public Form1()
{
FilePath = "IRIS_Train.csv";
Tuple <double, double, double, double> Norm = Tuple.Create(10.0, 0.0, 3.0, 0.0);
List <double[]> input = new List <double[]>();
List <double[]> output = new List <double[]>();
var t = Data.ReadData("IRIS.csv", 4, 1, Norm, false);
input = t.Item1;
output = t.Item2;
PerceptronNetwork network = new MLP.PerceptronNetwork(new int[] { 4, 5, 5, 1 });
NetworkController = new MLP.Network_Control(network, output, input, 0.02, false, Norm, 4, 1);
InitializeComponent();
//Set image to the starting network's image
if (!cbdisabledrawing.Checked)
{
var Image = NetworkImage.Draw(network, network.Layers.Count);
IMG_NetworkTopography.Image = Image;
IMG_NetworkTopography.SizeMode = PictureBoxSizeMode.Zoom;
}
rtbConsole.Text = "";
NetworkController.Paused = true;
}
private void bControllerReset_Click(object sender, EventArgs e)
{
int inputCount;
int outputCount;
var structure = ParseStructureBox(out inputCount, out outputCount);
Tuple <double, double, double, double> Norm = Tuple.Create(10.0, 0.0, 3.0, 0.0);
List <double[]> input = new List <double[]>();
List <double[]> output = new List <double[]>();
var t = Data.ReadData(FilePath, inputCount, outputCount, Norm, rbMNIST.Checked);
input = t.Item1;
output = t.Item2;
PerceptronNetwork network = new MLP.PerceptronNetwork(structure);
NetworkController = new MLP.Network_Control(network, output, input, Double.Parse(tbLearningRate.Text), false, Norm, inputCount, outputCount);
cartesianChart1.Series[0].Values.Clear();
cartesianChart2.Series[0].Values.Clear();
cartesianChart3.Series[0].Values.Clear();
if (!cbdisabledrawing.Checked)
{
var Image = NetworkImage.Draw(NetworkController.Network, NetworkController.Network.Layers.Count);
IMG_NetworkTopography.Image = Image;
}
}
/// <summary>
/// Draws the given Network structure
/// </summary>
/// <param name="Network"></param>
/// <param name="layerCount"></param>
/// <returns></returns>
public static Bitmap Draw(PerceptronNetwork Network, int layerCount)
{
Bitmap m = new Bitmap(layerCount * 200, Network.Layers.Max(item => item.Layer_Neurons.Count) * 200);
Graphics g = Graphics.FromImage(m);
int row = 0;
int column = 0;
int previous_count = 0;
foreach (Layer l in Network.Layers)
{
Font drawFont = new Font("Arial", 10);
SolidBrush drawBrush = new SolidBrush(Color.Black);
PointF layerPoint = new PointF(200 + column * 150 + 50 / 2 - 10, 180);
g.DrawString("Layer " + column, drawFont, drawBrush, layerPoint);
row = 0;
foreach (Perceptron p in l.Layer_Neurons)
{
int offset = 200;
int mult = 150;
int lineOffset = 50;
int x = offset + column * mult;
int y = offset + row * mult;
//Going from left to right place ellipses in place of perceptrons and add labels to them
g.DrawEllipse(Pens.Black, x, y, 100, 100);
PointF drawPoint = new PointF(x + (lineOffset / 2) - 10, y + (lineOffset / 2) + 10);
g.DrawString("Bias" + Math.Round(p.Bias, 3) + "\nOutput:" + Math.Round(p.Output, 2), drawFont, drawBrush, drawPoint);
//Add lines
if (column != 0)
{
for (int b = 0; b < previous_count; b++)
{
g.DrawLine(Pens.Black, new Point(x, y + lineOffset), new Point(x - mult + 2 * lineOffset, offset + b * mult + lineOffset));
}
}
row++;
}
column++;
previous_count = l.Layer_Neurons.Count;
}
m.Save("Network.png", ImageFormat.Png);
return(m);
}
private void rbMNIST_CheckedChanged(object sender, EventArgs e)
{
//btnbatch.Enabled = false;
Tuple <double, double, double, double> Norm = Tuple.Create(255.0, 0.0, 9.0, 0.0);
List <double[]> input = new List <double[]>();
List <double[]> output = new List <double[]>();
var t = Data.ReadData("mnist_test.csv", 785, 1, Norm, true);
input = t.Item1;
output = t.Item2;
//https://www.kaggle.com/donfuzius/vectordigits
PerceptronNetwork network = new MLP.PerceptronNetwork(new int[] { 784, 784, 1 });
NetworkController = new MLP.Network_Control(network, output, input, double.Parse(tbLearningRate.Text), false, Norm, 784, 1);
tbStructure.Text = "784 784 1";
/*
* Tuple<double, double, double, double> Norm = Tuple.Create(255.0, 0.0, 9.0, 0.0);
* List<double[]> input = new List<double[]>();
* List<double[]> output = new List<double[]>();
* var t = Data.ReadData(path, 784, 1, Norm);
* input = t.Item1;
* output = t.Item2;
*
* int errors = 0;
* for (int i = 0; i < input.Count; i++)
* {
* for (int j = 0; j < output[i].Length; j++)
* {
* var expected = Data.inverseNormalize(output[i][j], Norm.Item4, Norm.Item3);
*
* var a = NetworkController.Network.Activate(input[i]); //Get the result from the given outputs
* var bbb = Data.inverseNormalize(a[j], Norm.Item4, Norm.Item3); //Inverse normalize the result to fit it between output values
*
* if (Math.Round(bbb) != expected) { errors++; }//Error?
