constraint and a 0/1 variable that is true iff
209 // lower_bound <= sum variable * coeff <= upper_bound,
210 // and returns the identifier of that variable.
212 int64_t lower_bound, int64_t upper_bound,
213 const std::vector<std::pair<int, double>>& weighted_variables) {
214 const int reification_ct = AddLinearConstraint(1, 1, {});
215 if (std::numeric_limits<int64_t>::min() < lower_bound) {
216 const int under_lower_bound = AddVariable(0, 1);
217 SetCoefficient(reification_ct, under_lower_bound, 1);
218 const int under_lower_bound_ct =
219 AddLinearConstraint(std::numeric_limits<int64_t>::min(),
220 lower_bound - 1, weighted_variables);
221 SetEnforcementLiteral(under_lower_bound_ct, under_lower_bound);
222 }
223 if (upper_bound < std::numeric_limits<int64_t>::max()) {
224 const int above_upper_bound = AddVariable(0, 1);
225 SetCoefficient(reification_ct, above_upper_bound, 1);
226 const int above_upper_bound_ct = AddLinearConstraint(
227 upper_bound + 1, std::numeric_limits<int64_t>::max(),
228 weighted_variables);
229 SetEnforcementLiteral(above_upper_bound_ct, above_upper_bound);
230 }
231 const int within_bounds = AddVariable(0, 1);
232 SetCoefficient(reification_ct, within_bounds, 1);
233 const int within_bounds_ct =
234 AddLinearConstraint(lower_bound, upper_bound, weighted_variables);
235 SetEnforcementLiteral(within_bounds_ct, within_bounds);
236 return within_bounds;
237 }
238};
239
241 public:
242 RoutingGlopWrapper(bool is_relaxation, const glop::GlopParameters& parameters)
243 : is_relaxation_(is_relaxation) {
244 lp_solver_.SetParameters(parameters);
245 linear_program_.SetMaximizationProblem(false);
246 }
247 void Clear() override {
248 linear_program_.Clear();
249 linear_program_.SetMaximizationProblem(false);
250 allowed_intervals_.clear();
251 }
253 return linear_program_.CreateNewVariable().value();
254 }
255 bool SetVariableBounds(int index, int64_t lower_bound,
256 int64_t upper_bound) override {
257 DCHECK_GE(lower_bound, 0);
258 // When variable upper bounds are greater than this threshold, precision
259 // issues arise in GLOP. In this case we are just going to suppose that
260 // these high bound values are infinite and not set the upper bound.
261 const int64_t kMaxValue = 1e10;
262 const double lp_min = lower_bound;
263 const double lp_max =
264 (upper_bound > kMaxValue) ? glop::kInfinity : upper_bound;
265 if (lp_min <= lp_max) {
266 linear_program_.SetVariableBounds(glop::ColIndex(index), lp_min, lp_max);
267 return true;
268 }
269 // The linear_program would not be feasible, and it cannot handle the
270 // lp_min > lp_max case, so we must detect infeasibility here.
271 return false;
272 }
273 void SetVariableDisjointBounds(int index, const std::vector<int64_t>& starts,
274 const std::vector<int64_t>& ends) override {
275 // TODO(user): Investigate if we can avoid rebuilding the interval list
276 // each time (we could keep a reference to the forbidden interval list in
277 // RoutingDimension but we would need to store cumul offsets and use them
278 // when checking intervals).
