/// <summary>
/// Return true if one of the constraints gets violated
/// </summary>
/// <param name="ea">Algorithm to check constraints on</param>
/// <param name="maxRuntime">Maximum algorithm runtime</param>
/// <param name="maxEpochs">Maximum algorithm iterations</param>
/// <param name="maxStagnation">Maximum time no change occurs</param>
/// <param name="minFitness">Minimum achieved fitness</param>
/// <returns>True if algorithm should stop</returns>
private bool StopCondition(
EvolutionaryAlgorithm ea,
int?maxRuntime,
int?maxEpochs,
int?maxStagnation,
double?minFitness)
{
bool stop = false;
if (maxRuntime.HasValue)
{
stop |= ea.TimeRan > TimeSpan.FromSeconds(maxRuntime.Value);
}
if (maxEpochs.HasValue)
{
stop |= ea.Epoch >= maxEpochs.Value;
}
if (maxStagnation.HasValue)
{
stop |= ea.StagnationCount >= maxStagnation.Value;
}
if (minFitness.HasValue)
{
stop |= ea.Fitness >= minFitness.Value;
}
return(stop);
}
private void LoadSerialzedCanvasIfExists(EvolutionaryAlgorithm algorithm, string fileName = "saves/Suspended.canvas")
{
if (File.Exists(fileName))
{
LoadSerializedCanvas(algorithm, fileName);
}
}
/// <summary>
/// Print results for a single epoch.
/// </summary>
/// <param name="algorithm">algorithm to print for</param>
private static void PrintEpochResults(EvolutionaryAlgorithm algorithm)
{
// print fittest
Console.Write("Epoch {0, -4}: {1,5:N5}", algorithm.Epoch, algorithm.Fitness);
if (algorithm.StagnationCount > 0)
{
for (int i = 0; i < algorithm.StagnationCount; i++)
{
Console.Write('*');
}
}
Console.WriteLine();
}
/// <summary>
/// Create a complete new population with totally random values for all chromosomes.
/// </summary>
/// <param name="numberOfRepopulations"></param>
/// <param name="populationSize"></param>
/// <param name="numberOfTrials"></param>
/// <param name="gameSpeed"></param>
/// <param name="tTest"></param>
/// <param name="nuberOfCrossOvers">Make one point cross over on the specified number of best candidates, and randomly pick the half of the new chromosomes.</param>
public Trainer(int numberOfRepopulations, int populationSize, int numberOfTrials, int gameSpeed, Boolean tTest, int nuberOfCrossOvers)
{
_numberOfRepopulations = numberOfRepopulations;
_populationSize = populationSize;
_numberOfTrials = numberOfTrials;
_nuberOfCrossOvers = nuberOfCrossOvers;
_ea = new EvolutionaryAlgorithm(_populationSize);
_gameSpeed = gameSpeed;
if (tTest)
PlayWithRandomChromosomes();//TTest();
else
Train();
}
private void LoadSerializedCanvas(EvolutionaryAlgorithm algorithm, string fileName = "saves/Suspended.canvas")
{
var canvas = File.ReadAllBytes(fileName).AsCanvas(algorithm);
if (algorithm.Population == null || algorithm.Population.Count == 0)
{
algorithm.Population = new List <ICanvas>()
{
canvas
};
}
else
{
algorithm.Population.Add(canvas);
}
}
private void GenerateDiffImageAndSave(EvolutionaryAlgorithm algorithm)
{
var mainImage = ToBitmap(MainImage.Source as BitmapImage).AsByteArray();
var seedImage = ToBitmap(SeedImage.Source as BitmapImage).AsByteArray();
