public Layer(LayerCreationInfo layerCreationInfo, bool IsNetworkUsingBias, LayerType type, double minWeight, double maxWeight)
{
LayerNo = layerCreationInfo.LayerNo;
PreviousLayerNeuronsCount = layerCreationInfo.PreviousLayerNeuronsCount;
LayerActivationFunction = layerCreationInfo.LayerActivationFunction;
this.Type = type;
this.MinWeight = minWeight;
this.MaxWeight = maxWeight;
Neurons = new List <Neuron>();
for (int i = 0; i < layerCreationInfo.HowManyNeuronsPerLayer; i++)
{
Neurons.Add(new Neuron(
LayerNo == 0 ? 1 : PreviousLayerNeuronsCount,
layerCreationInfo.LayerNo,
false,
layerCreationInfo.LayerActivationFunction,
MinWeight,
MaxWeight));
}
if (IsNetworkUsingBias && type != LayerType.OUTPUT)
{
Neurons.Last().IsBias = true;
}
}
public static double Derivative(double x, NeuralNetwork.ActivationFunction activationFunction)
{
switch (activationFunction)
{
case NeuralNetwork.ActivationFunction.SIGMOID:
return(DerivativeSigmoid(x));
case NeuralNetwork.ActivationFunction.TANH:
return(DerivativeTanh(x));
default:
throw new Exception("Counting derivative for not implemented case");
}
}
public static double ActivationFunction(double value, NeuralNetwork.ActivationFunction activationFunction)
{
double result = 0.0;
switch (activationFunction)
{
case NeuralNetwork.ActivationFunction.TANH:
// someone says: "approximation is correct to 30 decimals", so:
if (value < -20.0)
{
result = -1.0;
}
else if (value > 20.0)
{
result = 1.0;
}
else
{
result = Math.Tanh(value);
}
break;
case NeuralNetwork.ActivationFunction.SIGMOID:
result = 1.0 / (1.0 + Math.Exp(-value));
break;
case NeuralNetwork.ActivationFunction.SINUSOID:
result = Math.Sin(value);
break;
case NeuralNetwork.ActivationFunction.BINARY_STEP:
result = value < 0 ? 0 : 1;
break;
case NeuralNetwork.ActivationFunction.NO_FUNCTION:
result = value;
break;
default:
throw new Exception("Function " + activationFunction.ToString() + "not implemented in CountOutput!");
}
return(result);
}
public Neuron(int inputsCount, int layerNo, bool isBias, NeuralNetwork.ActivationFunction activationFunction, double biasValue = -1.0)
{
IsBias = isBias;
BiasValue = biasValue;
LayerNo = layerNo;
this.ActivationFunction = activationFunction;
Inputs = new List <Input>();
PreviousInputs = new List <Input>();
for (int inputNo = 0; inputNo < inputsCount; inputNo++)
{
Inputs.Add(new Input(inputNo));
PreviousInputs.Add(new Input(inputNo));
}
ResetNeuron();
}
private static void LearnNetworkOption()
{
long DEFAULT_MAX_EPOCH_NO = 10000;
long DEFAULT_COLLECT_SPARSE_HISTORY_EVERY = 1000;
double DEFAULT_MIN_ERR_PERCENTAGE_TO_STOP = 0.99;
int inputCount = 2;
int outputCount = 1;
List <int> defaultHiddenLayersNeuronsCount = new List <int> {
20, 20
};
List <int> hiddenLayersNeuronsCount = new List <int>();
String filename = "data.csv";
String str = null;
bool networkUseBias = true;
NeuralNetwork.ActivationFunction activationFunction = NeuralNetwork.ActivationFunction.NO_FUNCTION;
double learningRatio = 0.1;
double momentum = 0.7;
double minErrorPercentageToStop = DEFAULT_MIN_ERR_PERCENTAGE_TO_STOP;