279 allowed_intervals_[index] =
280 absl::make_unique<SortedDisjointIntervalList>(starts, ends);
281 }
282 int64_t GetVariableLowerBound(int index) const override {
283 return linear_program_.variable_lower_bounds()[glop::ColIndex(index)];
284 }
285 int64_t GetVariableUpperBound(int index) const override {
286 const double upper_bound =
287 linear_program_.variable_upper_bounds()[glop::ColIndex(index)];
288 DCHECK_GE(upper_bound, 0);
289 return upper_bound == glop::kInfinity ? std::numeric_limits<int64_t>::max()
290 : static_cast<int64_t>(upper_bound);
291 }
292 void SetObjectiveCoefficient(int index, double coefficient) override {
293 linear_program_.SetObjectiveCoefficient(glop::ColIndex(index), coefficient);
294 }
295 double GetObjectiveCoefficient(int index) const override {
296 return linear_program_.objective_coefficients()[glop::ColIndex(index)];
297 }
298 void ClearObjective() override {
299 for (glop::ColIndex i(0); i < linear_program_.num_variables(); ++i) {
300 linear_program_.SetObjectiveCoefficient(i, 0);
301 }
302 }
303 int NumVariables() const override {
304 return linear_program_.num_variables().value();
305 }
306 int CreateNewConstraint(int64_t lower_bound, int64_t upper_bound) override {
307 const glop::RowIndex ct = linear_program_.CreateNewConstraint();
308 linear_program_.SetConstraintBounds(
309 ct,
310 (lower_bound == std::numeric_limits<int64_t>::min()) ? -glop::kInfinity
311 : lower_bound,
312 (upper_bound == std::numeric_limits<int64_t>::max()) ? glop::kInfinity
313 : upper_bound);
314 return ct.value();
315 }
316 void SetCoefficient(int ct, int index, double coefficient) override {
317 linear_program_.SetCoefficient(glop::RowIndex(ct), glop::ColIndex(index),
318 coefficient);
319 }
320 bool IsCPSATSolver() override { return false; }
321 void AddObjectiveConstraint() override {
322 const int ct = CreateNewConstraint(0, GetObjectiveValue());
323 for (int variable = 0; variable < NumVariables(); variable++) {
324 const double coefficient = GetObjectiveCoefficient(variable);
325 if (coefficient != 0) {
326 SetCoefficient(ct, variable, coefficient);
327 }
328 }
329 }
330 void AddMaximumConstraint(int max_var, std::vector<int> vars) override {}
331 void AddProductConstraint(int product_var, std::vector<int> vars) override {}
332 void SetEnforcementLiteral(int ct, int condition) override{};
333 DimensionSchedulingStatus Solve(absl::Duration duration_limit) override {
334 lp_solver_.GetMutableParameters()->set_max_time_in_seconds(
335 absl::ToDoubleSeconds(duration_limit));
336
337 // Because we construct the lp one constraint at a time and we never call
338 // SetCoefficient() on the same variable twice for a constraint, we know
339 // that the columns do not contain duplicates and are already ordered by
340 // constraint so we do not need to call linear_program->CleanUp() which can
341 // be costly. Note that the assumptions are DCHECKed() in the call below.
342 linear_program_.NotifyThatColumnsAreClean();
343 VLOG(2) << linear_program_.Dump();
344 const glop::ProblemStatus status = lp_solver_.Solve(linear_program_);
345 if (status != glop::ProblemStatus::OPTIMAL &&
346 status != glop::ProblemStatus::IMPRECISE) {
348 }
349 if (is_relaxation_) {
351 }
352 for (const auto& allowed_interval : allowed_intervals_) {
353 const double value_double = GetValue(allowed_interval.first);
354 const int64_t value =
355 (value_double >= std::numeric_limits<int64_t>::max())
356 ? std::numeric_limits<int64_t>::max()
357 : MathUtil::FastInt64Round(value_double);
358 const SortedDisjointIntervalList* const interval_list =
359 allowed_interval.second.get();
360 const auto it = interval_list->FirstIntervalGreaterOrEqual(value);
361 if (it == interval_list->end() || value < it->start) {
363 }
364 }
366 }
367 int64_t GetObjectiveValue() const override {
368 return MathUtil::FastInt64Round(lp_solver_.GetObjectiveValue());
369 }
370 double GetValue(int index) const override {
371 return lp_solver_.variable_values()[glop::ColIndex(index)];
372 }
373 bool SolutionIsInteger() const override {
374 return linear_program_.SolutionIsInteger(lp_solver_.variable_values(),
375 /*absolute_tolerance*/ 1e-3);
376 }
377
378 private:
379 const bool is_relaxation_;
380 glop::LinearProgram linear_program_;
381 glop::LPSolver lp_solver_;
382 absl::flat_hash_map<int, std::unique_ptr<SortedDisjointIntervalList>>
383 allowed_intervals_;
384};
385
387 public:
389 parameters_.set_num_search_workers(1);
390 // Keeping presolve but with 0 iterations; as of 11/2019 it is
391 // significantly faster than both full presolve and no presolve.
