/// <summary>
/// Randomizes the weights of all connections between all neurons in range of -5.0 to 5.0
/// </summary>
public void RandomizeWeights()
{
MyRandom r = new MyRandom(-0.5, 0.5);
//set random weights for output
foreach (Neuron itOut in outputLayer)
{
List <Connection> con = ((WorkingNeuron)(itOut)).connections;
foreach (Connection c in con)
{
c.SetWeight(r.NextDouble());
}
}
//set random weights for hidden
foreach (Neuron itHid in hiddenLayer)
{
WorkingNeuron wn = itHid as WorkingNeuron;
if (wn != null)
{
List <Connection> con = wn.connections;
foreach (Connection c in con)
{
c.SetWeight(r.NextDouble());
}
}
}
}
/// <summary>
/// Creates the mesh of the connection the neurons are maintaining, called after deserialization!
/// </summary>
public void CreateMesh()
{
foreach (Neuron n in hiddenLayer)
{
WorkingNeuron wn = n as WorkingNeuron;
if (wn != null)
{
foreach (Connection c in wn.connections)
{
c.sourceNeuron = this.GetNeuronByName(c.sourceNeuronName);
}
}
}
foreach (Neuron n in outputLayer)
{
WorkingNeuron wn = n as WorkingNeuron;
if (wn != null)
{
foreach (Connection c in wn.connections)
{
c.sourceNeuron = this.GetNeuronByName(c.sourceNeuronName);
}
}
}
}
public override Neuron Copy()
{
WorkingNeuron n = new WorkingNeuron();
n.name = this.name;
n.value = this.value;
n.activationfunction = activationfunction;
n.parentNetwork = this.parentNetwork;
return(n);
}
public void Invalidate()
{
for (int l = layers.Count - 1; l > 0; --l)
{
foreach (Neuron n in layers.ElementAt(l))
{
WorkingNeuron wn = n as WorkingNeuron;
if (wn != null)
{
wn.Invalidate();
}
}
}
}
public void Invalidate()
{
foreach (Neuron itOut in outputLayer)
{
((WorkingNeuron)(itOut)).Invalidate();
}
foreach (Neuron itHid in hiddenLayer)
{
WorkingNeuron n = itHid as WorkingNeuron;
if (n != null)
{
n.Invalidate();
}
}
}
/// <summary>
/// Creates the mesh of the connection the neurons are maintaining, called after deserialization!
/// </summary>
public void CreateMesh()
{
for (int l = layers.Count - 1; l > 0; --l)
{
foreach (Neuron n in layers.ElementAt(l))
{
WorkingNeuron wn = n as WorkingNeuron;
if (wn != null)
{
foreach (Connection c in wn.connections)
{
c.sourceNeuron = this.GetNeuronByName(c.sourceNeuronName);
}
}
}
}
}
/// <summary>
/// Function for Network Manipulation
/// connect each hidden neuron with each input neuron
/// and each output neuron with each hidden neuron
/// </summary>
public void CreateFullMesh()
{
//connect each hidden neuron to each input neuron
for (int l = layers.Count - 1; l > 0; --l)
{
foreach (Neuron n in layers.ElementAt(l))
{
WorkingNeuron wn = n as WorkingNeuron;
if (wn != null)
{
foreach (Neuron itIn in layers.ElementAt(l - 1))
{
wn.AddConnection(itIn, 1.0);
}
}
}
}
}
/// <summary>
/// Randomizes the weights of all connections between all neurons in range of -5.0 to 5.0
/// </summary>
public void RandomizeWeights()
{
MyRandom r = new MyRandom(-0.5, 0.5);
for (int l = layers.Count - 1; l > 0; --l)
{
foreach (Neuron n in layers.ElementAt(l))
{
WorkingNeuron wn = n as WorkingNeuron;
if (wn != null)
{
List <Connection> con = wn.connections;
foreach (Connection c in con)
{
c.SetWeight(r.NextDouble());
}
}
}
}
}
/// <summary>
/// Function for Network Manipulation
/// connect each hidden neuron with each input neuron
/// and each output neuron with each hidden neuron
/// </summary>
public void CreateFullMesh()
{
//Connect each output neuron to each hidden layer neuron
foreach (Neuron itOut in outputLayer)
{
foreach (Neuron itHid in hiddenLayer)
{
((WorkingNeuron)(itOut)).AddConnection(itHid, 1.0);
}
}
//connect each hidden neuron to each input neuron
foreach (Neuron itHid in hiddenLayer)
{
WorkingNeuron wn = itHid as WorkingNeuron;
if (wn != null)
{
foreach (Neuron itIn in inputLayer)
{