var diffImage = new byte[mainImage.Length];
for (int i = 0; i < mainImage.Length; i += 4)
{
var dB = Math.Abs((int)mainImage[i] - (int)seedImage[i]);
var dG = Math.Abs((int)mainImage[i + 1] - (int)seedImage[i + 1]);
var dR = Math.Abs((int)mainImage[i + 2] - (int)seedImage[i + 2]);
var dT = (dR + dG + dB) / 3;
diffImage[i / 4] = (byte)(255 - dT);
}
SaveBitmap(diffImage.AsBitmap(algorithm.CanvasWidth, algorithm.CanvasHeight), CompareImage);
}
static void Phase2()
{
EvolutionaryAlgorithm eva = new EvolutionaryAlgorithm();
MapBlueprintIndividual input = (MapBlueprintIndividual)Individual.FromFile(@"C:\Users\Jan\Source\Repos\RoguelikeEva\Results\door test\_RESULT 9.bin");
var graph = ExtractRoomAdjacencyGraph(input.Rooms.First().Key, input);
eva.Operators.Add(new ExtendBranchMutation(1));
/*eva.Operators.Add(new CompressBranchMutation(1));
* eva.Operators.Add(new ReplaceBranchMutation(1));
* eva.Operators.Add(new ReverseBranchMutation(1));
* eva.Operators.Add(new ShiftStartingRoomMutation(1));*/
eva.MultiObjective.Add(new BranchRatioFitness(3.0 / 2));
eva.MultiObjective.Add(new BranchEntryPointsFitness());
eva.SampleIndividual = new AdjacencyGraphIndividual(new List <byte>(input.Rooms.Keys), 3, 3);
eva.HvIndicator = new BiobjectiveHvIndicator();
using (StreamWriter file = new StreamWriter(@"C:\Users\Jan\Source\Repos\RoguelikeEva\Results\_out.txt"))
{
file.WriteLine("Gen\tHv");
int solutionNo = 0;
IEnumerable <Individual> result = eva.Run(e => e.Population.No == 10, 100, file, (writer, gen, hv) =>
{
writer.WriteLine("{0}\t{1}", gen.No, hv);
Console.Clear();
Console.WriteLine("Current generation: " + gen.No);
});
foreach (Individual ind in result)
{
AdjacencyGraphIndividual map = (AdjacencyGraphIndividual)ind;
// TODO:
// transform "map" to correct MapBlueprint
// "correct MapBlueprint".SaveImageToFile(@"C:\Users\Jan\Source\Repos\RoguelikeEva\Results\_RESULT " + solutionNo + ".png", 1536, 1024);
map.SaveToFile(@"C:\Users\Jan\Source\Repos\RoguelikeEva\Results\_RESULT " + (solutionNo++) + ".bin");
}
}
}
/// <summary>
/// Entry point method
/// </summary>
/// <param name="args">Not used</param>
public static void Main(string[] args)
{
Console.WriteLine("Starting algorithm...");
var algorithm = new EvolutionaryAlgorithm();
// subscribe to algorithm started (save seed image)
algorithm.AlgorithmStarted += (s, e) => algorithm.Seed.Save("Seed.bmp");
// subscribe to epoch done (print results each epoch)
algorithm.EpochCompleted += (s, e) => PrintEpochResults(algorithm);
// subscribe to algorithm done even (save result to bitmap)
algorithm.AlgorithmCompleted += (s, e) => Painter.Paint(algorithm.Population.CalculateFittest()).Save("Fittest.bmp");
algorithm.Run(() => algorithm.StagnationCount > 10);
Console.WriteLine("Press any key to exit...");
Console.ReadKey();
}
static void Phase1()
{
EvolutionaryAlgorithm eva = new EvolutionaryAlgorithm();
eva.Operators.Add(new FloorTypeMutation(0.5));
eva.Operators.Add(new JoinRoomsMutation(0.8));
eva.Operators.Add(new RemoveRoomMutation(0.8));
eva.Operators.Add(new FillHoleMutation(1));
