public NeuralTrainSet(NeuralNetwork neuralNetwork, float[] input, float[] target)
{
NeuralNetwork = neuralNetwork;
Layers = new NeuralTrainLayer[neuralNetwork.Layers.Length];
for (int i = 0; i < Layers.Length; i++)
{
if (i == 0)
{
Layers[i] = new NeuralTrainLayer(neuralNetwork.Layers[i], input);
}
else
{
Layers[i] = new NeuralTrainLayer(neuralNetwork.Layers[i], Layers[i - 1].Output);
}
}
output = Layers.Last().Output;
this.target = target;
bool firstLayer = true;
for (int i = Layers.Length - 1; i >= 0; i--)
{
NeuralTrainLayer layer = Layers[i];
target = layer.CalculateBackpropagateValues(target, firstLayer);
firstLayer = false;
}
}
internal void Apply(NeuralTrainLayer neuralTrainLayer, float learningRate)
{
if (neuralTrainLayer.TrainingCount == 0)
{
return;
}
float multiplyer = learningRate / neuralTrainLayer.TrainingCount;
float[] bShift = neuralTrainLayer.DCDB.Multiply(multiplyer);
float[] wShift = neuralTrainLayer.DCDW.Multiply(multiplyer);
Bias = Bias.Minus(bShift);
Weight = Weight.Minus(wShift);
}
public NeuralTrainSet(NeuralTrainSet a, NeuralTrainSet b)
{
if (a.NeuralNetwork != b.NeuralNetwork)
{
throw new Exception("Cannot blend sets that belong to a different network.");
}
NeuralNetwork = a.NeuralNetwork;
int layerCount = a.Layers.Length;
Layers = new NeuralTrainLayer[layerCount];
for (int i = 0; i < layerCount; i++)
{
Layers[i] = new NeuralTrainLayer(a.Layers[i], b.Layers[i]);
}
}
public NeuralTrainLayer(NeuralTrainLayer a, NeuralTrainLayer b)
{
DCDB = a.DCDB.Plus(b.DCDB);
DCDW = a.DCDW.Plus(b.DCDW);
TrainingCount = a.TrainingCount + b.TrainingCount;
}