392 parameters_.set_cp_model_presolve(true);
393 parameters_.set_max_presolve_iterations(0);
394 parameters_.set_catch_sigint_signal(false);
395 parameters_.set_mip_max_bound(1e8);
396 parameters_.set_search_branching(sat::SatParameters::LP_SEARCH);
397 }
399 void Clear() override {
400 model_.Clear();
401 response_.Clear();
402 objective_coefficients_.clear();
403 }
405 const int index = model_.variables_size();
406 sat::IntegerVariableProto* const variable = model_.add_variables();
407 variable->add_domain(0);
408 variable->add_domain(static_cast<int64_t>(parameters_.mip_max_bound()));
409 return index;
410 }
411 bool SetVariableBounds(int index, int64_t lower_bound,
412 int64_t upper_bound) override {
413 DCHECK_GE(lower_bound, 0);
414 const int64_t capped_upper_bound =
415 std::min<int64_t>(upper_bound, parameters_.mip_max_bound());
416 if (lower_bound > capped_upper_bound) return false;
417 sat::IntegerVariableProto* const variable = model_.mutable_variables(index);
418 variable->set_domain(0, lower_bound);
419 variable->set_domain(1, capped_upper_bound);
420 return true;
421 }
422 void SetVariableDisjointBounds(int index, const std::vector<int64_t>& starts,
423 const std::vector<int64_t>& ends) override {
424 DCHECK_EQ(starts.size(), ends.size());
425 const int ct = CreateNewConstraint(1, 1);
426 for (int i = 0; i < starts.size(); ++i) {
427 const int variable = CreateNewPositiveVariable();
428 SetVariableBounds(variable, 0, 1);
429 SetCoefficient(ct, variable, 1);
430 const int window_ct = CreateNewConstraint(starts[i], ends[i]);
431 SetCoefficient(window_ct, index, 1);
432 model_.mutable_constraints(window_ct)->add_enforcement_literal(variable);
433 }
434 }
435 int64_t GetVariableLowerBound(int index) const override {
436 return model_.variables(index).domain(0);
437 }
438 int64_t GetVariableUpperBound(int index) const override {
439 const auto& domain = model_.variables(index).domain();
440 return domain[domain.size() - 1];
441 }
442 void SetObjectiveCoefficient(int index, double coefficient) override {
443 if (index >= objective_coefficients_.size()) {
444 objective_coefficients_.resize(index + 1, 0);
445 }
446 objective_coefficients_[index] = coefficient;
447 sat::FloatObjectiveProto* const objective =
448 model_.mutable_floating_point_objective();
449 objective->add_vars(index);
450 objective->add_coeffs(coefficient);
451 }
452 double GetObjectiveCoefficient(int index) const override {
453 return (index < objective_coefficients_.size())
454 ? objective_coefficients_[index]
455 : 0;
456 }
457 void ClearObjective() override {
458 model_.mutable_floating_point_objective()->Clear();
459 }
460 int NumVariables() const override { return model_.variables_size(); }
461 int CreateNewConstraint(int64_t lower_bound, int64_t upper_bound) override {
462 sat::LinearConstraintProto* const ct =
463 model_.add_constraints()->mutable_linear();
464 ct->add_domain(lower_bound);
465 ct->add_domain(upper_bound);
466 return model_.constraints_size() - 1;
467 }
468 void SetCoefficient(int ct_index, int index, double coefficient) override {
469 sat::LinearConstraintProto* const ct =
470 model_.mutable_constraints(ct_index)->mutable_linear();
471 ct->add_vars(index);
472 const int64_t integer_coefficient = coefficient;
473 ct->add_coeffs(integer_coefficient);
474 }
475 bool IsCPSATSolver() override { return true; }
476 void AddObjectiveConstraint() override {
477 // Scale the objective to get its integer representation.
478 const sat::FloatObjectiveProto& fp_objective =
479 model_.floating_point_objective();
480 std::vector<std::pair<int, double>> terms;
481 for (int i = 0; i < fp_objective.vars_size(); ++i) {
482 terms.push_back({fp_objective.vars(i), fp_objective.coeffs(i)});
483 }
484 SolverLogger logger;
485 const bool status =
486 sat::ScaleAndSetObjective(parameters_, terms, fp_objective.offset(),
487 fp_objective.maximize(), &model_, &logger);
488 DCHECK(status);
489 const sat::CpObjectiveProto& objective = model_.objective();
490 int64_t activity = 0;
491 for (int i = 0; i < objective.vars_size(); ++i) {
492 activity += response_.solution(objective.vars(i)) * objective.coeffs(i);
493 }
494 const int ct = CreateNewConstraint(0, activity);
495 for (int i = 0; i < objective.vars_size(); ++i) {
496 SetCoefficient(ct, objective.vars(i), objective.coeffs(i));
497 }
498 model_.clear_objective();
499 }
500 void AddMaximumConstraint(int max_var, std::vector<int> vars) override {
501 sat::LinearArgumentProto* const ct =
502 model_.add_constraints()->mutable_lin_max();
503 ct->mutable_target()->add_vars(max_var);
504 ct->mutable_target()->add_coeffs(1);
505 for (const int var : vars) {
506 sat::LinearExpressionProto* const expr = ct->add_exprs();
507 expr->add_vars(var);
508 expr->add_coeffs(1);
509 }
510 }
511 void AddProductConstraint(int product_var, std::vector<int> vars) override {
