/**
* \brief Load the generation
* \details load the complete population and the best individual in two files. The path depends on GeneratePath()
*/
public void LoadGeneration(string path, int generationIndex, float[,] floatArr, int rowIndex)
{
UIO.LoadFloatArray(path + generationIndex, "/population.gene", floatArr);
Matrix <float> partialPopulation = Matrix <float> .Build.DenseOfArray(floatArr);
population.SetSubMatrix(rowIndex, 0, partialPopulation);
}
public void LoadGeneration(int generationIndex, float[,] floatArr)
{
UIO.LoadFloatArray(GeneratePath() + generationIndex, "/population.gene", floatArr);
population = Matrix <float> .Build.DenseOfArray(floatArr);
//population.SetSubMatrix(rowIndex, 0, Matrix<float>.Build.DenseOfArray(floatArr));
}
public override void CloseSessionWriter()
{
if (save)
{
UIO.CloseStreamWriter(writerSession);
}
}
public override void Build()
{
if (save)
{
writerEnv = UIO.CreateStreamWriter(GeneratePath(task, true), "GeneSessionResults.csv", false);
UIO.WriteLine(writerEnv, "Generation;Angle Random Rotation;Wind");
}
Debug.Log("Build DroneSession");
signal = new ControlSignal();
tsignal = new ThrustSignal();
tmpBuildCustomWeights = new List <Matrix <float> >();
externalEvaluations = Vector <float> .Build.Dense(populationSize);
//Debug.Log(fromTask);
taskObject = (DroneTask)Activator.CreateInstance(Type.GetType("Lexmou.MachineLearning.Drone" + task), rndGenerator, fromTask);
//Debug.Log(taskObject.fromTask);
dronePopulation = new GameObject[populationSize];
droneRigid = new Rigidbody[populationSize];
targetPosition = new Vector3[populationSize];
mlpPopulation = new MultiLayerMathsNet[populationSize];
gene = new Genetic(seed, rndGenerator, populationSize, taskObject.individualSize, initialValueWeights, mutationRate, randomIndividualsRate, bestIndividualsRate, emptyRate, GeneratePath(task, false), save);
if (loadGeneration != 0)
{
float[,] floatArr = new float[taskObject.individualSize - taskObject.rowIndex, populationSize];
Debug.Log(taskObject.fromTask);
//Debug.Log(taskObject.individualSize);
gene.LoadGeneration(GeneratePath(taskObject.fromTask, true), loadGeneration, floatArr, taskObject.rowIndex);
gene.generation = loadGeneration;
}
}
public override void OnDestroy()
{
if (save)
{
Debug.Log("Destroy Genetic Writer");
UIO.CloseStreamWriter(gene.writer);
UIO.CloseStreamWriter(writerEnv);
}
}
/**
*
*/
public Genetic(int seed, SystemRandomSource rndGenerator, int populationSize, int individualSize, float initialValueWeights, float mutationRate = 0.1f, float randomIndividualsRate = 0.05f, float bestIndividualsRate = 0.05f, float emptyRate = 0.0f, string path = "", bool save = true)
{
this.save = save;
tmp = Matrix <float> .Build.Dense(individualSize, populationSize);
this.path = path;
this.bestScore = 0.0f;
this.generation = 1;
this.seed = seed;
this.initialValueWeights = initialValueWeights;
if (rndGenerator != null)
{
this.rndGenerator = rndGenerator;
}
else
{
this.rndGenerator = new SystemRandomSource(seed);
}
distribution = new ContinuousUniform(-initialValueWeights, initialValueWeights, rndGenerator);
subDistribution = new ContinuousUniform(-0.01, 0.01, rndGenerator);
drawDistribution = new ContinuousUniform(0.0f, 1.0f, rndGenerator);
this.populationSize = populationSize;
this.individualSize = individualSize;
this.population = Matrix <float> .Build.Random(individualSize, populationSize, distribution);
for (int i = 0; i < populationSize; i++)
{
distEmpty = Combinatorics.GenerateCombination(individualSize, Mathf.RoundToInt(emptyRate * individualSize), rndGenerator);
for (int j = 0; j < individualSize; j++)
{
if (distEmpty[j])
{
population[j, i] = 0.000f;
}
}
}
this.evaluations = Vector <float> .Build.Dense(populationSize);
this.sumEvaluations = 0.0f;
this.mutationRate = mutationRate;
