private static void displayConfusionData(ConfusionData cd)
{
Console.WriteLine("Confusion data from learning set : ");
int maxLengthName = cd.UniqueCat.Aggregate("", (max, cur) => max.Length > cur.Length ? max : cur).Length;
for (int i = 0; i < cd.data.Count; i++)
{
Console.Write(cd.UniqueCat[i]);
for (int k = 0; k < maxLengthName - cd.UniqueCat[i].Length; k++)
{
Console.Write(" ");
}
Console.Write(" | ");
int wellPredicted = cd.data[i][i];
int badPredicted = 0;
for (int j = 0; j < cd.data[0].Count; j++)
{
Console.Write(cd.data[i][j].ToString("000") + " | ");
if (j != i)
{
badPredicted += cd.data[i][j];
}
}
if (wellPredicted + badPredicted == 0)
{
Console.WriteLine();
}
else
{
Console.WriteLine("(" + (wellPredicted * 100 / (wellPredicted + badPredicted)).ToString("00.0") + "%)");
}
}
}
static void Main(String[] args)
{
//Initialize a new motor
RandomForest RFmotor = new RandomForest();
//Enable or disable suspect-detection. Elements with less than 30% of prediction will be classed as "suspect"
RFmotor.enableOtherClassCreation(30, "suspect");
RFmotor.disableOtherClassCreation();
//Do something when a tree is created
RFmotor._TreeCreated += RFmotor__TreeCreated;
do
{
Console.Clear();
//Initialize random data to predict
generateDataToPredict(10);
//Initialize random learning set
generateLearningData(50);
//Evaluate the learning (optional, usefull when creating a new learning)
modulo = 100;
Console.Write("Training : ");
ConfusionData cd = RFmotor.evaluateCrossValidation(5, learningData, learningClasses, 500, 100);
Console.WriteLine();
displayConfusionData(cd);
Console.WriteLine();
//Train the motor with learning data
modulo = 10;
Console.Write("Training : ");
RFmotor.train(learningData, learningClasses, 500, 100);
Console.WriteLine();
//Predict sample's class
Console.WriteLine("Predicted data : ");
for (int i = 0; i < dataToPredict.GetLength(0); i++)
{
double[] row = new double[dataToPredict.GetLength(1)];
for (int j = 0; j < dataToPredict.GetLength(1); j++)
{
row[j] = dataToPredict[i, j];
}
Prediction predicted = RFmotor.predict(row);
displayPrediction(predicted, row);
}
} while (Console.ReadKey().Key != ConsoleKey.Escape);
Console.ReadLine();
}
public ConfusionData evaluateCrossValidation(int itterations, double[,] samples, String[] samplesCategories, int forestSize, int percentParameters = 0)
{
int error = 0;
ConfusionData cd = new ConfusionData();
var categories = samplesCategories.Distinct().ToList();
if (this._createOtherClass)
{
categories.Add(this._otherClassName);
}
if (!PredictCuts)
{
categories.Add(CUT_CLASS_NAME);
}
cd.UniqueCat = categories;
//On crée deux sous tableaux constitués d'échantillons tirés aléatoirement
double[][] samplesArray = ToJaggedArray <double>(samples);
for (int itt = 0; itt < itterations; itt++)
{
//On crée des index aléatoires pour créer les tableaux
int[] indexes = getRandomSamplesOutOfTrainingSet(samplesArray.GetLength(0), (int)(samplesArray.Length * 0.6));
int sizeA = (int)(samplesArray.Length * 0.6);
int sizeB = (int)(samplesArray.Length * 0.4);
while (sizeA + sizeB != samplesArray.Length)
{
sizeB++;
}
double[][] A = new double[sizeA][];
double[][] B = new double[sizeB][];
int j = 0;
int k = 0;
String[] ACat = new string[sizeA];
String[] BCat = new string[sizeB];
for (int i = 0; i < samplesArray.Length; i++)
{
if (indexes.Contains(i))
{
ACat[j] = samplesCategories[i];
A[j] = samplesArray[i];
j++;
}
else
{
BCat[k] = samplesCategories[i];
B[k] = samplesArray[i];
k++;
}
}
//Les deux tableaux sont crées
train(To2DArray <double>(A), ACat, forestSize, percentParameters);
for (int s = 0; s < B.Length; s++)
{
Prediction pred = this.predict(B[s]);
if (pred.predictedClass != BCat[s])
{
error++;
}
cd.data[cd.UniqueCat.IndexOf(BCat[s])][cd.UniqueCat.IndexOf(pred.predictedClass)]++;
if (pred.predictedClass != this.OtherClassName)
{
var percentage = cd.pourcentages.First(p => p.categorie == pred.predictedClass);
percentage.predicted += pred.treePredictedClass;
percentage.total += pred.totalTrees;
}
}
train(To2DArray <double>(B), BCat, forestSize, percentParameters);
for (int s = 0; s < A.Length; s++)
{
Prediction pred = this.predict(A[s]);
if (pred.predictedClass != ACat[s])
{
error++;
}
cd.data[cd.UniqueCat.IndexOf(ACat[s])][cd.UniqueCat.IndexOf(pred.predictedClass)]++;
if (pred.predictedClass != this.OtherClassName)
{
var percentage = cd.pourcentages.First(p => p.categorie == pred.predictedClass);
percentage.predicted += pred.treePredictedClass;
percentage.total += pred.totalTrees;
}
}
}
// ((double)error / (samples.GetLength(0))) / (double)itterations * 100.0;
return(cd);
}