512 sat::LinearArgumentProto* const ct =
513 model_.add_constraints()->mutable_int_prod();
514 ct->mutable_target()->add_vars(product_var);
515 ct->mutable_target()->add_coeffs(1);
516 for (const int var : vars) {
517 sat::LinearExpressionProto* expr = ct->add_exprs();
518 expr->add_vars(var);
519 expr->add_coeffs(1);
520 }
521 }
522 void SetEnforcementLiteral(int ct, int condition) override {
523 DCHECK_LT(ct, model_.constraints_size());
524 model_.mutable_constraints(ct)->add_enforcement_literal(condition);
525 }
526 DimensionSchedulingStatus Solve(absl::Duration duration_limit) override {
527 parameters_.set_max_time_in_seconds(absl::ToDoubleSeconds(duration_limit));
528 VLOG(2) << model_.DebugString();
529 if (hint_.vars_size() == model_.variables_size()) {
530 *model_.mutable_solution_hint() = hint_;
531 }
532 sat::Model model;
533 model.Add(sat::NewSatParameters(parameters_));
534 response_ = sat::SolveCpModel(model_, &model);
535 VLOG(2) << response_.DebugString();
536 if (response_.status() == sat::CpSolverStatus::OPTIMAL ||
537 (response_.status() == sat::CpSolverStatus::FEASIBLE &&
538 !model_.has_floating_point_objective())) {
539 hint_.Clear();
540 for (int i = 0; i < response_.solution_size(); ++i) {
541 hint_.add_vars(i);
542 hint_.add_values(response_.solution(i));
543 }
545 }
547 }
548 int64_t GetObjectiveValue() const override {
549 return MathUtil::FastInt64Round(response_.objective_value());
550 }
551 double GetValue(int index) const override {
552 return response_.solution(index);
553 }
554 bool SolutionIsInteger() const override { return true; }
555
556 private:
557 sat::CpModelProto model_;
558 sat::CpSolverResponse response_;
559 sat::SatParameters parameters_;
560 std::vector<double> objective_coefficients_;
561 sat::PartialVariableAssignment hint_;
562};
563
564// Utility class used in Local/GlobalDimensionCumulOptimizer to set the linear
565// solver constraints and solve the problem.
567 public:
569 bool use_precedence_propagator);
570
571 // In the OptimizeSingleRoute() and Optimize() methods, if both "cumul_values"
572 // and "cost" parameters are null, we don't optimize the cost and stop at the
573 // first feasible solution in the linear solver (since in this case only
574 // feasibility is of interest).
576 int vehicle, const std::function<int64_t(int64_t)>& next_accessor,
577 RoutingLinearSolverWrapper* solver, std::vector<int64_t>* cumul_values,
578 std::vector<int64_t>* break_values, int64_t* cost, int64_t* transit_cost,
579 bool clear_lp = true);
580
581 std::vector<DimensionSchedulingStatus> OptimizeSingleRouteWithResources(
582 int vehicle, const std::function<int64_t(int64_t)>& next_accessor,
583 const std::vector<RoutingModel::ResourceGroup::Resource>& resources,
584 const std::vector<int>& resource_indices, bool optimize_vehicle_costs,
586 std::vector<int64_t>* costs_without_transits,
587 std::vector<std::vector<int64_t>>* cumul_values,
588 std::vector<std::vector<int64_t>>* break_values, bool clear_lp = true);
589
591 const std::function<int64_t(int64_t)>& next_accessor,
592 RoutingLinearSolverWrapper* solver, std::vector<int64_t>* cumul_values,
593 std::vector<int64_t>* break_values,
594 std::vector<std::vector<int>>* resource_indices_per_group, int64_t* cost,
595 int64_t* transit_cost, bool clear_lp = true);
596
598 const std::function<int64_t(int64_t)>& next_accessor,
599 RoutingLinearSolverWrapper* solver, std::vector<int64_t>* cumul_values,
600 std::vector<int64_t>* break_values,
601 std::vector<std::vector<int>>* resource_indices_per_group);
602
604 int vehicle, const std::function<int64_t(int64_t)>& next_accessor,
606 RoutingLinearSolverWrapper* solver, std::vector<int64_t>* cumul_values,
607 std::vector<int64_t>* break_values);
608
609 const RoutingDimension* dimension() const { return dimension_; }
610
611 private:
612 // Initializes the containers and given solver. Must be called prior to
613 // setting any contraints and solving.
614 void InitOptimizer(RoutingLinearSolverWrapper* solver);
615
616 // Computes the minimum/maximum of cumuls for nodes on "route", and sets them
617 // in current_route_[min|max]_cumuls_ respectively.
618 // If the propagator_ is not null, uses the bounds tightened by the
619 // propagator.
620 // Otherwise, the bounds are computed by going over the nodes on the route
621 // using the CP bounds, and the fixed transits are used to tighten them.
622 bool ComputeRouteCumulBounds(const std::vector<int64_t>& route,
623 const std::vector<int64_t>& fixed_transits,
624 int64_t cumul_offset);
625
626 // Sets the constraints for all nodes on "vehicle"'s route according to
627 // "next_accessor". If optimize_costs is true, also sets the objective
628 // coefficients for the LP.
629 // Returns false if some infeasibility was detected, true otherwise.
