/// <summary>
/// Applies this crossover operation.
/// </summary>
/// <param name="solver"></param>
/// <param name="parent1"></param>
/// <param name="parent2"></param>
/// <returns></returns>
public Individual <List <int>, GeneticProblem, Fitness> CrossOver(
Solver <List <int>, GeneticProblem, Fitness> solver,
Individual <List <int>, GeneticProblem, Fitness> parent1,
Individual <List <int>, GeneticProblem, Fitness> parent2)
{
OsmSharp.Math.TSP.Problems.IProblem tsp_problem = solver.Problem.BaseProblem;
double[][] weights = tsp_problem.WeightMatrix;
// first create E_a
AsymmetricCycles e_a = new AsymmetricCycles(parent1.Genomes.Count + 1);
e_a.AddEdge(0, parent1.Genomes[0]);
for (int idx = 0; idx < parent1.Genomes.Count - 1; idx++)
{
int from = parent1.Genomes[idx];
int to = parent1.Genomes[idx + 1];
e_a.AddEdge(from, to);
}
e_a.AddEdge(parent1.Genomes[parent1.Genomes.Count - 1], 0);
// create E_b
int[] e_b = new int[parent2.Genomes.Count + 1];
e_b[parent2.Genomes[0]] = 0;
for (int idx = 0; idx < parent2.Genomes.Count - 1; idx++)
{
int from = parent2.Genomes[idx];
int to = parent2.Genomes[idx + 1];
e_b[to] = from;
}
e_b[0] = parent2.Genomes[parent2.Genomes.Count - 1];
// create cycles.
AsymmetricAlternatingCycles cycles = new AsymmetricAlternatingCycles(
parent2.Genomes.Count + 1);
for (int idx = 0; idx < e_b.Length; idx++)
{
int a = e_a[idx];
int b = e_b[a];
if (idx != b)
{
cycles.AddEdge(idx, a, b);
}
}
// the cycles that can be selected.
List <int> selectable_cycles = new List <int>(cycles.Cycles.Keys);
int generated = 0;
Individual best = null;
while (generated < _max_offspring &&
selectable_cycles.Count > 0)
{
// select some random cycles.
List <int> cycle_starts = this.SelectCycles(selectable_cycles);
// copy if needed.
AsymmetricCycles a = null;
if (_max_offspring > 1)
{
a = e_a.Clone();
}
else
{
a = e_a;
}
// take e_a and remove all edges that are in the selected cycles and replace them by the eges
int[] next_array_a = a.NextArray;
foreach (int start in cycle_starts)
{
int current = start;
KeyValuePair <int, int> current_next = cycles.Next(current);
do
{
a.AddEdge(current_next.Value, current_next.Key);
current = current_next.Value;
current_next = cycles.Next(current);
} while(current != start);
}
// connect all subtoures.
int cycle_count = a.Cycles.Count;
while (cycle_count > 1)
{
// get the smallest tour.
KeyValuePair <int, int> current_tour =
new KeyValuePair <int, int>(-1, int.MaxValue);
foreach (KeyValuePair <int, int> tour in a.Cycles)
{
if (tour.Value < current_tour.Value)
{
current_tour = tour;
}
}
// first try nn approach.
double weight = double.MaxValue;
int selected_from1 = -1;
int selected_from2 = -1;
int selected_to1 = -1;
int selected_to2 = -1;
bool[] ignore_list = new bool[a.Length];
int from;
int to;
from = current_tour.Key;
ignore_list[from] = true;
//ignore_list.Add(from);
to = next_array_a[from];
do
{
// step to the next ones.
from = to;
to = next_array_a[from];
//ignore_list.Add(from);
ignore_list[from] = true;
} while (from != current_tour.Key);
if (_nn)
{ // only try tours containing nn.
from = current_tour.Key;
to = next_array_a[from];
double weight_from_to = weights[from][to];
do
{
// check the nearest neighbours of from
foreach (int nn in tsp_problem.Get10NearestNeighbours(from))
{
int nn_to = next_array_a[nn];
//if (!ignore_list.Contains(nn) &&
// !ignore_list.Contains(nn_to))
if (!ignore_list[nn] &&
!ignore_list[nn_to])
{
double merge_weight =
(weights[from][nn_to] + weights[nn][to]) -
(weight_from_to + weights[nn][nn_to]);
if (weight > merge_weight)
{
weight = merge_weight;
selected_from1 = from;
selected_from2 = nn;
selected_to1 = to;
selected_to2 = nn_to;
}
}
}
// step to the next ones.