eva.Operators.Add(new ExtendEdgeMutation(1));
eva.MultiObjective.Add(new AdjacencyFitness());
eva.MultiObjective.Add(new RoomFitness(20));
eva.MultiObjective.Add(new SpaceFitness());
eva.SampleIndividual = new MapBlueprintIndividual(30, 20, 0.5);
eva.HvIndicator = new HvSweep3D();
//eva.MatingSelector = new RouletteWheelSelector();
using (StreamWriter file = new StreamWriter(@"C:\Users\Jan\Source\Repos\RoguelikeEva\Results\_out.txt"))
{
file.WriteLine("Gen\tHv");
int solutionNo = 0;
IEnumerable <Individual> result = eva.Run(e => e.Population.No == 10, 100, file, (writer, gen, hv) =>
{
writer.WriteLine("{0}\t{1}", gen.No, hv);
Console.Clear();
Console.WriteLine("Current generation: " + gen.No);
});
foreach (Individual ind in result)
{
MapBlueprintIndividual map = (MapBlueprintIndividual)ind;
AddDoors(map.Gene);
// TODO: skip map which are not correct (discontinuous)
map.Gene.SaveImageToFile(@"C:\Users\Jan\Source\Repos\RoguelikeEva\Results\_RESULT " + solutionNo + ".png", 1536, 1024);
map.SaveToFile(@"C:\Users\Jan\Source\Repos\RoguelikeEva\Results\_RESULT " + (solutionNo++) + ".bin");
}
}
}
public void EvolSearchTravelingSalesman()
{
/*int test = 5;
* object obj = new Object();
* obj = test;
* test++;
* int so;*/
/*Place place = new Place(10,20);
* PlaceList list = new PlaceList();
* list.Add(place);
* XmlSerializer serializer = new XmlSerializer(typeof(PlaceList));
* TextWriter writer = new StreamWriter( @"F:\test.xml" );
* serializer.Serialize( writer, list );
* writer.Close();*/
/*TextReader reader = new StreamReader("places.xml");
* XmlSerializer serializer = new XmlSerializer(typeof(PlaceList));
* PlaceList list = (PlaceList)serializer.Deserialize(reader);
* reader.Close();*/
// http://www.zib.de/groetschel/news/tsp_loesungen_version2.html
// min: 2069
PlaceList list = new PlaceList();
list.Add(new Place(144, 158));
list.Add(new Place(188, 80));
list.Add(new Place(190, 134));
list.Add(new Place(238, 125));
list.Add(new Place(61, 265));
list.Add(new Place(175, 205));
list.Add(new Place(248, 216));
list.Add(new Place(315, 195));
list.Add(new Place(303, 205));
list.Add(new Place(222, 300));
list.Add(new Place(340, 285));
list.Add(new Place(69, 399));
list.Add(new Place(120, 354));
list.Add(new Place(153, 430));
list.Add(new Place(250, 472));
list.Add(new Place(163, 57));
list.Add(new Place(83, 140));
list.Add(new Place(210, 104));
list.Add(new Place(298, 85));
list.Add(new Place(266, 100));
list.Add(new Place(100, 235));
list.Add(new Place(205, 208));
list.Add(new Place(249, 194));
list.Add(new Place(359, 205));
list.Add(new Place(65, 288));
list.Add(new Place(158, 270));
list.Add(new Place(186, 254));
list.Add(new Place(281, 259));
list.Add(new Place(353, 242));
list.Add(new Place(131, 321));
list.Add(new Place(252, 298));
list.Add(new Place(137, 340));
list.Add(new Place(52, 363));
list.Add(new Place(133, 394));
list.Add(new Place(184, 368));
list.Add(new Place(240, 384));
list.Add(new Place(98, 477));
list.Add(new Place(222, 458));
list.Add(new Place(325, 441));
