chore: minor improvements

Author: zepfred

core/src/main/java/ai/timefold/solver/core/impl/evolutionaryalgorithm/DefaultEvolutionaryAlgorithmPhaseFactory.java

@@ -405,9 +405,9 @@ private static <Solution_> Phase<Solution_> buildLocalSearchPhase(HeuristicConfi
         }
         if (updatedLocalSearchPhaseConfig.getTerminationConfig() == null) {
             var terminationConfig = new TerminationConfig();
-            var windowTime = isComplex ? 20L : 1L;
+            var windowTime = isComplex ? 30L : 1L;
             terminationConfig.setDiminishedReturnsConfig(new DiminishedReturnsTerminationConfig()
-                    .withMinimumImprovementRatio(0.01).withSlidingWindowSeconds(windowTime));
+                    .withMinimumImprovementRatio(0.001).withSlidingWindowSeconds(windowTime));
             updatedLocalSearchPhaseConfig.setTerminationConfig(terminationConfig);
         }
         var clearNearbyClass = updatedLocalSearchPhaseConfig.getMoveSelectorConfig() == null;

core/src/main/java/ai/timefold/solver/core/impl/evolutionaryalgorithm/decider/HybridGeneticSearchDecider.java

@@ -141,8 +141,8 @@ public void phaseStarted(EvolutionaryAlgorithmPhaseScope<Solution_> phaseScope)
         if (exploratoryRate == -1) { // Not set by the user
             var scaleLog = phaseScope.getSolverScope().getScoreDirector().getValueRangeManager().getProblemSizeStatistics()
                     .approximateProblemSizeLog();
-            // It is expected that the conservative profile will be called 90% of the time for complex problems
-            exploratoryRate = scaleLog < 427.0 ? 0.9 : 0.1;
+            // It is expected that the conservative profile will be called 95% of the time for complex problems
+            exploratoryRate = scaleLog < 427.0 ? 0.95 : 0.05;
         }
         this.worker = new HybridGeneticSearchWorker<>(HybridGeneticSearchWorkerContext.of(exploratoryRate, context),
                 bestSolutionUpdater, (DelegatingSplittableRandomGenerator) workerSolverScope.getWorkingRandom(),

core/src/main/java/ai/timefold/solver/core/impl/evolutionaryalgorithm/population/individual/generator/basic/BasicRuinRecreateIndividualStrategy.java

@@ -102,6 +102,10 @@ void applyRuinRecreate(InnerScoreDirector<Solution_, Score_> scoreDirector,
             EvolutionaryAlgorithmPhaseScope<Solution_> phaseScope, Individual<Solution_, Score_> bestIndividual) {
         var bestSolutionState = solutionStateManager.saveSolutionState(scoreDirector, bestIndividual);
         solutionStateManager.restoreSolutionState(scoreDirector, bestSolutionState);
+        if (workingRandom.nextBoolean()) {
+            // The method can also maintain the current best solution and restart the search from it
+            return;
+        }
         applyRuinPhase(scoreDirector, workingRandom, bestIndividual);
         updateScope(phaseScope);
         applyPhases(phaseScope, shuffledFirstFitConstructionPhase);

core/src/main/java/ai/timefold/solver/core/impl/evolutionaryalgorithm/population/individual/generator/list/ListRuinRecreateIndividualStrategy.java

@@ -104,6 +104,10 @@ public Phase<Solution_> getLocalSearchPhase() {
     void applyRuinRecreate(InnerScoreDirector<Solution_, Score_> scoreDirector, Individual<Solution_, Score_> bestIndividual) {
         var bestSolutionState = solutionStateManager.saveSolutionState(scoreDirector, bestIndividual);
         solutionStateManager.restoreSolutionState(scoreDirector, bestSolutionState);
+        if (workingRandom.nextBoolean()) {
+            // The method can also maintain the current best solution and restart the search from it
+            return;
+        }
         var listVariableDescriptor = Objects.requireNonNull(scoreDirector.getSolutionDescriptor().getListVariableDescriptor());
         var listVariableMetaModel = listVariableDescriptor.getVariableMetaModel();
         var valueRangeManager = scoreDirector.getValueRangeManager();