public NeuralNetwork(int nbInputs, int nbOuputs, List <int> innerLayerDescr)
{
_inputs = new NeuralLayer(nbInputs);
_outputs = new NeuralLayer(nbOuputs);
for (int idxLayer = 0; idxLayer < innerLayerDescr.Count; idxLayer++)
{
_innerLayers.Add(new NeuralLayer(innerLayerDescr[idxLayer]));
}
NeuralLayer prevLayer = _outputs;
for (int idxLayer = innerLayerDescr.Count - 1; idxLayer >= 0; idxLayer--)
{
NeuralLayer postLayer = idxLayer > 0 ? _innerLayers[idxLayer - 1] : _inputs;
for (int idxNeuron = 0; idxNeuron < innerLayerDescr[idxLayer]; idxNeuron++)
{
_innerLayers[idxLayer].Neurons.Add(new Neuron(postLayer, prevLayer));
}
_innerLayers[idxLayer].SetBias(new NeuronBias(prevLayer));
prevLayer = _innerLayers[idxLayer];
}
for (int idxInput = 0; idxInput < nbInputs; idxInput++)
{
_inputs.Neurons.Add(new NeuronInput(prevLayer, ""));
}
_inputs.SetBias(new NeuronBias(prevLayer));
for (int idxOutput = 0; idxOutput < nbOuputs; idxOutput++)
{
_outputs.Neurons.Add(new NeuronOutput(_innerLayers.Last(), ""));
}
}
void AnnVisualizerForm_Paint(object sender, PaintEventArgs e)
{
Pen pen = new Pen(Color.FromArgb(255, 0, 0, 0));
int idxNeuron = 0;
Label curDest = null;
int idxLayer = 0;
int nextLayerIdx = _ann._outputs.Neurons.Count;
foreach (var nn in _ann._outputs.Neurons)
{
var nextOrigin = findWeightLabelOrig(idxNeuron++);
int idxWeight = 0;
Value max = nn.Weights.Max();
foreach (var nextNn in _ann._innerLayers[0].Neurons)
{
curDest = findWeightLabelDest(nextLayerIdx++);
double strength = Math.Abs(nn.Weights[idxWeight++].X / max.X);
pen.Color = Color.FromArgb(255, (byte)(strength * 255.0), 0, (byte)((1.0 - strength) * 255.0));
e.Graphics.DrawLine(pen, nextOrigin.Location.X, nextOrigin.Location.Y, curDest.Location.X, curDest.Location.Y + 10);
}
}
foreach (var layer in _ann._innerLayers)
{
int idxCurLayerNn = idxNeuron + layer.Neurons.Count;
foreach (var nn in layer.Neurons)
{
var nextOrigin = findWeightLabelOrig(idxNeuron++);
NeuralLayer nextLayer = null;
if (_ann._innerLayers.Count > idxLayer + 1)
{
nextLayer = _ann._innerLayers[idxLayer + 1];
}
else
{
nextLayer = _ann._inputs;
}
nextLayerIdx = idxCurLayerNn;
if (nn.Weights.Count > 0)
{
Value max = nn.Weights.Max();
int idxWeight = 0;
foreach (var nextNn in nextLayer.Neurons)
{
curDest = findWeightLabelDest(nextLayerIdx++);
double strength = Math.Abs(nn.Weights[idxWeight++].X / max.X);
pen.Color = Color.FromArgb(255, (byte)(strength * 255.0), 0, (byte)((1.0 - strength) * 255.0));
e.Graphics.DrawLine(pen, nextOrigin.Location.X, nextOrigin.Location.Y, curDest.Location.X, curDest.Location.Y + 10);
}
}
}
idxLayer++;
}
}
public AnnVisualizerForm(NeuralNetwork ann)
{
_ann = ann;
InitializeComponent();
ResizeEnd += AnnVisualizerForm_Resize;
Paint += AnnVisualizerForm_Paint;
var tlayer = new NeuralLayer(1);
Neuron input = new NeuronInput(tlayer, "X");
tlayer.Neurons.Add(input);
