public void UseGAN(int generatedNumber)
{
var balls = gameSystem.GetBallsStatus();
float[,] conditionsAll = new float[generatedNumber, balls.Count * 3];
for (int iball = 0; iball < balls.Count; ++iball)
{
for (int i = 0; i < generatedNumber; ++i)
{
conditionsAll[i, iball *3] = balls[iball].x;
conditionsAll[i, iball * 3 + 1] = balls[iball].y;
conditionsAll[i, iball * 3 + 2] = balls[iball].z;
}
}
float[,] generated = (float[, ])modelRef.GenerateBatch(conditionsAll, MathUtils.GenerateWhiteNoise(generatedNumber, -1, 1, modelRef.inputNoiseShape));
dataPlane.RemovePointsOfType(1);
dataPlane.RemovePointsOfType(0);
for (int i = 0; i < generatedNumber; ++i)
{
dataPlane.AddDatapoint(new Vector2(generated[i, 0] / 2, generated[i, 1]) / 2, 1);
}
dataPlane.AddDatapoint(new Vector2(0.5f, 0.5f), 0);
dataPlane.AddDatapoint(new Vector2(-0.5f, -0.5f), 0);
dataPlane.AddDatapoint(new Vector2(0.5f, -0.5f), 0);
dataPlane.AddDatapoint(new Vector2(-0.5f, 0.5f), 0);
dataPlane.AddDatapoint(new Vector2(0, 0), 0);
}
public void UseGAN(int generatedNumber)
{
float[,] generated = (float[, ])modelRef.GenerateBatch(null, MathUtils.GenerateWhiteNoise(generatedNumber, -1f, 1f, modelRef.inputNoiseShape));
dataPlane.RemovePointsOfType(1);
for (int i = 0; i < generatedNumber; ++i)
{
dataPlane.AddDatapoint(new Vector2(generated[i, 0], generated[i, 1]), 1);
}
}
public float TrainDiscriminatorBatch(int batchSize)
{
//fetch data from data buffer
var fetchesForFake = new List <ValueTuple <string, int, string> >();
var fetchesForReal = new List <ValueTuple <string, int, string> >();
fetchesForReal.Add(new ValueTuple <string, int, string>("Target", 0, "Target"));
if (ganReference.HasConditionInput)
{
fetchesForFake.Add(new ValueTuple <string, int, string>("Condition", 0, "Condition"));
fetchesForReal.Add(new ValueTuple <string, int, string>("Condition", 0, "Condition"));
}
var samplesForFake = dataBuffer.RandomSample(batchSize, fetchesForFake.ToArray());
var samplesForReal = dataBuffer.RandomSample(batchSize, fetchesForReal.ToArray());
Array conditionsForFake = samplesForFake.ContainsKey("Condition") ? samplesForFake["Condition"] : null;
Array conditionsForReal = samplesForReal.ContainsKey("Condition") ? samplesForReal["Condition"] : null;
Array targetsForReal = samplesForReal["Target"];
//generate noise
Array noise = null;
if (ganReference.HasNoiseInput)
{
noise = MathUtils.GenerateWhiteNoise(batchSize, -1, 1, ganReference.inputNoiseShape);
}
//genrate fake data
Array generatedFake = ganReference.GenerateBatch(conditionsForFake, noise);
//genreate labels
float[,] fakeLabels = new float[batchSize, 1];
float[,] realLabels = new float[batchSize, 1];
for (int i = 0; i < batchSize; ++i)
{
fakeLabels[i, 0] = 0;
realLabels[i, 0] = 1;
}
//train with fake labels
float l1 = ganReference.TrainDiscriminatorBatch(generatedFake, fakeLabels, conditionsForFake);
float l2 = ganReference.TrainDiscriminatorBatch(targetsForReal, realLabels, conditionsForReal);
return((l1 + l2) / 2);
}