this.randomIndividualsRate = randomIndividualsRate;
this.bestIndividualsRate = bestIndividualsRate;
this.emptyRate = emptyRate;
if (save)
{
writer = UIO.CreateStreamWriter(GeneratePath(), "GeneticResults.csv", false);
UIO.WriteLine(writer, "Generation;Best;Mean;Std Deviation;Median");
}
indexPermutation = Combinatorics.GeneratePermutation(populationSize);
}
public override void RunEachIntervalUpdate()
{
generationInfos = gene.Run(externalEvaluations, theoricBestScore, intervalSave);
hudManager.UpdateTextLayout("Generation", generationInfos);
theoricBestScore = 0;
if (save)
{
UIO.WriteLine(writerEnv, gene.generation + ";" + taskObject.AngleRandomRotation + ";" + taskObject.WindStrength);
}
}
public override void SetParametersFromCommandLine()
{
UIO.CheckBeforeReplaceCommandLineArguments(this, "save");
UIO.CheckBeforeReplaceCommandLineArguments(this, "task");
UIO.CheckBeforeReplaceCommandLineArguments(this, "fromTask");
UIO.CheckBeforeReplaceCommandLineArguments(this, "seed");
UIO.CheckBeforeReplaceCommandLineArguments(this, "intervalSave");
UIO.CheckBeforeReplaceCommandLineArguments(this, "loadGeneration");
UIO.CheckBeforeReplaceCommandLineArguments(this, "timeScale");
}
private void btnOK_Click(object sender, EventArgs e)
{
var strategyFilter = new StrategyFilter();
strategyFilter.Days = Day;
strategyFilter.Sensors = Sensors;
var str = UXmlConvert.GetString(strategyFilter);
var filename = UWorkspace.GetStrategyXmlName(_wksdef);
UIO.SaveFile(str, filename);
DialogResult = DialogResult.OK;
Close();
}
public override void BuildSessionWriter()
{
if (save)
{
writerSession = UIO.CreateStreamWriter(GeneratePath(task, true), "DroneSessionParams.csv", false);
UIO.WriteLine(writerSession, "Task : " + task + ";" +
"Seed : " + seed + ";" +
"Population Size : " + populationSize + ";" +
"Individual Size : " + taskObject.individualSize + ";" +
"Mutation Rate : " + mutationRate + ";" +
"Random Rate : " + randomIndividualsRate + ";" +
"Best Individual Rate : " + bestIndividualsRate + ";" +
"Empty Coeff Individual Rate : " + emptyRate + " (" + Mathf.RoundToInt(emptyRate * taskObject.individualSize) + ");" +
"MLP layers : [" + string.Join("-", taskObject.shapes.Select(x => x.ToString()).ToArray()) + "] ; Intitial Value Weights : [" + -initialValueWeights + "," + initialValueWeights + "]");
}
}
public string Run(Vector <float> externalEvaluations, float theoricBestScore, int intervalSave = 0)
{
string generationInfos;
Evaluation(externalEvaluations, "max");
if (save)
{
UIO.WriteLine(writer, generation + ";" + bestScore + ";" + mean + ";" + stdDeviation + ";" + median);
}
if (intervalSave > 0 && generation % intervalSave == 0)
{
Debug.Log("Save in Progress");
SaveGeneration();
}
generationInfos = generation + " | Best Score : " + bestScore + "/" + theoricBestScore + " | Index : " + bestIndividualIndex
+ "\r\n" + "Score Mean : " + (sumEvaluations / populationSize);
Selection();
CrossOver();
Mutation();
generation += 1;
return(generationInfos);
}
public static void LoadBest(string path, int generationIndex, float[] floatArr)
{
UIO.LoadFloatArray(path + generationIndex, "/best.gene", floatArr);
}
/**
* \brief Load the best individual
* \details load the complete population and the best individual in two files. The path depends on GeneratePath()
*/
public void LoadBest(int generationIndex, float[] floatArr)
{
UIO.LoadFloatArray(GeneratePath() + generationIndex, "/best.gene", floatArr);
}
/**
* \brief Save the generation
* \details Save the complete population and the best individual in two files. The path depends on GeneratePath()
*/
public void SaveGeneration()
{
UIO.SaveFloatArray(GeneratePath() + generation, "population.gene", population.ToArray());
UIO.SaveFloatArray(GeneratePath() + generation, "best.gene", GetBestIndividual().ToArray());
}