from = to;
to = next_array_a[from];
} while (from != current_tour.Key);
}
if (selected_from2 < 0)
{
// check the nearest neighbours of from
foreach (int customer in parent1.Genomes)
{
int customer_to = next_array_a[customer];
//if (!ignore_list.Contains(customer) &&
// !ignore_list.Contains(customer_to))
if (!ignore_list[customer] &&
!ignore_list[customer_to])
{
double merge_weight =
(weights[from][customer_to] + weights[customer][to]) -
(weights[from][to] + weights[customer][customer_to]);
if (weight > merge_weight)
{
weight = merge_weight;
selected_from1 = from;
selected_from2 = customer;
selected_to1 = to;
selected_to2 = customer_to;
}
}
}
}
//if (selected_from1 >= 0)
//{
a.AddEdge(selected_from1, selected_to2);
a.AddEdge(selected_from2, selected_to1);
//}
//else
//{
// throw new Exception();
//}
cycle_count--;
}
//if (a.Cycles.Values.First<int>() != a.Length)
//{
// throw new Exception();
//}
List <int> new_genome = new List <int>(a.Length);
int next = next_array_a[0];
do
{
new_genome.Add(next);
next = next_array_a[next];
}while (next != 0);
Individual individual = new Individual(new_genome);
individual.CalculateFitness(solver.Problem, solver.FitnessCalculator);
if (best == null ||
best.Fitness.CompareTo(individual.Fitness) > 0)
{
best = individual;
}
generated++;
}
if (best == null)
{
List <int> new_genome = new List <int>();
int next = e_a[0];
do
{
new_genome.Add(next);
next = e_a[next];
}while (next != 0);
Individual individual = new Individual(new_genome);
individual.CalculateFitness(solver.Problem, solver.FitnessCalculator);
if (best == null ||
best.Fitness.CompareTo(individual.Fitness) > 0)
{
best = individual;
}
}
return(best);
}
/// <summary>
/// Solves the problem using a GA.
/// </summary>
/// <param name="problem"></param>
/// <returns></returns>
protected override IRoute DoSolve(OsmSharp.Math.TSP.Problems.IProblem problem)
{
//int population_size = 10;
//if (problem.Size < 100)
//{
// population_size = System.Math.Max(problem.Size * 3, 10);
// if (problem.Size < 10)
// {
// population_size = 1;
// }
//}
//if (problem.Size < 1000)
//{
// population_size = problem.Size / 4;
//}
// create the settings.
SolverSettings settings = new SolverSettings(
_stagnation_count,
_population,
1000000000,
_eltism,
_cross,
_mutation);
//List<IMutationOperation<List<int>, GeneticProblem, Fitness>> mutators = new List<IMutationOperation<int,GeneticProblem,Fitness>>();
////mutators.Add(new DefaultMutationOperation());
////mutators.Add(new BestPlacementMutationOperation());
//mutators.Add(new BestDetailedPlacementMutationOperation());
//List<double> probabilities = new List<double>();
//probabilities.Add(1);
////probabilities.Add(0.5);
////probabilities.Add(0.3);
//CombinedMutation<List<int>, GeneticProblem, Fitness> mutation = new CombinedMutation<int,GeneticProblem,Fitness>(
// StaticRandomGenerator.Get(),
// mutators,
// probabilities);
////SequentialContructiveCrossoverOperator cross_over = new SequentialContructiveCrossoverOperator();
//BestDetailedPlacementCrossOverOperation cross_over = new BestDetailedPlacementCrossOverOperation();
////BestPlacementCrossOverOperation cross_over = new BestPlacementCrossOverOperation();
////EdgeRecombinationCrossOverOperation cross_over = new EdgeRecombinationCrossOverOperation();
//BestPlacementGenerationOperation generation = new BestPlacementGenerationOperation();
////RandomGenerationOperation generation = new RandomGenerationOperation();
ISelector <List <int>, GeneticProblem, Fitness> selector = new RandomSelector <List <int>, GeneticProblem, Fitness>();
//ISelector<List<int>, GeneticProblem, Fitness> selector = new TournamentBasedSelector<List<int>, GeneticProblem, Fitness>(75, 0.01);
solver =
new Solver <List <int>, GeneticProblem, Fitness>(
new GeneticProblem(problem),
settings,
selector,
_mutation_operation,
_cross_over_operation,
_generation_operation,
new FitnessCalculator(),
true, false);
solver.NewFittest += new Solver <List <int>, GeneticProblem, Fitness> .NewFittestDelegate(solver_NewFittest);
solver.NewGeneration += new Solver <List <int>, GeneticProblem, Fitness> .NewGenerationDelegate(solver_NewGeneration);
List <int> result = new List <int>(solver.Start(null).Genomes);
result.Insert(0, 0);
return(new SimpleAsymmetricRoute(result, true));
}
/// <summary>
/// Creates a new genetic problem.