630 bool SetRouteCumulConstraints(
631 int vehicle, const std::function<int64_t(int64_t)>& next_accessor,
632 int64_t cumul_offset, bool optimize_costs,
633 RoutingLinearSolverWrapper* solver, int64_t* route_transit_cost,
634 int64_t* route_cost_offset);
635
636 // Sets the global constraints on the dimension, and adds global objective
637 // cost coefficients if optimize_costs is true.
638 // NOTE: When called, the call to this function MUST come after
639 // SetRouteCumulConstraints() has been called on all routes, so that
640 // index_to_cumul_variable_ and min_start/max_end_cumul_ are correctly
641 // initialized.
642 // Returns false if some infeasibility was detected, true otherwise.
643 bool SetGlobalConstraints(
644 const std::function<int64_t(int64_t)>& next_accessor,
645 int64_t cumul_offset, bool optimize_costs,
647
648 void SetValuesFromLP(const std::vector<int>& lp_variables, int64_t offset,
650 std::vector<int64_t>* lp_values) const;
651
652 void SetResourceIndices(
654 std::vector<std::vector<int>>* resource_indices_per_group) const;
655
656 // This function packs the routes of the given vehicles while keeping the cost
657 // of the LP lower than its current (supposed optimal) objective value.
658 // It does so by setting the current objective variables' coefficient to 0 and
659 // setting the coefficient of the route ends to 1, to first minimize the route
660 // ends' cumuls, and then maximizes the starts' cumuls without increasing the
661 // ends.
662 DimensionSchedulingStatus PackRoutes(std::vector<int> vehicles,
664
665 std::unique_ptr<CumulBoundsPropagator> propagator_;
666 std::vector<int64_t> current_route_min_cumuls_;
667 std::vector<int64_t> current_route_max_cumuls_;
668 const RoutingDimension* const dimension_;
669 // Scheduler variables for current route cumuls and for all nodes cumuls.
670 std::vector<int> current_route_cumul_variables_;
671 std::vector<int> index_to_cumul_variable_;
672 // Scheduler variables for current route breaks and all vehicle breaks.
673 // There are two variables for each break: start and end.
674 // current_route_break_variables_ has variables corresponding to
675 // break[0] start, break[0] end, break[1] start, break[1] end, etc.
676 std::vector<int> current_route_break_variables_;
677 // Vector all_break_variables contains the break variables of all vehicles,
678 // in the same format as current_route_break_variables.
679 // It is the concatenation of break variables of vehicles in [0, #vehicles).
680 std::vector<int> all_break_variables_;
681 // Allows to retrieve break variables of a given vehicle: those go from
682 // all_break_variables_[vehicle_to_all_break_variables_offset_[vehicle]] to
683 // all_break_variables[vehicle_to_all_break_variables_offset_[vehicle+1]-1].
684 std::vector<int> vehicle_to_all_break_variables_offset_;
685 // The following vector contains indices of resource-to-vehicle assignment
686 // variables. For every resource group, stores indices of
687 // num_resources*num_vehicles boolean variables indicating whether resource #r
688 // is assigned to vehicle #v.
689 std::vector<std::vector<int>>
690 resource_group_to_resource_to_vehicle_assignment_variables_;
691
692 int max_end_cumul_;
693 int min_start_cumul_;
694 std::vector<std::pair<int64_t, int64_t>>
695 visited_pickup_delivery_indices_for_pair_;
696};
697
698// Class used to compute optimal values for dimension cumuls of routes,
699// minimizing cumul soft lower and upper bound costs, and vehicle span costs of
700// a route.
701// In its methods, next_accessor is a callback returning the next node of a
702// given node on a route.
704 public:
707 RoutingSearchParameters::SchedulingSolver solver_type);
708
709 // If feasible, computes the optimal cost of the route performed by a vehicle,
710 // minimizing cumul soft lower and upper bound costs and vehicle span costs,
711 // and stores it in "optimal_cost" (if not null).
712 // Returns true iff the route respects all constraints.
714 int vehicle, const std::function<int64_t(int64_t)>& next_accessor,
715 int64_t* optimal_cost);
716
717 // Same as ComputeRouteCumulCost, but the cost computed does not contain
718 // the part of the vehicle span cost due to fixed transits.
720 int vehicle, const std::function<int64_t(int64_t)>& next_accessor,
721 int64_t* optimal_cost_without_transits);
722
723 std::vector<DimensionSchedulingStatus>
725 int vehicle, const std::function<int64_t(int64_t)>& next_accessor,
726 const std::vector<RoutingModel::ResourceGroup::Resource>& resources,
727 const std::vector<int>& resource_indices, bool optimize_vehicle_costs,
728 std::vector<int64_t>* optimal_costs_without_transits,
729 std::vector<std::vector<int64_t>>* optimal_cumuls,
730 std::vector<std::vector<int64_t>>* optimal_breaks);
731
732 // If feasible, computes the optimal values for cumul and break variables
733 // of the route performed by a vehicle, minimizing cumul soft lower, upper
734 // bound costs and vehicle span costs, stores them in "optimal_cumuls"
735 // (if not null), and optimal_breaks, and returns true.
