private void AddSynapse(Neuron inputNeuron)
{
var synapse = new Synapse(inputNeuron, this);
inputNeuron.OutputSynapses.Add(synapse);
InputSynapses.Add(synapse);
}
/// <summary>
/// Connects to input (left) layer
/// </summary>
/// <param name="inputNeuron">Neuron form left(previous) layer</param>
/// <param name="weight">Value of synapse between input and current neuron</param>
public void ConnectToInputNeuron(Neuron inputNeuron, float weight)
{
var synapse = new Synapse(inputNeuron, this, weight);
InputSynapses.Add(synapse);
inputNeuron.OutputSynapses.Add(synapse);
}
public double CalculateValue()
{
return(Value = _act(
InputSynapses.Select(
synapse => synapse.Weight * synapse.InputNeuron.Value
).Sum() + Bias
));
}
public Neuron(IEnumerable <Neuron> inputNeurons) : this()
{
foreach (var inputNeuron in inputNeurons)
{
var synapse = new Synapse(inputNeuron, this);
inputNeuron.OutputSynapses.Add(synapse);
InputSynapses.Add(synapse);
}
}
public void Initialize(int numberInputs)
{
for (int i = 0; i < numberInputs; i++)
{
InputSynapses.Add(new Synapse(i + 1, this, Util.r.NextDouble()));
}
LearningRate = 1;
}
public Neuron(IEnumerable <Neuron> inputNeurons, Type activationFunctionType = null) : this(activationFunctionType)
{
foreach (var inputNeuron in inputNeurons)
{
var synapse = new Synapse(inputNeuron, this);
inputNeuron.OutputSynapses.Add(synapse);
InputSynapses.Add(synapse);
}
}
//Tworzenie neuronu wraz z podłączeniem neuronów wejściowych
public Neuron(List <Neuron> inputNeurons, IActivationFunction activationFunction) : this(activationFunction)
{
foreach (var inputNeuron in inputNeurons)
{
var synapse = new Synapse(inputNeuron, this);
inputNeuron.OutputSynapses.Add(synapse);
InputSynapses.Add(synapse);
}
}
public double GetOuputValue()
{
double sum = 0;
for (int i = 0; i < InputSynapses.Count; i++)
{
sum += InputSynapses[i].Weight * InputSynapses[i].InputNeuron.Output;
}
return(Output = Sigmoid.Output(InputSynapses.Sum(a => a.Weight * a.InputNeuron.Output) + BiasWeight));
}
/// <summary>
/// Gets bias connected to current neuron
/// </summary>
/// <returns>Connection</returns>
public Synapse GetInputBiasSynapse()
{
if (InputSynapses.Count == 0)
{
return(null);
}
var biasSynapse = InputSynapses.Skip(InputSynapses.Count - 1).Take(1).Single();
return(biasSynapse);
}
/// <summary>
/// Get all synapses connected to current neuron
/// </summary>
/// <returns>List of synapses</returns>
public List <Synapse> GetInputWeightSynapses()
{
if (InputSynapses.Count == 0)
{
return(new List <Synapse>());
}
var synapses = InputSynapses.Take(InputSynapses.Count - 1).ToList();
return(synapses);
}
/// <summary>
/// Initializes a new instance of the <see cref="Neuron"/> class.
/// </summary>
/// <param name="inputNeurons">The input neurons.</param>
public Neuron(IEnumerable <Neuron> inputNeurons) : this()
{
Ensure.That(inputNeurons).IsNotNull();
foreach (var inputNeuron in inputNeurons)
{
var synapse = new Synapse(inputNeuron, this);
inputNeuron?.OutputSynapses?.Add(synapse);
InputSynapses?.Add(synapse);
}
}
public Neuron(
IEnumerable <Neuron> inputNeurons,
Func <double, double> activationFunction,
Func <double, double> dActivationFunction
) : this(activationFunction, dActivationFunction)
{
foreach (var inputNeuron in inputNeurons)
{
InputSynapses.Add(new Synapse(inputNeuron, this));
inputNeuron.OutputSynapses.Add(new Synapse(inputNeuron, this));
}
}
public Neuron(NeuralNetwork NN, IEnumerable <Neuron> inputNeurons)
: this()
{
Network = NN;
foreach (var inputNeuron in inputNeurons)
{
var synapse = new Synapse(inputNeuron, this);
inputNeuron.OutputSynapses.Add(synapse);
InputSynapses.Add(synapse);
}
}
public float CalculateValue()
{
/*for (int i = 0; i < InputSynapses.Count; i++)
* {
* xSum += InputSynapses[i].Weight * InputSynapses[i].InputNeuron.Value;
* }
****** é o mesmo que fazer o Sum do LINQ
*/
float xSum = InputSynapses.Sum(x => x.Weight * x.InputNeuron.Value);
float u = xSum + Bias;
Value = Sigmoid.Output(u);
return(Value);
}
public void UpdateWeights() => InputSynapses.Do(synapse => synapse.Update());
