public LinearWeightLayer(INodeLayer inputLayer, INodeLayer outputLayer)
{
Connections = new List <IConnection>();
InputNodes = inputLayer.Nodes;
OutputNodes = outputLayer.Nodes;
foreach (var inode in inputLayer.Nodes)
{
foreach (var onode in outputLayer.Nodes.Where(n => !(n is InputNode)))
{
Connections.Add(new LinearWeight(inode, onode));
}
}
}
public LinearWeightLayer(INodeLayer inputLayer, INodeLayer outputLayer)
{
Connections = new List<IConnection>();
InputNodes = inputLayer.Nodes;
OutputNodes = outputLayer.Nodes;
foreach (var inode in inputLayer.Nodes)
{
foreach (var onode in outputLayer.Nodes.Where(n => !(n is InputNode)))
{
Connections.Add(new LinearWeight(inode, onode));
}
}
}
public MLP(InputLayer inputLayer, INodeLayer[] layers)
{
Create(inputLayer, layers);
}
private void Create(InputLayer inputLayer, INodeLayer[] layers)
{
NodeLayers = new INodeLayer[layers.Length + 1];
ConnectionLayers = new List<IConnectionLayer>();
var nodes = new List<List<INode>>();
NumberOfInputs = inputLayer.Input.Length;
NumberOfOutputs = layers.Last().Output.Length;
NodeLayers[0] = inputLayer;
nodes.Add(inputLayer.Nodes);
for (int i = 0; i < layers.Length; i++)
{
var ih = layers[i];
NodeLayers[i+1] = ih;
nodes.Add(ih.Nodes);
}
for (var i = 0; i < NodeLayers.Length - 1; i++)
{
var cl = new LinearWeightLayer(NodeLayers[i], NodeLayers[i + 1]);
ConnectionLayers.Add(cl);
}
StagedNodes = nodes.Select(n => n.ToArray()).ToArray();
Nodes = nodes.SelectMany(s => s).ToList();
}
public double Train(INodeLayer NodeLayer, int Epochs = 1000)
{
if (Epochs < 1)
throw new ArgumentException("At least 1 epoch is required.");
if (DataSets.Length < 1)
throw new ArgumentException("No DataSets have been loaded.");
if (InputCount != NodeLayer.Inputs)
throw new ArgumentException("The given INodeLayer does not have the same number of Inputs as the DataSets.");
if (OutputCount != NodeLayer.Outputs)
throw new ArgumentException("The given INodeLayer does not have the same number of Outputs as the DataSets.");
for (int i = 0; i < Epochs; i++)
{
for (int n = 0; n < NodeLayer.Outputs; n++)
{
foreach(var dataSet in DataSets)
{
double error = 0;
NodeLayer.Nodes[n].Calculate(dataSet.Inputs, dataSet.Outputs[n], ref error);
double delta = error * LearningRate;
AdjustNode(NodeLayer.Nodes[n], dataSet.Inputs, delta);
}
}
}
double SSE = 0;
foreach (var dataSet in DataSets)
{
double error = 0;
NodeLayer.Calculate(dataSet.Inputs, dataSet.Outputs, ref error);
SSE += Math.Pow(error, 2);
}
return SSE;
}
