public static void drawNeurons(List <List <Neuron> > layers)
{
UInt16 x = (Visualization.nrnSize * 2), y = x;
List <List <Byte> > adjustedLayers = new List <List <Byte> >();
IntPtr dc = Visualization.GetDC(IntPtr.Zero);
using (Pen pen = new Pen(Color.Black))
using (Pen pen0 = new Pen(Color.White))
using (Graphics g = Graphics.FromHdc(dc)) {
g.FillRectangle(pen0.Brush, 0, 0, Visualization.width, Visualization.height);
foreach (List <Neuron> layer in layers)
{
if (layer.Count <= Visualization.maxLayerLengthToDraw)
{
adjustedLayers.Add(layer.Select(x0 => x0.activation).ToList());
continue;
}
UInt16 split = (UInt16)(layer.Count / Visualization.maxLayerLengthToDraw), j = 0, ctr = 0;
Byte[] newNeurons = new Byte[Visualization.maxLayerLengthToDraw];
Byte i = 0;
UInt32 sum = 0;
while (i < Visualization.maxLayerLengthToDraw)
{
while (j < split)
{
sum += layer[ctr].activation;
++j;
++ctr;
}
newNeurons[i] = (Byte)(sum / j);
sum = 0;
j = 0;
++i;
}
adjustedLayers.Add(newNeurons.ToList());
}
foreach (List <Byte> layer in adjustedLayers)
{
foreach (Byte b in layer)
{
g.DrawEllipse(pen, new RectangleF(x, y, Visualization.nrnSize, Visualization.nrnSize));
g.DrawString(b.ToString(), Visualization.font, pen.Brush, x, y + (Visualization.nrnSize / 4));
y += (Visualization.nrnSize * 2);
}
x += (Visualization.nrnSize * 2);
y = (Visualization.nrnSize * 2);
}
}
Visualization.ReleaseDC(IntPtr.Zero, dc);
}
/// <summary>
/// Make a prediction
/// </summary>
/// <param name="input"></param>
/// <param name="desired"></param>
/// <param name="learn"></param>
/// <returns></returns>
public List <Byte> makePrediction(IEnumerable <Byte> input, Byte[] desired, Boolean learn, Boolean visualize = false, Action onFail = null)
{
//Clear all potential old neuron values
foreach (List <Neuron> layer in this.layers.Skip(1))
{
foreach (Neuron nrn in layer)
{
nrn.activation = 0;
}
}
//Restructure input
List <Byte> _input;
if (input.GetType() == typeof(List <Byte>))
{
_input = (List <Byte>)input;
}
else
{
_input = new List <Byte>(input);
}
//Fill input layers
UInt16 i = 0;
while (i < this.layerData[0]) /* possible data loss if input.Length>layerData[0] */
{
this.layers[0][i].activation = _input[i];
++i;
}
i = 0;
Single sum = 0;
while (i < (this.layerCount - 1))
{
foreach (Neuron neuron in this.layers[i + 1])
{
foreach (Neuron neuron0 in this.layers[i])
{
// Console.WriteLine("Activation:"+neuron0.activation.ToString()+",bias:"+neuron0.bias.ToString()+",weight:"+this.weights[neuron0][neuron].ToString());
sum += (neuron0.activation + neuron0.bias) * (this.weights[neuron0][neuron]) * 0.5F;
}
neuron.activation = sum.sigmoid();
// Console.WriteLine("Sum:"+sum.ToString()+",Sigmoided:"+sum.sigmoid().ToString());
sum = 0;
}
++i;
}
if (visualize)
{
Visualization.drawNeurons(this.layers);
}
if (learn)
{
this.backPropogate(desired, onFail);
}
return(this.layers[this.layerCount - 1].Select(x => x.activation).ToList());
}