list.Add(new Place(195, 499));
int countPlaces = list.Count;
// Anzahl aller Distanzen: ((AnzahlOrte*AnzahlOrte)-AnzahlOrte)/2
int countDistances = (countPlaces * countPlaces - countPlaces) / 2;
Matrix distanceMatrix = new Matrix(countPlaces, countPlaces);
double distance;
for (int i = 0; i < countPlaces; i++)
{
distanceMatrix.SetValue(i, i, 0.0);
for (int j = i + 1; j < countPlaces; j++)
{
distance = list[i].GetDistance(list[j]);
distanceMatrix.SetValue(i, j, distance);
distanceMatrix.SetValue(j, i, distance);
}
}
// Distancematrix kann als static zugewiesen werden, da sie f¸r alle solutions gleich ist
TravelingSalesmanSolution.DistanceMatrix = distanceMatrix;
TravelingSalesmanProblem problem = new TravelingSalesmanProblem(10, 40);
EvolutionaryAlgorithm evol = new EvolutionaryAlgorithm(problem);
//evol.Run();
// Anlegen von 10 Ausgangsindividuen
/*TravelingSalesmanSolution[] tspList = new TravelingSalesmanSolution[10];
*
* // Speicher reservieren
* for(int i=0; i<tspList.GetLength(0); i++)
* {
* tspList[i] = new TravelingSalesmanSolution();
* }*/
/*Random random = new Random(unchecked((int)DateTime.Now.Ticks));
* int countPlaces = 100;
*
* // Anzahl aller Distanzen: ((AnzahlOrte*AnzahlOrte)-AnzahlOrte)/2
* int countDistances = (countPlaces*countPlaces-countPlaces)/2;
* int[] distanceList = new int[countDistances];
* Matrix distanceMatrix = new Matrix(countPlaces, countPlaces);
*
* for (int i=0; i<distanceList.GetLength(0); i++)
* {
* distanceList[i] = Convert.ToInt32(Math.Round(random.NextDouble()*500));
* }
*
* int index = 0;
* for (int i=0; i<countPlaces; i++)
* {
* distanceMatrix.SetValue(i, i, 0);
* for (int j=i+1;j<countPlaces;j++)
* {
* distanceMatrix.SetValue(i, j, distanceList[index]);
* distanceMatrix.SetValue(j, i, distanceList[index]);
* index++;
* }
* }
*
* TravelingSalesmanSolution.DistanceMatrix = distanceMatrix;
*
* TravelingSalesmanSolution[] tspList = new TravelingSalesmanSolution[10];
* for(int i=0; i<tspList.GetLength(0); i++)
* {
* tspList[i] = new TravelingSalesmanSolution();
* }
*
* int a = 10;*/
}
/// <summary>
/// Use the best optimized agents to play StarCraft Broodwar (using the fittest unit agent optimized properties loaded from an XML file).
/// </summary>
/// <param name="numberOfRestarts"></param>
/// <param name="populationSize"></param>
/// <param name="gameSpeed"></param>
public Trainer(int numberOfRestarts, int populationSize, int gameSpeed)
{
_numberOfRepopulations = numberOfRestarts;
_populationSize = populationSize;
_unitAgentXMLFileName = "";
var xmlFileNames = new List<String>() {"Terran_Goliath.xml", "Terran_Vulture.xml", "Terran_SiegeTank.xml" };
_gameSpeed = gameSpeed;
_ea = new EvolutionaryAlgorithm(_populationSize, xmlFileNames);
PlayWithoutTraining(_gameSpeed, numberOfRestarts, true);//loadAgent: TRUE: CALL PLAY and use the current best optimized XMLfile, else use test method to load some standard values.
}
/// <summary>
/// Use the best optimized agent to play StarCraft Broodwar (using the fittest unit agent optimized properties loaded from an XML file).