736 // Returns false if the route is not feasible.
738 int vehicle, const std::function<int64_t(int64_t)>& next_accessor,
739 std::vector<int64_t>* optimal_cumuls,
740 std::vector<int64_t>* optimal_breaks);
741
742 // Similar to ComputeRouteCumuls, but also tries to pack the cumul values on
743 // the route, such that the cost remains the same, the cumul of route end is
744 // minimized, and then the cumul of the start of the route is maximized.
745 // If 'resource' is non-null, the packed route must also respect its start/end
746 // time window.
748 int vehicle, const std::function<int64_t(int64_t)>& next_accessor,
750 std::vector<int64_t>* packed_cumuls, std::vector<int64_t>* packed_breaks);
751
753 return optimizer_core_.dimension();
754 }
755
756 private:
757 std::vector<std::unique_ptr<RoutingLinearSolverWrapper>> solver_;
758 DimensionCumulOptimizerCore optimizer_core_;
759};
760
762 public:
765 RoutingSearchParameters::SchedulingSolver solver_type);
766 // If feasible, computes the optimal cost of the entire model with regards to
767 // the optimizer_core_'s dimension costs, minimizing cumul soft lower/upper
768 // bound costs and vehicle/global span costs, and stores it in "optimal_cost"
769 // (if not null).
770 // Returns true iff all the constraints can be respected.
772 const std::function<int64_t(int64_t)>& next_accessor,
773 int64_t* optimal_cost_without_transits);
774 // If feasible, computes the optimal values for cumul, break and resource
775 // variables, minimizing cumul soft lower/upper bound costs and vehicle/global
776 // span costs, stores them in "optimal_cumuls" (if not null), "optimal_breaks"
777 // and "optimal_resource_indices_per_group", and returns true.
778 // Returns false if the routes are not feasible.
780 const std::function<int64_t(int64_t)>& next_accessor,
781 std::vector<int64_t>* optimal_cumuls,
782 std::vector<int64_t>* optimal_breaks,
783 std::vector<std::vector<int>>* optimal_resource_indices_per_group);
784
785 // Similar to ComputeCumuls, but also tries to pack the cumul values on all
786 // routes, such that the cost remains the same, the cumuls of route ends are
787 // minimized, and then the cumuls of the starts of the routes are maximized.
789 const std::function<int64_t(int64_t)>& next_accessor,
790 std::vector<int64_t>* packed_cumuls, std::vector<int64_t>* packed_breaks,
791 std::vector<std::vector<int>>* resource_indices_per_group);
792
794 return optimizer_core_.dimension();
795 }
796
797 private:
798 std::unique_ptr<RoutingLinearSolverWrapper> solver_;
799 DimensionCumulOptimizerCore optimizer_core_;
800};
801
803 public:
806 LocalDimensionCumulOptimizer* mp_optimizer);
807
808 // Returns the cost resulting from the min-cost assignment of resources to
809 // (used) vehicles, or -1 if the assignment is impossible.
810 // For each vehicle v and resource r, the cost of assigning r to v is equal to
811 // - primary_vehicle_to_resource_assignment_costs[v][r] if
812 // primary_vehicle_to_resource_assignment_costs[v] is not empty,
813 // - secondary_vehicle_to_resource_assignment_costs[v][r] otherwise
814 // (secondary_vehicle_to_resource_assignment_costs[v] can never be empty).
815 // If non-null, 'resource_indices' contains the index of the resource assigned
816 // to each vehicle.
818 const std::vector<std::vector<int64_t>>&
819 primary_vehicle_to_resource_assignment_costs,
820 const std::vector<std::vector<int64_t>>&
821 secondary_vehicle_to_resource_assignment_costs,
822 const std::function<bool(int)>& use_primary_for_vehicle,
823 std::vector<int>* resource_indices) const;
824
825 // Computes the vehicle-to-resource assignment costs for the given vehicle to
826 // all resources in the group, and sets these costs in assignment_costs (if
827 // non-null).
828 // optimize_vehicle_costs indicates if the costs should be optimized or if
829 // we merely care about feasibility (cost of 0) and infeasibility (cost of -1)
830 // of the assignments.
831 // The cumul and break values corresponding to the assignment of each resource
832 // are also set in cumul_values and break_values, if non-null.
