static void Main(string[] args)
{
var data = HealthInfo.DeserialiseData(HEART_DATASET);
var testData = new List <HealthInfo>();
while (testData.Count < 25)
{
int index = _randEngine.Next(data.Count - 1);
testData.Add(data[index]);
data.RemoveAt(index);
}
var agent = new Boost <HealthInfo>();
agent.Add(new Agent <HealthInfo>(100));
int Nbsucces = 0;
int bestNbSuccess = 0;
int run = 0;
int nbRunSinceLastSuccess = 0;
int nbAgent = 1;
while ((double)Nbsucces / data.Count < 0.85)
{
Nbsucces = 0;
for (int i = 0; i < 1; i++)
{
agent.Fit(data);
}
foreach (var item in data)
{
if ((agent.MakePrediction(item) > 0) == (item.Target > 0))
{
Nbsucces++;
}
}
if (Nbsucces > bestNbSuccess)
{
Console.WriteLine($"Last best result at : {DateTime.Now.ToString()}");
Console.WriteLine($"The agent made {Nbsucces} prediction succefully on a total of {data.Count} at run {run}.");
Console.WriteLine($"This is accurate at {(double)Nbsucces / data.Count}");
Console.WriteLine($"The Agent spec");
Console.WriteLine(agent.ToString());
Console.WriteLine("Result");
Console.WriteLine($"{data[0].Header()} Prediction");
int lineToBePrintedCount = 303;
for (int i = 0; i < data.Count; i++)
{
if (lineToBePrintedCount > 0 && (_randEngine.Next() % 303 == 0 || data.Count - i < lineToBePrintedCount))
{
Console.WriteLine($"{data[i].ToString()} {agent.MakePrediction(data[i])}");
lineToBePrintedCount--;
}
}
Console.WriteLine($"This is accurate at {(double)Nbsucces / data.Count}");
bestNbSuccess = Nbsucces;
nbRunSinceLastSuccess = -1;
}
nbRunSinceLastSuccess++;
if (nbRunSinceLastSuccess >= 500)
{
if (_randEngine.Next() % 2 == 0)
{
agent = new Boost <HealthInfo>();
agent.Add(new Agent <HealthInfo>(100));
Console.WriteLine($"New Agent at run {run}");
}
nbRunSinceLastSuccess = 0;
nbAgent++;
}
run++;
}
Console.WriteLine($"The tranning is complete after {run} run and {nbAgent} agents");
int dataEvaluation = 0;
foreach (var d in testData)
{
if ((agent.MakePrediction(d) > 0) == (d.Target > 0))
{
dataEvaluation++;
}
}
Console.WriteLine($"Final score is {(double)dataEvaluation/testData.Count} or {dataEvaluation} / {testData.Count}");
Console.ReadLine();
}
private double run(double[][] inputs, int[] outputs, double[] sampleWeights)
{
double error = 0;
double weightSum = 0;
int[] actualOutputs = new int[outputs.Length];
do
{
// Create and train a classifier
TModel model = Creation(sampleWeights);
if (model == null)
{
break;
}
// Determine its current accuracy
for (int i = 0; i < actualOutputs.Length; i++)
{
actualOutputs[i] = model.Compute(inputs[i]);
}
error = 0;
for (int i = 0; i < actualOutputs.Length; i++)
{
if (actualOutputs[i] != outputs[i])
{
error += sampleWeights[i];
}
}
if (error >= threshold)
{
break;
}
// AdaBoost
// double w = 0.5 * System.Math.Log((1.0 - error) / error);
double w = error / (1 - error);
double sum = 0;
for (int i = 0; i < sampleWeights.Length; i++)
{
if (actualOutputs[i] != outputs[i])
{
sum += sampleWeights[i] *= w;
}
else
{
sum += sampleWeights[i];
}
}
// Update sample weights
for (int i = 0; i < sampleWeights.Length; i++)
{
sampleWeights[i] /= sum;
}
classifier.Add(w, model);
weightSum += w;
convergence.NewValue = error;
} while (!convergence.HasConverged);
// Normalize weights for confidence calculation
for (int i = 0; i < classifier.Models.Count; i++)
{
classifier.Models[i].Weight /= weightSum;
}
return(ComputeError(inputs, outputs));
}