/// </summary>
/// <param name="numberOfRestarts"></param>
/// <param name="populationSize"></param>
/// <param name="unitAgentXMLFileName"></param>
/// <param name="gameSpeed"></param>
public Trainer(int numberOfRestarts, int populationSize, String unitAgentXMLFileName, int gameSpeed)
{
_numberOfRepopulations = numberOfRestarts;
_populationSize = populationSize;
_unitAgentXMLFileName = unitAgentXMLFileName;
_gameSpeed = gameSpeed;
_ea = new EvolutionaryAlgorithm(_populationSize, _unitAgentXMLFileName);
PlayWithoutTraining(_gameSpeed, numberOfRestarts, true);//loadAgent: TRUE: CALL PLAY and use the current best optimized XMLfile, else use test method to load some standard values.
}
/// <summary>
/// Create a population of existing chromosomes from a file with a possible mixture of new random chromosomes.
/// </summary>
/// <param name="totalNumberOfRepopulations"></param>
/// <param name="startGenerationOn"></param>
/// <param name="percentOfOpPropsToLoad"></param>
/// <param name="populationSize"></param>
/// <param name="numberOfTrials"></param>
/// <param name="unitAgentXMLFileName"></param>
/// <param name="gameSpeed"></param>
/// <param name="nuberOfCrossOvers">Make one point cross over on the specified number of best candidates, and randomly pick the half of the new chromosomes.</param>
public Trainer(int totalNumberOfRepopulations, int startGenerationOn, int percentOfOpPropsToLoad, int populationSize, int numberOfTrials, String unitAgentXMLFileName, int gameSpeed, int nuberOfCrossOvers)
{
if (totalNumberOfRepopulations > startGenerationOn)
_numberOfRepopulations = totalNumberOfRepopulations - startGenerationOn;
else
{
Logger.Logger.AddAndPrint("numberOfRepopulations needs to be bigger than startGenerationOn.");
throw new Exception("numberOfRepopulations needs to be bigger than startGenerationOn.");
}
_populationSize = populationSize;
_numberOfTrials = numberOfTrials;
_nuberOfCrossOvers = nuberOfCrossOvers;
_unitAgentXMLFileName = unitAgentXMLFileName;
_startGenerationOn = startGenerationOn;
_gameSpeed = gameSpeed;
_ea = new EvolutionaryAlgorithm(_populationSize, percentOfOpPropsToLoad, _unitAgentXMLFileName);
Train();
}
private void SaveSerializedCanvas(EvolutionaryAlgorithm algorithm, string fileName = "saves/Suspended.canvas")
{
var bytes = algorithm.Population.CalculateFittest().AsByteArray();
File.WriteAllBytes(fileName, bytes);
}
/// <summary>
/// Called when run button is clicked
/// </summary>
/// <param name="sender">Event initiator</param>
/// <param name="e">Event arguments</param>
private async void Run_Click(object sender, RoutedEventArgs e)
{
if (SeedImage.Source == null)
{
Label_Status.Content = "No image set!";
return;
}
if (ctc != null)
{
ctc.Cancel();
ctc = null;
Button_Run.Content = "Run";
}
else
{
Button_Run.Content = "Stop";
ctc = new CancellationTokenSource();
saveWhileRunning = CheckBox_SaveWhileRunning.IsChecked == true;
var algorithm = new EvolutionaryAlgorithm()
{
CanvasWidth = int.Parse(TextBox_CanvasSizeX.Text),
CanvasHeight = int.Parse(TextBox_CanvasSizeY.Text),
CanvasCount = int.Parse(TextBox_PopulationSize.Text),
PolygonCount = int.Parse(TextBox_PolygonCount.Text),
PolygonEdgeCount = int.Parse(((ComboBoxItem)ComboBox_PolygonType.SelectedItem).Tag.ToString()),
WeightPositionChange = double.Parse(this.TextBox_PositionChange.Text),
WeightColorChange = double.Parse(this.TextBox_ColorChange.Text),
WeightIndexChange = double.Parse(this.TextBox_ZIndexChange.Text),
WeightRandomChange = double.Parse(this.TextBox_RandomChange.Text),
Seed = ToBitmap((BitmapImage)SeedImage.Source)
};
// If a saves/Serialized.canvas exists, load it.