_transferFunc = new Neuron(tlayer, new NeuralLayer(0), "transfer");
_transferFunc.Weights[0].X = 1.0;
Recompute();
}
public AnnVisualizerForm(NeuralNetwork ann)
{
_ann = ann;
InitializeComponent();
ResizeEnd += AnnVisualizerForm_Resize;
Paint += AnnVisualizerForm_Paint;
var tlayer = new NeuralLayer(1);
Neuron input = new NeuronInput(tlayer, "X");
tlayer.Neurons.Add(input);
_transferFunc = new Neuron(tlayer, new NeuralLayer(0), "transfer");
_transferFunc.Weights[0].X = 1.0;
Recompute();
}
public Neuron(NeuralLayer children, NeuralLayer parentLayer, string label = "hidden")
{
_parents = parentLayer;
_children = children;
_label = label;
int size = children.Size;
while (size-- > 0)
{
Weights.Add(new Value());
}
if (children.Size > 0) // Bias neuron
{
Weights.Add(new Value());
}
size = parentLayer.Size;
while (size-- > 0)
{
Deltas.Add(new Value());
}
}
public NeuronBias(NeuralLayer nextLayer)
: base(nextLayer, "bias")
{
Value.X = 1.0;
}
public NeuronInput(NeuralLayer firstInnerLayer, string label)
: base(new NeuralLayer(0), firstInnerLayer, label)
{
}
public NeuronOutput(NeuralLayer outputLayer, string label)
: base(outputLayer, new NeuralLayer(0), label)
{
}
public NeuralNetwork(int nbInputs, int nbOuputs, List<int> innerLayerDescr)
{
_inputs = new NeuralLayer(nbInputs);
_outputs = new NeuralLayer(nbOuputs);
for (int idxLayer = 0; idxLayer < innerLayerDescr.Count; idxLayer++)
_innerLayers.Add(new NeuralLayer(innerLayerDescr[idxLayer]));
NeuralLayer prevLayer = _outputs;
for (int idxLayer = innerLayerDescr.Count - 1; idxLayer >= 0; idxLayer--)
{
NeuralLayer postLayer = idxLayer > 0 ? _innerLayers[idxLayer - 1] : _inputs;
for (int idxNeuron = 0; idxNeuron < innerLayerDescr[idxLayer]; idxNeuron++)
_innerLayers[idxLayer].Neurons.Add(new Neuron(postLayer, prevLayer));
_innerLayers[idxLayer].SetBias(new NeuronBias(prevLayer));
prevLayer = _innerLayers[idxLayer];
}
for (int idxInput = 0; idxInput < nbInputs; idxInput++)
_inputs.Neurons.Add(new NeuronInput(prevLayer, ""));
_inputs.SetBias(new NeuronBias(prevLayer));
for (int idxOutput = 0; idxOutput < nbOuputs; idxOutput++)
_outputs.Neurons.Add(new NeuronOutput(_innerLayers.Last(), ""));
}
public NeuronOutput(NeuralLayer outputLayer, string label)
: base(outputLayer, new NeuralLayer(0), label)
{
}
public NeuronInput(NeuralLayer firstInnerLayer, string label)
: base(new NeuralLayer(0), firstInnerLayer, label)
{
}
public NeuronBias(NeuralLayer nextLayer)
: base(nextLayer, "bias")
{
Value.X = 1.0;
}
public Neuron(NeuralLayer children, NeuralLayer parentLayer, string label = "hidden")
{
_parents = parentLayer;
_children = children;
_label = label;
int size = children.Size;
while (size--> 0)
Weights.Add(new Value());
if (children.Size > 0) // Bias neuron
Weights.Add(new Value());
size = parentLayer.Size;
while (size--> 0)
Deltas.Add(new Value());
}