//основной конструктор, размер входного слоя, размеры скрытых слоев и тд
public NeuralNetwork(int inSize, List <int> hiddSizes,
int outSize, floatFun actFun, bool isB = false, Pair <float> r = null)
{
if (isB)
{
inSize++;
}
inputLayer = new Layer(inSize, 0, ActivationFun.linearFunction);
int prevSize = inSize;
foreach (int size in hiddSizes)
{
int sizeLayer = size;
if (isB)
{
sizeLayer++;
}
hiddenLayers.Add(new Layer(sizeLayer, prevSize, actFun));
prevSize = sizeLayer;
}
outputLayer = new Layer(outSize, prevSize, actFun);
if (r != null)
{
range = r;
}
layerSize = 2 + hiddenLayers.Count;
isBias = isB;
}
//Изменение настроек НН
private void changeNNSettings(object sender, EventArgs e)
{
float V;
float.TryParse(textBoxLearnRate.Text, out V);
Information.NN.LearnSpeed = V;
string funStr = comboBoxFunAct.Text;
if (funStr == "")
{
funStr = "Linear";
}
floatFun F = ActivationFun.sigmoidFunction;
if (funStr == "Linear")
{
F = ActivationFun.linearFunction;
}
if (funStr == "Sigmoid")
{
F = ActivationFun.sigmoidFunction;
}
if (funStr == "Tangent")
{
F = ActivationFun.tangentFunction;
}
Information.NN.SetActFun(F);
float M;
float.TryParse(textBoxMoment.Text, out M);
Information.NN.LearnMoment = M;
console.AppendText("fun = " + funStr + " V = " + Information.NN.LearnSpeed);
console.AppendText(" M = " + Information.NN.LearnMoment + "\r\n");
}
public void SetActFun(floatFun dF)
{
foreach (var l in hiddenLayers)
{
l.setActivationFunction(dF);
}
outputLayer.setActivationFunction(dF);
}
//по умолчанию
public NeuralNetwork(int size, floatFun actFun, bool isB = false, Pair <float> r = null)
{
int sizeOut = size;
if (isB)
{
size++;
}
inputLayer = new Layer(size, 0, ActivationFun.linearFunction);
hiddenLayers.Add(new Layer(size, size, actFun));
outputLayer = new Layer(sizeOut, size, actFun);
if (r != null)
{
range = r;
}
layerSize = 3;
isBias = isB;
}
//конструктор с инициализацией весов(выгрузка из файла)
public NeuralNetwork(int inSize, List <int> hiddSizes, int outSize,
floatFun actFun, List <List <float> > weights, bool isB = false)
{
isBias = isB;
inputLayer = new Layer(inSize, 0, ActivationFun.linearFunction);
int weightIndex = 1;
int prevSize = inSize;
foreach (int sizeLayer in hiddSizes)
{
hiddenLayers.Add(new Layer(sizeLayer, prevSize, actFun));
prevSize = sizeLayer;
hiddenLayers[weightIndex - 1].Weights = weights[weightIndex];
weightIndex++;
}
outputLayer = new Layer(outSize, prevSize, actFun);
outputLayer.Weights = weights[weightIndex];
layerSize = 2 + hiddenLayers.Count;
}
//конструктор слоя
public Layer(int size, int sizePrev, floatFun actFun, float a = -0.8f, float b = 0.8f)
{
sizeNeurons = size;
outputs = new List <float>(sizeNeurons);
activationFun = actFun;
Thread.Sleep(TimeSpan.FromMilliseconds(100));
Random rand = new Random(DateTime.Now.Millisecond);
if (sizePrev > 0)
{
sizeWeights = size * sizePrev;
weights = new List <float>(sizeWeights);
//deltaWeights = new List<float>(sizeWeights);
for (int i = 0; i < sizeWeights; i++)
{
weights.Add(a + (float)rand.NextDouble() * (b - a));
//deltaWeights.Add(0);
}
}
}
public void setActivationFunction(floatFun fun)
{
activationFun = fun;
}