/// </summary>
/// <param name="problem"></param>
public GeneticProblem(OsmSharp.Math.TSP.Problems.IProblem problem)
{
_problem = problem;
}
/// <summary>
/// Applies this operation.
/// </summary>
/// <param name="solver"></param>
/// <param name="parent1"></param>
/// <param name="parent2"></param>
/// <returns></returns>
public Individual <List <int>, GeneticProblem, Fitness> CrossOver(
Solver <List <int>, GeneticProblem, Fitness> solver,
Individual <List <int>, GeneticProblem, Fitness> parent1,
Individual <List <int>, GeneticProblem, Fitness> parent2)
{
OsmSharp.Math.TSP.Problems.IProblem tsp_problem = solver.Problem.BaseProblem;
// first create E_a
int[] e_a = new int[parent1.Genomes.Count + 1];
e_a[0] = parent1.Genomes[0];
for (int idx = 0; idx < parent1.Genomes.Count - 1; idx++)
{
int from = parent1.Genomes[idx];
int to = parent1.Genomes[idx + 1];
e_a[from] = to;
}
e_a[parent1.Genomes[parent1.Genomes.Count - 1]] = 0;
// create E_b
int[] e_b = new int[parent2.Genomes.Count + 1];
e_b[0] = parent2.Genomes[0];
for (int idx = 0; idx < parent2.Genomes.Count - 1; idx++)
{
int from = parent2.Genomes[idx];
int to = parent2.Genomes[idx + 1];
e_b[from] = to;
}
e_b[parent2.Genomes[parent2.Genomes.Count - 1]] = 0;
// create G_ab.
Dictionary <int, int> g_a =
new Dictionary <int, int>();
Dictionary <int, int> g_b_opposite =
new Dictionary <int, int>();
for (int idx = 0; idx < e_b.Length; idx++)
{
if (e_a[idx] != e_b[idx])
{
g_a.Add(idx, e_a[idx]);
g_b_opposite.Add(e_b[idx], idx);
}
}
List <Dictionary <int, KeyValuePair <int, int> > > cycles =
new List <Dictionary <int, KeyValuePair <int, int> > >();
while (g_a.Count > 0)
{
int first = g_a.Keys.First <int>();
KeyValuePair <int, int> next_pair;
Dictionary <int, KeyValuePair <int, int> > cycle =
new Dictionary <int, KeyValuePair <int, int> >();
while (!cycle.ContainsKey(first))
{
next_pair = new KeyValuePair <int, int>(g_a[first], g_b_opposite[g_a[first]]);
// remove.
g_a.Remove(first);
g_b_opposite.Remove(next_pair.Key);
cycle.Add(first, next_pair);
// get all the nexts ones.
first = next_pair.Value;
}
cycles.Add(cycle);
}
int generated = 0;
Individual best = null;
while (generated < _max_offspring)
{
// select some random cycles.
List <Dictionary <int, KeyValuePair <int, int> > > selected_cycles =
this.SelectCycles(cycles);
// take e_a and remove all edges that are in the selected cycles and replace them by the eges
// from e_b in the cycles.
foreach (Dictionary <int, KeyValuePair <int, int> > cycle in selected_cycles)
{
foreach (KeyValuePair <int, KeyValuePair <int, int> > pair in cycle)
{
e_a[pair.Value.Value] = pair.Value.Key;
}
}
// construct all subtours.