* rtbConsole.Text += "Expected: " + expected + " Got :" + Math.Round(bbb) + "\n";
* rtbConsole.SelectionStart = rtbConsole.Text.Length;
* rtbConsole.ScrollToCaret();
*
* }
*
*
* }
* rtbConsole.Text += "Total Number of Errors: " + errors + "/" + input.Count*output.Count + "\n";
*
*/
}
static void Main(string[] args)
{
PerceptronNetwork p;
var t = Data.ReadData(inputPath, inputCount, outputCount, Norm, false);
input = t.Item1;
output = t.Item2;
p = new PerceptronNetwork(new int[] { inputCount, 5, 5, outputCount });
while (!p.Learn(input, output, 0.05, 0.01, 3000))
{
p = new PerceptronNetwork(new int[] { inputCount, 5, 5, outputCount });
}
/*
* MLP.Network_Control control = new MLP.Network_Control(p, output, input, 0.01, false, Norm, inputCount, outputCount);
*
* for (int i = 0; i < 3000; i++)
* {
* control.Step();
* }
*
* MLP.NetworkImage.Draw(p, p.Layers.Count);
*
* var sst = control.TestValues(1,1,1,1);
* Console.WriteLine(sst);
*
* Console.WriteLine(control.CurrentError);
*/
while (true)
{
double[] val = new double[inputCount];
for (int i = 0; i < inputCount; i++)
{
Console.WriteLine("Input Test Value " + i + ": ");
val[i] = Data.normalize(double.Parse(Console.ReadLine()), Norm.Item2, Norm.Item1);
}
double[] sal = p.Activate(val);
for (int i = 0; i < outputCount; i++)
{
Console.Write("Result: " + i + ": " + Data.inverseNormalize(sal[i], Norm.Item4, Norm.Item3) + " ");
}
Console.WriteLine("");
}
}
private void rbIris_CheckedChanged(object sender, EventArgs e)
{
//btnbatch.Enabled = true;
Tuple <double, double, double, double> Norm = Tuple.Create(10.0, 0.0, 3.0, 0.0);
List <double[]> input = new List <double[]>();
List <double[]> output = new List <double[]>();
var t = Data.ReadData("IRIS_Train.csv", 4, 1, Norm, false);
input = t.Item1;
output = t.Item2;
PerceptronNetwork network = new MLP.PerceptronNetwork(new int[] { 4, 5, 5, 1 });
NetworkController = new MLP.Network_Control(network, output, input, double.Parse(tbLearningRate.Text), false, Norm, 4, 1);
tbStructure.Text = "4 5 5 1";
}
/// <summary>
/// Network controller, static networks are non-interactive but somewhat faster
/// </summary>
/// <param name="network"></param>
/// <param name="isStatic"></param>
public Network_Control(PerceptronNetwork network, List <double[]> output, List <double[]> input, double learningRate, bool isStatic, Tuple <double, double, double, double> normalizer, int inputcount, int outputcount)
{
Network = network;
Paused = false;
CurrentIteration = 0;
CurrentError = Int32.MaxValue;
BestError = CurrentError;
Error_Historian = new List <double>();
Delay_Historian = new List <System.TimeSpan>();
LearningRate = learningRate;
Output = new List <double[]>();
Input = new List <double[]>();
InputCount = inputcount;
OutputCount = outputcount;
Output.AddRange(output);
Input.AddRange(input);
Norm = normalizer;
}
