public void calcActivDerivZ()
{
for (int i = 1; i < layers.Count; i++)
{
layers[i].activDerivZ = NetworkMath.activationFunctionDeriv(layers[i].z);
}
}
/// <summary>
/// calculates the value of a single layer's neuron values
/// </summary>
/// <param name="preLayer"></param>
/// <param name="curLayer"></param>
public void calcLayer(Layer preLayer, Layer curLayer)
{
double[] temp = NetworkMath.multiply(curLayer.weight, preLayer.neuron); //multiplies the weights and the value of the previous layers neurons
curLayer.z = NetworkMath.add(curLayer.bias, temp); //adds the appropriate biases to the product of the weights and neurons
curLayer.neuron = NetworkMath.activationFunction(curLayer.z); //applies an activation function on the previous sum
}
public void updateWeightsAndBiases(double rate, double numTests)
{
for (int layerNum = 1; layerNum < layers.Count; layerNum++) //loops through each layers starting with the first hidden layer
{ //WEIGHT UPDATE
double[,] sumWeight = new double[layers[layerNum].derivWeight[0].GetLength(0), layers[layerNum].derivWeight[0].GetLength(1)]; //holds sum for the weight derivatives
for (int i = 0; i < sumWeight.GetLength(0); i++) //this goes through the first index
{
for (int j = 0; j < sumWeight.GetLength(1); j++) //this goes throguh the second index
{
sumWeight[i, j] = 0; //sets every value to zero
}
}
foreach (double[,] d in layers[layerNum].derivWeight) //loops through each derivative in the list and adds it to the sum
{
sumWeight = NetworkMath.add(sumWeight, d); //adds the derivatives to the sum
}
for (int i = 0; i < sumWeight.GetLength(0); i++) //this goes through the first index
{
for (int j = 0; j < sumWeight.GetLength(1); j++) //this goes throguh the second index
{
sumWeight[i, j] /= numTests; //divides the sum by number of tests to calc average
layers[layerNum].weight[i, j] -= sumWeight[i, j] * rate; //the actual update of each weight
}
}
//BIAS UPDATE
double[] sumBias = new double[layers[layerNum].derivBias[0].GetLength(0)]; //creates a sum array for the bias derivatives
for (int i = 0; i < sumBias.Length; i++) //loops through the bias sum array
{
sumBias[i] = 0; //sets every element in the array to zero
}
foreach (double[] d in layers[layerNum].derivBias) //loops through all the derivatives from every test
{
sumBias = NetworkMath.add(sumBias, d); //adds the derivatives to the sum
}
for (int i = 0; i < sumBias.Length; i++) //loops through each element of the sum array
{
sumBias[i] /= numTests; //divides the sum by the number of tests to calculate the average
layers[layerNum].bias[i] -= sumBias[i] * rate; //the actual update of each bias
}
layers[layerNum].derivWeight.Clear(); //removes the list of derivatives used in this update
layers[layerNum].derivBias.Clear();
layers[layerNum].derivNeuron.Clear();
}
}
/// <summary>
/// calculates all neuron values in the network
/// </summary>
/// <param name="input"></param>
public void calcNetwork(double[] input)
{
layers[0].neuron = input; //sets the correct values of the nodes of the input layer
for (int i = 1; i < (layers.Count); i++) //loops through each layer (starting with the first hidden layer)
{
//calcLayer(layers[i - 1], layers[i]);
double[] temp = NetworkMath.multiply(layers[i].weight, layers[i - 1].neuron); //multiplies the weights and the value of the previous layers neurons
layers[i].z = NetworkMath.add(layers[i].bias, temp); //adds the appropriate biases to the product of the weights and neurons
layers[i].neuron = NetworkMath.activationFunction(layers[i].z); //applies an activation function on the previous sum
}
}