834 int v, const std::function<int64_t(int64_t)>& next_accessor,
835 bool optimize_vehicle_costs, std::vector<int64_t>* assignment_costs,
836 std::vector<std::vector<int64_t>>* cumul_values,
837 std::vector<std::vector<int64_t>>* break_values);
838
839 const RoutingDimension* const dimension() const {
840 return optimizer_.dimension();
841 }
842
843 private:
845 LocalDimensionCumulOptimizer& mp_optimizer_;
846 const RoutingModel& model_;
847 const RoutingModel::ResourceGroup& resource_group_;
848};
849
850} // namespace operations_research
851
852#endif // OR_TOOLS_CONSTRAINT_SOLVER_ROUTING_LP_SCHEDULING_H_
const RoutingDimension & dimension() const
bool PropagateCumulBounds(const std::function< int64_t(int64_t)> &next_accessor, int64_t cumul_offset)
CumulBoundsPropagator(const RoutingDimension *dimension)
std::vector< DimensionSchedulingStatus > OptimizeSingleRouteWithResources(int vehicle, const std::function< int64_t(int64_t)> &next_accessor, const std::vector< RoutingModel::ResourceGroup::Resource > &resources, const std::vector< int > &resource_indices, bool optimize_vehicle_costs, RoutingLinearSolverWrapper *solver, std::vector< int64_t > *costs_without_transits, std::vector< std::vector< int64_t > > *cumul_values, std::vector< std::vector< int64_t > > *break_values, bool clear_lp=true)
DimensionSchedulingStatus Optimize(const std::function< int64_t(int64_t)> &next_accessor, RoutingLinearSolverWrapper *solver, std::vector< int64_t > *cumul_values, std::vector< int64_t > *break_values, std::vector< std::vector< int > > *resource_indices_per_group, int64_t *cost, int64_t *transit_cost, bool clear_lp=true)
DimensionCumulOptimizerCore(const RoutingDimension *dimension, bool use_precedence_propagator)
DimensionSchedulingStatus OptimizeAndPack(const std::function< int64_t(int64_t)> &next_accessor, RoutingLinearSolverWrapper *solver, std::vector< int64_t > *cumul_values, std::vector< int64_t > *break_values, std::vector< std::vector< int > > *resource_indices_per_group)
DimensionSchedulingStatus OptimizeSingleRoute(int vehicle, const std::function< int64_t(int64_t)> &next_accessor, RoutingLinearSolverWrapper *solver, std::vector< int64_t > *cumul_values, std::vector< int64_t > *break_values, int64_t *cost, int64_t *transit_cost, bool clear_lp=true)
DimensionSchedulingStatus OptimizeAndPackSingleRoute(int vehicle, const std::function< int64_t(int64_t)> &next_accessor, const RoutingModel::ResourceGroup::Resource *resource, RoutingLinearSolverWrapper *solver, std::vector< int64_t > *cumul_values, std::vector< int64_t > *break_values)
GlobalDimensionCumulOptimizer(const RoutingDimension *dimension, RoutingSearchParameters::SchedulingSolver solver_type)
DimensionSchedulingStatus ComputeCumuls(const std::function< int64_t(int64_t)> &next_accessor, std::vector< int64_t > *optimal_cumuls, std::vector< int64_t > *optimal_breaks, std::vector< std::vector< int > > *optimal_resource_indices_per_group)
DimensionSchedulingStatus ComputeCumulCostWithoutFixedTransits(const std::function< int64_t(int64_t)> &next_accessor, int64_t *optimal_cost_without_transits)
DimensionSchedulingStatus ComputePackedCumuls(const std::function< int64_t(int64_t)> &next_accessor, std::vector< int64_t > *packed_cumuls, std::vector< int64_t > *packed_breaks, std::vector< std::vector< int > > *resource_indices_per_group)
DimensionSchedulingStatus ComputePackedRouteCumuls(int vehicle, const std::function< int64_t(int64_t)> &next_accessor, const RoutingModel::ResourceGroup::Resource *resource, std::vector< int64_t > *packed_cumuls, std::vector< int64_t > *packed_breaks)
DimensionSchedulingStatus ComputeRouteCumulCost(int vehicle, const std::function< int64_t(int64_t)> &next_accessor, int64_t *optimal_cost)
DimensionSchedulingStatus ComputeRouteCumuls(int vehicle, const std::function< int64_t(int64_t)> &next_accessor, std::vector< int64_t > *optimal_cumuls, std::vector< int64_t > *optimal_breaks)
DimensionSchedulingStatus ComputeRouteCumulCostWithoutFixedTransits(int vehicle, const std::function< int64_t(int64_t)> &next_accessor, int64_t *optimal_cost_without_transits)
LocalDimensionCumulOptimizer(const RoutingDimension *dimension, RoutingSearchParameters::SchedulingSolver solver_type)
std::vector< DimensionSchedulingStatus > ComputeRouteCumulCostsForResourcesWithoutFixedTransits(int vehicle, const std::function< int64_t(int64_t)> &next_accessor, const std::vector< RoutingModel::ResourceGroup::Resource > &resources, const std::vector< int > &resource_indices, bool optimize_vehicle_costs, std::vector< int64_t > *optimal_costs_without_transits, std::vector< std::vector< int64_t > > *optimal_cumuls, std::vector< std::vector< int64_t > > *optimal_breaks)