long maxEpochNo = 0;
long writeOnConsoleEveryEpoch = 1000;
bool randomOrderOfLearningData = true;
List <LearnSample> learningData = new List <LearnSample>();
long collectSparseHistoryEvery = 1000;
bool dataLoopRunning = true;
List <Layer> layers = new List <Layer>();
#region Data preparation
while (dataLoopRunning)
{
learningRatio = 0.1;
//minErrorPercentageToStop = DEFAULT_MIN_ERR_PERCENTAGE_TO_STOP;
filename = "data.csv";
Console.Clear();
Console.WriteLine("Creating and learning Neural Networks");
bool ok = false;
while (!ok)
{
Console.Write("Input neurons count: ");
str = TAKE_PARAMETERS_FROM_USER ? Console.ReadLine() : inputCount.ToString();
ok = int.TryParse(str, out inputCount);
if (ok)
{
ok = (inputCount > 0);
}
if (!ok)
{
Console.WriteLine("This value is not allowed.");
}
}
ok = false;
while (!ok)
{
Console.Write("Hidden layers neurons count, coma sepparated: [20, 20] ");
str = TAKE_PARAMETERS_FROM_USER ? Console.ReadLine() : "";
ok = true;
if (str.Length > 0)
{
String[] tab = str.Split(',');
for (int i = 0; i < tab.Length; i++)
{
int val = 0;
ok = int.TryParse(tab[i], out val);
if (ok)
{
ok = (val > 0);
}
if (!ok)
{
break;
}
hiddenLayersNeuronsCount.Add(val);
}
}
else
{
hiddenLayersNeuronsCount = defaultHiddenLayersNeuronsCount;
}
if (!ok)
{
Console.WriteLine("This value is not allowed.");
}
}
ok = false;
while (!ok)
{
Console.Write("Output neurons count: ");
str = TAKE_PARAMETERS_FROM_USER ? Console.ReadLine() : outputCount.ToString();
ok = int.TryParse(str, out outputCount);
if (ok)
{
ok = (outputCount > 0);
}
if (!ok)
{
Console.WriteLine("This value is not allowed.");
}
}
ok = false;
while (!ok)
{
Console.Write("Activation function of neurons Sigmoid/Tanh: [S/t] ");
str = TAKE_PARAMETERS_FROM_USER ? Console.ReadLine() : "";
str = str.ToLower();
if (str.Length == 0)
{
str = "s";
}
ok = (str == "s" || str == "t");
if (ok)
{
activationFunction = (str == "s") ? NeuralNetwork.ActivationFunction.SIGMOID : NeuralNetwork.ActivationFunction.TANH;
}
else
{
Console.WriteLine("This value is not allowed.");
}
}
ok = false;
while (!ok)
{
Console.Write("Using bias? [n/Y] ");
str = TAKE_PARAMETERS_FROM_USER ? Console.ReadLine() : "";
str = str.ToLower();
if (str.Length == 0)
{
str = "y";
}
ok = (str == "n" || str == "y");
if (ok)
{
networkUseBias = (str == "y");
}
else
{
Console.WriteLine("This value is not allowed.");
}
}
Console.WriteLine("What should stops learning?");
Console.WriteLine("0 - minimum percentage error or maximum number of epochs");
Console.WriteLine("1 - minimum percentage error");
Console.WriteLine("2 - maximum number of epochs");
Console.WriteLine("3 - only F2 key");
ok = false;
while (!ok)
{
int choice = 0;
Console.Write("Your choice? [3] ");
str = TAKE_PARAMETERS_FROM_USER ? Console.ReadLine() : "";
if (str.Length == 0)
{
choice = 3;
ok = true;
}
else
{
ok = int.TryParse(str, out choice);
}
if (ok)
{
switch (choice)
{
case 0:
ok = false;
while (!ok)
{
Console.Write("Maximum number of epochs: [" + DEFAULT_MAX_EPOCH_NO + "] ");