public virtual void CalculateValue() => Value = Activation(InputSynapses.Sum(synapse => synapse.Weight * synapse.InputNeuron.Value));
//public void UpdateWeights(double learnRate, double momentum)
//{
// var prevDelta = BiasDelta;
// BiasDelta = learnRate * Gradient;
// Bias += BiasDelta + momentum * prevDelta;
// foreach (var synapse in InputSynapses)
// {
// prevDelta = synapse.WeightDelta;
// synapse.WeightDelta = learnRate * Gradient * synapse.InputNeuron.Value;
// synapse.Weight += synapse.WeightDelta + momentum * prevDelta;
// }
//}
#endregion
public void ResetNeuron()
{
Initialization();
InputSynapses.ForEach(synapse => synapse.Initialization());
}
public virtual double CalculateValue()
{
var d = InputSynapses.Sum(a => a.Weight * a.InputNeuron.Value); // + Bias;
return(Value = IsHidden ? NeuralNetwork.SigmoidFunction(d) : NeuralNetwork.IdentityFunction(d));
}
public void AddInput(Synapse input)
{
InputSynapses.Add(input);
}
//计算输出值 ，输入值加上偏置，然后应用sigmoid函数
public virtual double CalculateValue()
{
this.InputValue = InputSynapses.Sum(a => a.Weight * a.InputNeuron.OutputValue);
return(OutputValue = Sigmoid.Output(InputValue + Bias));
}
public virtual double CalculateHiddenValue()
{
return(Value = Sigmoid.Relu(InputSynapses.Sum(a => a.Weight * a.InputNeuron.Value)));
}
public float CalculateValue()
{
return(Value = Sigmoid.Output(InputSynapses.Sum(x => x.Weight * x.InputNeuron.Value) + Bias));
}
public void CalculateValueRecurrent()
{
OutputValue = Sigmoid.Output(InputSynapses.Sum(syn => syn.Weight * syn.InputNeuron.LastValue) + Bias);
}
/// <summary>
/// Calculate the value of the neuron based on the weights and values of the Previous Layer
/// VALUE = SIGMOIDSQUISH(WEIGHT * VALUE) of every previous neuron + bias
/// </summary>
/// <returns></returns>
public float CalculateValue()
{
return(Value = NeuralMath.SigmoidSquish(InputSynapses.Sum(s => s.Weight * s.InputNeuron.Value) + Bias));
}
public virtual double CalculateValue()
{
var inputSignals = InputSynapses.Sum(a => a.Weight * a.InputNeuron.Value);
return(Value = ActivationFunction.Output(inputSignals + Bias));
}
public virtual double CalculateValue() //This function is used in forward propogation process.
{
return(Value = Sigmoid.Output(InputSynapses.Sum(a => a.Weight * a.InputNeuron.Value) + Bias)); //Calculate by ably activation function to (weight*input)+bias
}
public void Mutate()
{
#if false
int ran = NeuralNet.RandomGenerator.Next(1, 8);
if (ran == 1)
{
return;
}
else if (ran > 4)
{
InputSynapses.ForEach(syn =>
{
if (NeuralNet.RandomGenerator.Next(1, 2) == 1)
{
syn.Weight *= 1 + (NeuralNet.RandomGenerator.NextDouble() - 0.5) / 10;
}
});
}
else if (ran == 2)
{
InputSynapses.ForEach(syn =>
{
if (NeuralNet.RandomGenerator.Next(1, 5) == 1)
{
syn.Weight *= -1;
}
});
}
else if (ran == 3)
{
Bias *= 1 + (NeuralNet.RandomGenerator.NextDouble() - 0.5) / 5;
}
else if (ran == 4)
{
Bias *= -1;
}
#else
InputSynapses.ForEach(syn =>
{
var ran = NeuralNet.RandomGenerator.NextDouble();
if (ran < Constants.Con.mutate_percent_of_synapses_uniform)
{
syn.Weight += NeuralNet.RandomGenerator.NextGaussian(0, Constants.Con.mutation_power_synapse);
}
else if (ran < Constants.Con.mutate_percent_of_synapses_uniform + Constants.Con.mutate_percent_of_synapses_new_values)
{
syn.Weight = NeuralNet.RandomGenerator.NextGaussian(0, Constants.Con.init_stdDev_synapse);
}
syn.Weight = Math.Max(-30, Math.Min(30, syn.Weight));
});
{//bias
var ran = NeuralNet.RandomGenerator.NextDouble();
if (ran < Constants.Con.mutate_percent_of_biases_uniform)
{
Bias += NeuralNet.RandomGenerator.NextGaussian(0, Constants.Con.mutation_power_bias);
}
else if (ran < Constants.Con.mutate_percent_of_synapses_uniform + Constants.Con.mutate_percent_of_biases_new_values)
{
Bias = NeuralNet.RandomGenerator.NextGaussian(0, Constants.Con.init_stdDev_bias);
}
Bias = Math.Max(-30, Math.Min(30, Bias));
}
#endif
}
public virtual double CalculateValue()
{
return(Value = Sigmoid.Output(InputSynapses.Sum(a => a.Weight * a.InputNeuron.Value) + Bias));
}
public virtual double CalculateValue()
{
return(Value = NeuralNetwork.SigmoidFunction(InputSynapses.Sum(a => a.Weight * a.InputNeuron.Value) + Bias));
}