ResourceAssignmentOptimizer(const RoutingModel::ResourceGroup *resource_group, LocalDimensionCumulOptimizer *optimizer, LocalDimensionCumulOptimizer *mp_optimizer)
const RoutingDimension *const dimension() const
bool ComputeAssignmentCostsForVehicle(int v, const std::function< int64_t(int64_t)> &next_accessor, bool optimize_vehicle_costs, std::vector< int64_t > *assignment_costs, std::vector< std::vector< int64_t > > *cumul_values, std::vector< std::vector< int64_t > > *break_values)
int64_t ComputeBestAssignmentCost(const std::vector< std::vector< int64_t > > &primary_vehicle_to_resource_assignment_costs, const std::vector< std::vector< int64_t > > &secondary_vehicle_to_resource_assignment_costs, const std::function< bool(int)> &use_primary_for_vehicle, std::vector< int > *resource_indices) const
void SetCoefficient(int ct_index, int index, double coefficient) override
bool SetVariableBounds(int index, int64_t lower_bound, int64_t upper_bound) override
double GetValue(int index) const override
DimensionSchedulingStatus Solve(absl::Duration duration_limit) override
void AddMaximumConstraint(int max_var, std::vector< int > vars) override
void AddProductConstraint(int product_var, std::vector< int > vars) override
void SetEnforcementLiteral(int ct, int condition) override
double GetObjectiveCoefficient(int index) const override
int64_t GetVariableUpperBound(int index) const override
void SetVariableDisjointBounds(int index, const std::vector< int64_t > &starts, const std::vector< int64_t > &ends) override
int CreateNewConstraint(int64_t lower_bound, int64_t upper_bound) override
void SetObjectiveCoefficient(int index, double coefficient) override
int64_t GetVariableLowerBound(int index) const override
Dimensions represent quantities accumulated at nodes along the routes.
Definition: routing.h:2590
RoutingGlopWrapper(bool is_relaxation, const glop::GlopParameters &parameters)
bool SetVariableBounds(int index, int64_t lower_bound, int64_t upper_bound) override
double GetValue(int index) const override
DimensionSchedulingStatus Solve(absl::Duration duration_limit) override
void SetCoefficient(int ct, int index, double coefficient) override
void AddMaximumConstraint(int max_var, std::vector< int > vars) override
void AddProductConstraint(int product_var, std::vector< int > vars) override
void SetEnforcementLiteral(int ct, int condition) override
double GetObjectiveCoefficient(int index) const override
int64_t GetVariableUpperBound(int index) const override
void SetVariableDisjointBounds(int index, const std::vector< int64_t > &starts, const std::vector< int64_t > &ends) override
int CreateNewConstraint(int64_t lower_bound, int64_t upper_bound) override
void SetObjectiveCoefficient(int index, double coefficient) override
int64_t GetVariableLowerBound(int index) const override
virtual void SetCoefficient(int ct, int index, double coefficient)=0
int AddReifiedLinearConstraint(int64_t lower_bound, int64_t upper_bound, const std::vector< std::pair< int, double > > &weighted_variables)
virtual double GetObjectiveCoefficient(int index) const =0
virtual void AddProductConstraint(int product_var, std::vector< int > vars)=0
int AddVariable(int64_t lower_bound, int64_t upper_bound)
virtual int64_t GetObjectiveValue() const =0
virtual int CreateNewConstraint(int64_t lower_bound, int64_t upper_bound)=0
int AddLinearConstraint(int64_t lower_bound, int64_t upper_bound, const std::vector< std::pair< int, double > > &variable_coeffs)
virtual double GetValue(int index) const =0
virtual DimensionSchedulingStatus Solve(absl::Duration duration_limit)=0
virtual void SetVariableDisjointBounds(int index, const std::vector< int64_t > &starts, const std::vector< int64_t > &ends)=0
virtual bool SetVariableBounds(int index, int64_t lower_bound, int64_t upper_bound)=0
virtual void SetObjectiveCoefficient(int index, double coefficient)=0
virtual void SetEnforcementLiteral(int ct, int condition)=0
virtual int64_t GetVariableUpperBound(int index) const =0
virtual int64_t GetVariableLowerBound(int index) const =0
virtual void AddMaximumConstraint(int max_var, std::vector< int > vars)=0
A Resource sets attributes (costs/constraints) for a set of dimensions.
Definition: routing.h:419
A ResourceGroup defines a set of available Resources with attributes on one or multiple dimensions.
Definition: routing.h:398
Collection of objects used to extend the Constraint Solver library.