private void BuildGrids()
{
MlpNetwork network = _programLogic.ExaminedNetwork;
BuildGridColumns(uiInputData, network.InputCount + 1);
uiInputData.Rows.Add(2);
uiInputData.Rows[0].Cells[0].Value = "Input number (i)";
uiInputData.Rows[1].Cells[0].Value = "Original inputs (u(i))";
uiInputData.Rows[2].Cells[0].Value = "Normalized inputs (x(i))";
BuildGridColumns(uiTeachingProgress, network.OutputCount + 1);
uiTeachingProgress.Rows.Add(5);
uiTeachingProgress.Rows[0].Cells[0].Value = "Output number (j)";
uiTeachingProgress.Rows[1].Cells[0].Value = "Value before teaching (y(j))";
uiTeachingProgress.Rows[2].Cells[0].Value = "Value expected (z(j))";
uiTeachingProgress.Rows[3].Cells[0].Value = "Error before teaching (e(j))";
uiTeachingProgress.Rows[4].Cells[0].Value = "Value after teaching (y'(j))";
uiTeachingProgress.Rows[5].Cells[0].Value = "Error after teaching (e'(j))";
for (int i = 1; i < uiTeachingProgress.ColumnCount; i++)
{
DataGridViewCellStyle style =
uiTeachingProgress.Rows[2].Cells[i].Style;
style.BackColor = style.SelectionBackColor = Color.LightGreen;
style = uiTeachingProgress.Rows[3].Cells[i].Style;
style.BackColor = style.SelectionBackColor = Color.Pink;
style = uiTeachingProgress.Rows[5].Cells[i].Style;
style.BackColor = style.SelectionBackColor = Color.Pink;
}
}
internal void LoadTeachingSet(string fileName)
{
TeachingSet tset = NeuralNetworks.TeachingSet.FromFile(fileName);
if (tset == null || tset.Count == 0)
{
throw new ApplicationException("The file does not contain any data.");
}
_teachingSetFileName = fileName;
_teachingSet = tset;
_normalizedTeachingSet = new TeachingSet(tset);
_normalizedTeachingSet.Normalize();
_examinedNetwork = new MlpNetwork(tset.InputCount, false,
new int[] { tset.OutputCount }, typeof(BipolarNeuron));
InitializeTeaching();
}
private void CreateNetwork()
{
if (uiSuspendCreating.Checked)
{
return;
}
int layerCount = uiLayerCount.Value;
int outputCount = uiOutputCount.Value;
double weightsRange = GetWeightsRange(uiWeightsRange.Value);
double topWeight = weightsRange / 2.0;
double bottomWeight = topWeight - weightsRange;
double biasWeightsRange = GetWeightsRange(uiBiasWeightsRange.Value);
double bottomBiasWeight = -(biasWeightsRange / 2.0);
int[] neuronCounts = new int[layerCount];
for (int i = 0; i < neuronCounts.Length - 1; i++)
{
neuronCounts[i] = layerCount;
}
neuronCounts[neuronCounts.Length - 1] = outputCount;
Type neuronType;
switch (uiTransferFunctionType.SelectedIndex)
{
case 0: neuronType = typeof(BipolarNeuron); break;
case 1: neuronType = typeof(UnipolarNeuron); break;
case 2: neuronType = typeof(SigmoidalNeuron); break;
case 3: neuronType = typeof(TanhNeuron); break;
default: throw new InvalidOperationException();
}
_examinedNetwork = new MlpNetwork(2, true, neuronCounts, neuronType);
_examinedNetwork.Randomize(_randomGenerator, bottomWeight, topWeight);
if (uiDifferentBiasRange.Checked)
{
foreach (Neuron n in _examinedNetwork.Layers[0])
{
n.Weights[n.Weights.Length - 1] =
bottomBiasWeight +
_randomGenerator.NextDouble() * biasWeightsRange;
}
}
_netFun.NeuralNetwork = _examinedNetwork;
_netFun.OutputOffset = 0.5;
_netFun.OutputScale = 0.5;
_netFun.OutputColors.Clear();
for (int i = 0; i < outputCount; i++)
{
//{tga to paint only blue/red
_netFun.OutputColors.Add(Color.FromArgb(255, 0, 0));
//tga}
//_netFun.OutputColors.Add(Color.FromArgb(
// _randomGenerator.Next(255),
// _randomGenerator.Next(255),
// _randomGenerator.Next(255)));
}
CreateNeuronLines();
}
private void CreateNetwork()
{
int[] neuronCounts = new int[trackBar3.Value];
for (int i = 0; i < neuronCounts.Length - 1; i++)
{
neuronCounts[i] = trackBar3.Value;
}
neuronCounts[neuronCounts.Length - 1] = trackBar2.Value;
_network = new MlpNetwork(2, true, neuronCounts,
checkBox2.Checked ? typeof(SigmoidalNeuron) : typeof(BipolarNeuron));
if (checkBox2.Checked)
{
foreach (Neuron[] layer in _network.Layers)
{
foreach (Neuron n in layer)
{
((SigmoidalNeuron)n).Beta = 100.0;
}
}
}
_network.Randomize(_randomGenerator, -10.0, 10.0);
/*foreach (Neuron n in _network.Layers[0])
* n.Weights[n.Weights.Length - 1] =
* _randomGenerator.NextDouble() * 20.0 - 10.0;*/
netFun.NeuralNetwork = _network;
netFun.OutputColors.Clear();
if (trackBar2.Value == 2)
{
netFun.OutputColors.Add(Color.Red);
netFun.OutputColors.Add(Color.Green);
}
else
{
for (int i = 0; i < trackBar2.Value; i++)
{
netFun.OutputColors.Add(Color.FromArgb(
_randomGenerator.Next(255),
_randomGenerator.Next(255),
_randomGenerator.Next(255)));
}
}
_neuronLines.Data.Clear();
foreach (Neuron n in _network.Layers[0])
{
double a = n.Weights[0];
double b = n.Weights[1];
double c = n.Weights[2];
RectangleF vrect = chartPlotter4.VisibleRectangle;
PointF[] points = new PointF[2];
int currentPoint = 0;
double y = (-c - a * vrect.Left) / b;
if (y >= vrect.Top && y <= vrect.Bottom)
{
points[currentPoint++] = new PointF(vrect.Left, (float)y);
}
y = (-c - a * vrect.Right) / b;
if (y >= vrect.Top && y <= vrect.Bottom)
{
points[currentPoint++] = new PointF(vrect.Right, (float)y);
}
if (currentPoint < 2)
{
double x = (-c - b * vrect.Top) / a;
if (x >= vrect.Left && x <= vrect.Right)
{
points[currentPoint++] = new PointF((float)x, vrect.Top);
}
if (currentPoint < 2)
{
x = (-c - b * vrect.Bottom) / a;
if (x >= vrect.Left && x <= vrect.Right)
{
points[currentPoint++] = new PointF((float)x, vrect.Bottom);
}
}
}
_neuronLines.Data.Add(points[0]);
_neuronLines.Data.Add(points[1]);
/*_neuronLines.Data.Add(new PointF(i / 10f, i / 10f));
* _neuronLines.Data.Add(new PointF(i / 10f + 0.1f, i / 10f));*/
}
}