wn.AddConnection(itIn, 1.0);
}
}
}
}
public WorkingNeuron(WorkingNeuron n) : base(n)
{
valid = n.valid;
connections = new List <Connection>();
}
/// <summary>
/// (Deep) Copy Constructor for NeuroalNet
/// </summary>
/// <param name="net"></param>
public NeuralNetwork(NeuralNetwork net) : this(net.layers.Count - 2)
{
settings = net.settings;
name = "Copy of " + net.name;
score = -1.0f;
foreach (Neuron inNeuron in net.layers.ElementAt(0))
{
Neuron n;
if ((n = inNeuron as InputNeuron) != null)
{
AddInputNeuron(n.Copy());
}
else if ((n = inNeuron as BiasNeuron) != null)
{
AddInputNeuron(n.Copy());
}
}
foreach (Neuron outNeuron in net.layers.ElementAt(net.layers.Count - 1))
{
Neuron n;
if ((n = outNeuron as WorkingNeuron) != null)
{
AddOutputNeuron(n.Copy());
}
}
for (int l = 1; l <= net.layers.Count - 2; ++l)
{
foreach (Neuron hidNeuron in net.layers.ElementAt(l))
{
Neuron n;
if ((n = hidNeuron as WorkingNeuron) != null)
{
AddHiddenNeuron(n.Copy(), l);
}
else if ((n = hidNeuron as BiasNeuron) != null)
{
AddHiddenNeuron(n.Copy(), l);
}
}
}
//generate Full mesh
this.CreateFullMesh();
//copy weights
//hidden layers
for (int l = 1; l < layers.Count - 2; ++l)
{
for (int i = 0; i < layers.ElementAt(l).Count; ++i)
{
WorkingNeuron n = layers.ElementAt(l).ElementAt(i) as WorkingNeuron;
if (n != null)
{
for (int c = 0; c < ((WorkingNeuron)net.layers.ElementAt(l).ElementAt(i)).connections.Count; ++c)
{
n.connections.ElementAt(c).SetWeight(((WorkingNeuron)net.layers.ElementAt(l).ElementAt(i)).connections.ElementAt(c).GetWeight());
}
}
}
}
//output layer
for (int i = 0; i < layers.ElementAt(layers.Count - 1).Count; ++i)
{
WorkingNeuron n = layers.ElementAt(layers.Count - 1).ElementAt(i) as WorkingNeuron;
if (n != null)
{
for (int c = 0; c < ((WorkingNeuron)net.layers.ElementAt(layers.Count - 1).ElementAt(i)).connections.Count; ++c)
{
n.connections.ElementAt(c).SetWeight(((WorkingNeuron)net.layers.ElementAt(layers.Count - 1).ElementAt(i)).connections.ElementAt(c).GetWeight());
}
}
}
}
/// <summary>
/// (Deep) Copy Constructor for NeuroalNet
/// </summary>
/// <param name="net"></param>
public NeuralNetwork(NeuralNetwork net)
{
inputLayer = new List <Neuron>();
outputLayer = new List <Neuron>();
hiddenLayer = new List <Neuron>();
layers = new List <List <Neuron> >();
layers.Add(inputLayer);
layers.Add(hiddenLayer);
layers.Add(outputLayer);
settings = net.settings;
name = "Copy of " + net.name;
score = -1.0f;
foreach (Neuron inNeuron in net.inputLayer)
{
Neuron n;
if ((n = inNeuron as InputNeuron) != null)
{
AddInputNeuron(n.Copy());
}
else if ((n = inNeuron as BiasNeuron) != null)
{
AddInputNeuron(n.Copy());
}
}
foreach (Neuron outNeuron in net.outputLayer)
{
Neuron n;
if ((n = outNeuron as WorkingNeuron) != null)
{
AddOutputNeuron(n.Copy());
}
}
foreach (Neuron hidNeuron in net.hiddenLayer)
{
Neuron n;
if ((n = hidNeuron as WorkingNeuron) != null)
{
AddHiddenNeuron(n.Copy());
}
else if ((n = hidNeuron as BiasNeuron) != null)
{
AddHiddenNeuron(n.Copy());
}
}
//generate Full mesh
this.CreateFullMesh();
//copy weights
//hidden layer
for (int i = 0; i < hiddenLayer.Count; ++i)
{
WorkingNeuron n = hiddenLayer.ElementAt(i) as WorkingNeuron;
if (n != null)
{
for (int c = 0; c < ((WorkingNeuron)net.hiddenLayer.ElementAt(i)).connections.Count; ++c)
{
n.connections.ElementAt(c).SetWeight(((WorkingNeuron)net.hiddenLayer.ElementAt(i)).connections.ElementAt(c).GetWeight());
}
}
}
//output layer
for (int i = 0; i < outputLayer.Count; ++i)
{
WorkingNeuron n = outputLayer.ElementAt(i) as WorkingNeuron;
if (n != null)
{
for (int c = 0; c < ((WorkingNeuron)net.outputLayer.ElementAt(i)).connections.Count; ++c)
{
n.connections.ElementAt(c).SetWeight(((WorkingNeuron)net.outputLayer.ElementAt(i)).connections.ElementAt(c).GetWeight());
}
}
}
}