always-cautious/ortools-clone
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

[CP-SAT] misc improvements to the internals; more work on general constraints with enforcements

Browse Source
This commit is contained in:
Laurent Perron
2025-09-16 18:01:41 +02:00
committed by Corentin Le Molgat
parent cc94abdb08
commit a841258da7
21 changed files with 712 additions and 285 deletions
Download Patch File Download Diff File Expand all files Collapse all files

25
ortools/sat/integer_search.cc
View File

@@ -18,6 +18,7 @@
#include <cstdint>
#include <functional>
#include <random>
#include <string>
#include <tuple>
#include <vector>
@@ -259,7 +260,7 @@ std::function<BooleanOrIntegerLiteral()> LpPseudoCostHeuristic(Model* model) {
if (info.score > best_score) {
best_score = info.score;
// This direction works better than the inverse in the benchs. But
// This direction works better than the inverse in the benchmarks. But
// always branching up seems even better. TODO(user): investigate.
if (info.down_score > info.up_score) {
decision = BooleanOrIntegerLiteral(info.down_branch);
@@ -296,6 +297,9 @@ UnassignedVarWithLowestMinAtItsMinHeuristic(
std::function<BooleanOrIntegerLiteral()> SequentialSearch(
std::vector<std::function<BooleanOrIntegerLiteral()>> heuristics) {
for (const auto& h : heuristics) {
CHECK(h != nullptr);
}
return [heuristics]() {
for (const auto& h : heuristics) {
const BooleanOrIntegerLiteral decision = h();
@@ -990,10 +994,21 @@ std::function<BooleanOrIntegerLiteral()> RandomizeOnRestartHeuristic(
weights.push_back(1);
}
// Always add heuristic search.
policies.push_back(SequentialSearch({heuristics.heuristic_search, sat_policy,
heuristics.integer_completion_search}));
weights.push_back(1);
// Add heuristic search if present.
if (heuristics.heuristic_search != nullptr) {
policies.push_back(
SequentialSearch({heuristics.heuristic_search, sat_policy,
heuristics.integer_completion_search}));
weights.push_back(1);
}
if (policies.size() == 1) {
CHECK(heuristics.fixed_search != nullptr);
policies.push_back(
SequentialSearch({heuristics.fixed_search, sat_policy,
heuristics.integer_completion_search}));
weights.push_back(1);
}
// The higher weight for the sat policy is because this policy actually
// contains a lot of variation as we randomize the sat parameters.