str = TAKE_PARAMETERS_FROM_USER ? Console.ReadLine() : "";
if (str.Length > 0)
{
ok = long.TryParse(str, out maxEpochNo);
if (ok)
{
ok = (maxEpochNo > 0);
}
}
else
{
maxEpochNo = DEFAULT_MAX_EPOCH_NO;
ok = true;
}
if (!ok)
{
Console.WriteLine("This value is not allowed.");
}
}
ok = false;
while (!ok)
{
Console.Write("Minimum percentage error: [" + DEFAULT_MIN_ERR_PERCENTAGE_TO_STOP + "] ");
str = TAKE_PARAMETERS_FROM_USER ? Console.ReadLine() : "";
if (str.Length > 0)
{
ok = double.TryParse(str, out minErrorPercentageToStop);
if (ok)
{
ok = (minErrorPercentageToStop > 0);
}
}
else
{
minErrorPercentageToStop = DEFAULT_MIN_ERR_PERCENTAGE_TO_STOP;
ok = true;
}
if (!ok)
{
Console.WriteLine("This value is not allowed.");
}
}
break;
case 1:
ok = false;
while (!ok)
{
Console.Write("Minimum percentage error: [" + DEFAULT_MIN_ERR_PERCENTAGE_TO_STOP + "] ");
str = TAKE_PARAMETERS_FROM_USER ? Console.ReadLine() : "";
if (str.Length > 0)
{
ok = double.TryParse(str, out minErrorPercentageToStop);
if (ok)
{
ok = (minErrorPercentageToStop > 0);
}
}
else
{
minErrorPercentageToStop = DEFAULT_MIN_ERR_PERCENTAGE_TO_STOP;
ok = true;
}
if (!ok)
{
Console.WriteLine("This value is not allowed.");
}
}
maxEpochNo = 0;
break;
case 2:
ok = false;
while (!ok)
{
Console.Write("Maximum number of epochs: [" + DEFAULT_MAX_EPOCH_NO + "] ");
str = TAKE_PARAMETERS_FROM_USER ? Console.ReadLine() : "";
if (str.Length > 0)
{
ok = long.TryParse(str, out maxEpochNo);
if (ok)
{
ok = (maxEpochNo > 0);
}
}
else
{
maxEpochNo = DEFAULT_MAX_EPOCH_NO;
ok = true;
}
if (!ok)
{
Console.WriteLine("This value is not allowed.");
}
}
minErrorPercentageToStop = 0;
break;
case 3:
minErrorPercentageToStop = 0;
maxEpochNo = 0;
break;
default:
ok = false;
break;
}
}
if (!ok)
{
Console.WriteLine("This value is not allowed.");
}
}
ok = false;
while (!ok)
{
Console.Write("Learning ratio: [" + learningRatio + "] ");
str = TAKE_PARAMETERS_FROM_USER ? Console.ReadLine() : "";
str = str.Replace('.', ',');
double newLearningRatio = learningRatio;
if (str.Length > 0)
{
ok = double.TryParse(str, out newLearningRatio);
if (ok)
{
ok = (newLearningRatio > 0);
}
}
else
{
ok = true;
}
if (ok)
{
learningRatio = newLearningRatio;
}
else
{
Console.WriteLine("This value is not allowed.");
}
}
ok = false;
while (!ok)
{
Console.Write("Momentum: [" + momentum + "] ");
str = TAKE_PARAMETERS_FROM_USER ? Console.ReadLine() : "";
str = str.Replace('.', ',');
double newMomentum = momentum;
if (str.Length > 0)
{
ok = double.TryParse(str, out newMomentum);
if (ok)
{
ok = (newMomentum > 0);
}
}
else
{
ok = true;
}
if (ok)
{
momentum = newMomentum;
}
else
{
Console.WriteLine("This value is not allowed.");
}
}
ok = false;
while (!ok)
{
Console.Write("Should the order of the learning data be random? [n/Y] ");
str = TAKE_PARAMETERS_FROM_USER ? Console.ReadLine() : "";
str = str.ToLower();
if (str.Length == 0)
{
str = "y";
}
ok = (str == "n" || str == "y");
if (ok)
{
randomOrderOfLearningData = (str == "y");
}
else
{
Console.WriteLine("This value is not allowed.");
}