List <Dictionary <int, KeyValuePair <int, double> > > subtours =
new List <Dictionary <int, KeyValuePair <int, double> > >();
List <int> e_a_list = new List <int>(e_a);
int start_idx = 0;
while (start_idx < e_a_list.Count)
{
Dictionary <int, KeyValuePair <int, double> > current_tour = new Dictionary <int, KeyValuePair <int, double> >();
int from = start_idx;
int to = e_a_list[from];
while (to >= 0)
{
e_a_list[from] = -1;
current_tour.Add(from, new KeyValuePair <int, double>(to, tsp_problem.WeightMatrix[from][to]));
from = to;
to = e_a_list[from];
}
for (start_idx = start_idx + 1; start_idx < e_a_list.Count; start_idx++)
{
if (e_a_list[start_idx] >= 0)
{
break;
}
}
subtours.Add(current_tour);
}
while (subtours.Count > 1)
{
int size = e_a.Length;
Dictionary <int, KeyValuePair <int, double> > current_tour = null;
foreach (Dictionary <int, KeyValuePair <int, double> > tour in subtours)
{
if (current_tour == null ||
tour.Count < current_tour.Count)
{
current_tour = tour;
}
}
// merge two tours.
Dictionary <int, KeyValuePair <int, double> > target_tour = null;
double weight = double.MaxValue;
KeyValuePair <int, KeyValuePair <int, double> > from = new KeyValuePair <int, KeyValuePair <int, double> >();
KeyValuePair <int, KeyValuePair <int, double> > to = new KeyValuePair <int, KeyValuePair <int, double> >();
// first try nn approach.
if (_nn)
{
foreach (KeyValuePair <int, KeyValuePair <int, double> > from_pair in current_tour)
{
foreach (int nn in tsp_problem.Get10NearestNeighbours(from_pair.Key))
{
if (!current_tour.ContainsKey(nn))
{
foreach (Dictionary <int, KeyValuePair <int, double> > target in subtours)
{
KeyValuePair <int, double> to_pair_value;
if (target.TryGetValue(nn, out to_pair_value))
{
double merge_weight = (tsp_problem.WeightMatrix[from_pair.Key][to_pair_value.Key] +
tsp_problem.WeightMatrix[nn][from_pair.Value.Key]) -
(from_pair.Value.Value + to_pair_value.Value);
if (merge_weight < weight)
{
weight = merge_weight;
from = from_pair;
to = new KeyValuePair <int, KeyValuePair <int, double> >(nn, to_pair_value);
target_tour = target;
}
}
}
}
}
}
}
if (target_tour == null)
{
foreach (KeyValuePair <int, KeyValuePair <int, double> > from_pair in current_tour)
{
foreach (Dictionary <int, KeyValuePair <int, double> > target in subtours)
{
if (target != current_tour)
{
foreach (KeyValuePair <int, KeyValuePair <int, double> > to_pair in target)
{
double merge_weight = (tsp_problem.WeightMatrix[from_pair.Key][to_pair.Value.Key] +
tsp_problem.WeightMatrix[to_pair.Key][from_pair.Value.Key]) -
(from_pair.Value.Value + to_pair.Value.Value);
if (merge_weight < weight)
{
weight = merge_weight;
from = from_pair;
to = to_pair;
target_tour = target;
}
}
}
}
}
}
// merge the tours.
subtours.Remove(current_tour);
foreach (KeyValuePair <int, KeyValuePair <int, double> > for_target in current_tour)
{
target_tour.Add(for_target.Key, for_target.Value);
}
target_tour[from.Key] = new KeyValuePair <int, double>(to.Value.Key,
tsp_problem.WeightMatrix[from.Key][to.Value.Key]);
target_tour[to.Key] = new KeyValuePair <int, double>(from.Value.Key,
tsp_problem.WeightMatrix[to.Key][from.Value.Key]);
}
List <int> new_genome = new List <int>();
int next = subtours[0][0].Key;
do
{
new_genome.Add(next);
next = subtours[0][next].Key;
}while (next != 0);
Individual individual = new Individual(new_genome);
individual.CalculateFitness(solver.Problem, solver.FitnessCalculator);
if (best == null ||
best.Fitness.CompareTo(individual.Fitness) > 0)
{
best = individual;
}
generated++;
}
return(best);
}
/// <summary>
/// Creates a new genetic problem.
/// </summary>
/// <param name="problem"></param>
public GeneticProblem(OsmSharp.Math.TSP.Problems.IProblem problem)
{
_problem = problem;
}