if (CheckBox_LoadSusupended.IsChecked == true)
{
LoadSerialzedCanvasIfExists(algorithm);
}
// Get user defined values from the GUI where the checkbox/textbox structure applies
int? maxRuntime = GetIntIfFilledIn(CheckBox_MaxRuntime, TextBox_MaxRuntime);
int? maxEpochs = GetIntIfFilledIn(CheckBox_MaxEpochs, TextBox_MaxEpochs);
int? maxStagnation = GetIntIfFilledIn(Checkbox_MaxStagnation, TextBox_MaxStagnation);
double?minFitness = GetDoubleIfFilledIn(Checkbox_MinFitness, Textbox_MinFitness);
try
{
algorithm.DeleteStatsFileIfExists();
algorithm.AlgorithmStarted += (s, args) => Dispatcher.Invoke(new Action(() =>
{
try
{
algorithm.Seed.Save("img/Seed.bmp");
}
catch (Exception)
{
// do nothing
}
}));
algorithm.EpochCompleted += (s, args) => Dispatcher.Invoke(new Action(() =>
{
if (MainImage.Source == null)
{
return;
}
GenerateDiffImageAndSave(algorithm);
}));
algorithm.EpochCompleted += (s, args) => Dispatcher.Invoke(new Action(() => Label_Similar.Content = string.Format("{0:P2}", algorithm.Fitness)));
algorithm.EpochCompleted += (s, args) => Dispatcher.Invoke(new Action(() =>
{
if (algorithm.Epoch % 1000 == 1)
{
Painter.Paint(algorithm.Population.CalculateFittest()).Save(string.Format("img/Epoch_{0}K.bmp", algorithm.Epoch / 1000));
}
}));
algorithm.EpochCompleted += (s, args) => Dispatcher.Invoke(new Action(() => SaveBitmap(Painter.Paint(algorithm.Population.CalculateFittest()), MainImage)));
algorithm.EpochCompleted += (s, args) => Dispatcher.Invoke(new Action(() => Label_Status.Content = string.Format("Epoch: {0, 11}\nStagnation: {2, 11}\nFitness: {1,11:N6}\nRuntime: {3:d\\.hh\\:mm\\:ss}", algorithm.Epoch, algorithm.Fitness, algorithm.StagnationCount, algorithm.TimeRan)));
algorithm.EpochCompleted += (s, args) => Dispatcher.Invoke(new Action(() =>
{
if (saveWhileRunning)
{
if (algorithm.Epoch % 1000 == 0)
{
SaveSerializedCanvas(algorithm);
}
if (algorithm.Epoch % 100000 == 0)
{
SaveSerializedCanvas(algorithm, string.Format("saves/Epoch_{0}k", algorithm.Epoch));
}
}
}));
algorithm.AlgorithmCompleted += (s, args) => Dispatcher.Invoke(new Action(() =>
{
Painter.Paint(algorithm.Population.CalculateFittest()).Save(string.Format("img/Epoch_{0}K{1}.bmp", algorithm.Epoch / 1000, algorithm.Epoch % 1000));
if (saveWhileRunning)
{
SaveSerializedCanvas(algorithm);
}
Button_Run.Content = "Run";
Label_Status.Content += "\nFinished";
ctc = null;
}));
await algorithm.RunAsync(() => StopCondition(algorithm, maxRuntime, maxEpochs, maxStagnation, minFitness), ctc.Token);
}
catch (OperationCanceledException)
{
Painter.Paint(algorithm.Population.CalculateFittest()).Save(string.Format("img/Epoch_{0}K{1}.bmp", algorithm.Epoch / 1000, algorithm.Epoch % 1000));
if (saveWhileRunning)
{
SaveSerializedCanvas(algorithm);
}
algorithm.Matlabify();
Label_Status.Content += "\nCanceled";
}
}
}