}
ok = false;
while (!ok)
{
Console.Write("File with learning data: [" + filename + "] ");
str = TAKE_PARAMETERS_FROM_USER ? Console.ReadLine() : "";
if (str.Length == 0)
{
str = filename;
}
FileInfo fi = new FileInfo(str);
ok = fi.Exists;
if (!ok)
{
Console.WriteLine("File not exists.");
}
else
{
filename = str;
Console.WriteLine("Loading teaching elements from file...");
ok = LoadTeachingElements(inputCount, outputCount, out learningData, filename);
if (ok)
{
Console.WriteLine("Teaching elements loaded.");
dataLoopRunning = false;
}
else
{
Console.Write("Do you want to provide new data [d] or try another file [F]? ");
str = TAKE_PARAMETERS_FROM_USER ? Console.ReadLine() : "";
str = str.ToLower();
if (str.Length == 0)
{
str = "f";
}
ok = (str == "d" || str == "f");
if (ok)
{
if (str == "f")
{
ok = false;
}
else
{
break;
}
}
else
{
Console.WriteLine("This value is not allowed.");
}
}
}
}
}
#endregion
#region Layers creation
int layerNo = 0;
LayerCreationInfo inputLci = new LayerCreationInfo();
inputLci.HowManyNeuronsPerLayer = inputCount + (networkUseBias ? 1 : 0);
inputLci.LayerActivationFunction = activationFunction;
inputLci.LayerNo = layerNo++;
inputLci.PreviousLayerNeuronsCount = 0;
layers.Add(new Layer(inputLci, networkUseBias, Layer.LayerType.INPUT, -0.5, +0.5));
foreach (int hiddenLayerNeurons in hiddenLayersNeuronsCount)
{
LayerCreationInfo hiddenLci = new LayerCreationInfo();
hiddenLci.HowManyNeuronsPerLayer = hiddenLayerNeurons + (networkUseBias ? 1 : 0);
hiddenLci.LayerActivationFunction = activationFunction;
hiddenLci.PreviousLayerNeuronsCount = layers.ElementAt(layerNo - 1).Neurons.Count;
hiddenLci.LayerNo = layerNo++;
layers.Add(new Layer(hiddenLci, networkUseBias, Layer.LayerType.HIDDEN, -0.5, +0.5));
}
LayerCreationInfo outputLci = new LayerCreationInfo();
outputLci.HowManyNeuronsPerLayer = outputCount;
outputLci.LayerActivationFunction = activationFunction;
outputLci.PreviousLayerNeuronsCount = layers.ElementAt(layerNo - 1).Neurons.Count;
outputLci.LayerNo = layerNo++;
layers.Add(new Layer(outputLci, networkUseBias, Layer.LayerType.OUTPUT, -0.5, +0.5));
#endregion
NeuralNetwork neuralNetwork = new NeuralNetwork(
new Topology()
{
Layers = layers,
IsNetworkUsingBiases = networkUseBias,
OutputCount = outputCount
},
-0.5,
+0.5,
NeuralNetwork.LearningMethod.NOT_LINEAR,
"NN",
collectSparseHistoryEvery
);
LearnNetwork(neuralNetwork, networkUseBias, learningData, learningRatio, momentum,
minErrorPercentageToStop, maxEpochNo, randomOrderOfLearningData, writeOnConsoleEveryEpoch, true);
String prefix = DateTime.Now.ToString("yyyyMMdd_HHmmss");
String saveNetworkFilename = prefix + "_Network.nnc";
String saveErrorsFilename = prefix + "_Errors.csv";
neuralNetwork.SaveNetworkToFile(saveNetworkFilename);
// neuralNetwork.SaveErrorsToFile(saveErrorsFilename);
Console.WriteLine();
Console.Write("Press any key to go to main menu... ");
Console.ReadKey();
}
public Neuron(double value, NeuralNetwork.ActivationFunction activationFunction)
{
Value = activationFunction(value);
}
