static double test2()
{
LayerData.FullyConnected inputLayer = new LayerData.FullyConnected()
{
cntNeurons = 3,
tFuncType = TransferFuncType.NONE
};
LayerData.FullyConnected h1 = new LayerData.FullyConnected()
{
cntNeurons = 10,
tFuncType = TransferFuncType.RATIONALSIGMOID
};
LayerData.FullyConnected outputLayer = new LayerData.FullyConnected()
{
cntNeurons = 1,
tFuncType = TransferFuncType.TANH
};
int limit = 2000;
NeuralNetwork nn = new NeuralNetwork(limit * 25, 1, 0.01, 0.1, true, inputLayer, outputLayer);
double error = 0;
Random gen = new Random();
for (int i = 0; i < limit; i++)
{
error = 0;
bool displayOutput = false;
if (i % 1 == 0)
{
nn.RegisterOutput("__________");
displayOutput = true;
}
for (int j = 0; j < 25; j++)
{
double input = gen.NextDouble();
error += nn.Train(new double[] { input, 1, 1 }, new double[] { Math.Tanh(input) }, false);
}
if (displayOutput)
{
nn.RegisterOutput(string.Format("Error : {0}", error));
}
Thread.Sleep(100);
}
nn.WaitTillDone();
return(error);
}
static double test1()
{
LayerData.FullyConnected inputLayer = new LayerData.FullyConnected()
{
cntNeurons = 2,
tFuncType = TransferFuncType.NONE
};
LayerData.FullyConnected hidden = new LayerData.FullyConnected()
{
cntNeurons = 20,
tFuncType = TransferFuncType.RECTILINEAR
};
LayerData.FullyConnected outputLayer = new LayerData.FullyConnected()
{
cntNeurons = 1,
tFuncType = TransferFuncType.SIGMOID
};
int limit = 100;
NeuralNetwork nn = new NeuralNetwork(limit * 4, 1, 0.25, 0.1, true, inputLayer, hidden, outputLayer);
double error = 0;
for (int i = 0; i < limit; i++)
{
error = 0;
bool displayOutput = false;
if (i % (limit > 10 ? limit / 10 : 1) == 0)
{
Console.WriteLine("__________");
displayOutput = true;
}
error += nn.Train(new double[] { 0, 0 }, new double[] { 0 }, displayOutput);
error += nn.Train(new double[] { 0, 1 }, new double[] { 1 }, displayOutput);
error += nn.Train(new double[] { 1, 0 }, new double[] { 1 }, displayOutput);
error += nn.Train(new double[] { 1, 1 }, new double[] { 0 }, displayOutput);
if (displayOutput)
{
Console.WriteLine("Error : {0}", error);
}
Thread.Sleep(10);
}
return(error);
}
static double test3()
{
int limit = 0;
int height, width;
Dictionary <int, byte[, ]> data = new Dictionary <int, byte[, ]>();
Dictionary <int, byte> label = new Dictionary <int, byte>();
// Read Data
byte[] imgBytes = File.ReadAllBytes(@"train-images.idx3-ubyte");
byte[] labelBytes = File.ReadAllBytes(@"train-labels.idx1-ubyte");
int idx1 = 0, idx2 = 8;
int magicNumber = ReadInt(ref imgBytes, ref idx1);
limit = ReadInt(ref imgBytes, ref idx1);
height = ReadInt(ref imgBytes, ref idx1);
width = ReadInt(ref imgBytes, ref idx1);
for (int i = 0; i < limit; i++)
{
data[i] = new byte[height, width];
label[i] = labelBytes[idx2++];
for (int j = 0; j < height; j++)
{
for (int k = 0; k < width; k++)
{
data[i][j, k] = imgBytes[idx1++];
}
}
}
LayerData.FullyConnected inputLayer = new LayerData.FullyConnected()
{
cntNeurons = 784,
tFuncType = TransferFuncType.NONE
};
LayerData.FullyConnected hidden = new LayerData.FullyConnected()
{
cntNeurons = 20,
tFuncType = TransferFuncType.SIGMOID
};
LayerData.Convolutional conv1 = new LayerData.Convolutional()
{
filters = new int[]
{
14, 14, 14, 14, 14, 14
},
stride = 2,
padding = false
};
LayerData.MaxPool maxPool1 = new LayerData.MaxPool()
{
size = 2,
stride = 2
};
LayerData.Convolutional conv2 = new LayerData.Convolutional()
{
filters = new int[]
{
7, 7, 7, 7, 7, 7
},
stride = 1,
padding = false
};
LayerData.FullyConnected fully1 = new LayerData.FullyConnected()
{
cntNeurons = 30,
tFuncType = TransferFuncType.RECTILINEAR
};
LayerData.FullyConnected outputLayer = new LayerData.FullyConnected()
{
cntNeurons = 10,
tFuncType = TransferFuncType.SOFTMAX
};
NeuralNetwork nn = new NeuralNetwork(limit, 1000, 0.0001, 0.0, true, inputLayer, conv1, conv2, fully1, outputLayer);
//nn.Save("temp.xml", "temp");
//NeuralNetwork nn = NeuralNetwork.Load("Temp\\400.xml", true);
//nn.Save("TestingSaveModule.xml", "CNNTEST");
//NeuralNetwork nn = NeuralNetwork.Load("temp.xml", true);
//NeuralNetwork nn = new NeuralNetwork(new int[] { 784, 20, 10 },
// new TransferFuncType[] { TransferFuncType.NONE, TransferFuncType.SIGMOID, TransferFuncType.SOFTMAX }, 60000, 100);
nn.batchSize = 1;
double error = 0;
Random randGen = new Random();
//int saveCnt = 400;
//nn.Save("Temp\\" + saveCnt++ + ".xml", "ForGraph");
//limit = 180000;
int[] wrongDetectedCnt = new int[10];
for (int i = 0; i < 10; i++)
{
wrongDetectedCnt[i] = 0;
}
for (int i = nn.trainingCounter; i < limit; i++)
{
//learningRate = 0.005 * (1.00001 - (double)i / limit);
//momentum = 0.005 * (1.00001 - (double)i / limit);
error = 0;
bool displayOutput = false;
//if (i % (limit > 10 ? limit / 1000 : 1) == 0)
//{
// nn.RegisterOutput("_____________");
// displayOutput = true;
//}
double[] input = new double[height * width];
int idx = (int)(randGen.NextDouble() * 59999);
for (int j = 0; j < height; j++)
{
for (int k = 0; k < width; k++)
{
input[j * width + k] = (data[idx][j, k] / 255.0);// >= 140 ? 1 : 0);
}
}
double[] output = new double[10];
for (int j = 0; j < 10; j++)
{
output[j] = 0;
}
output[label[idx]] = 1;
if (displayOutput)
{
nn.RegisterOutput(string.Format("Label: {0}", label[i]));
}
error = nn.Train(input, output, displayOutput);
if (error > 0.4)
{
wrongDetectedCnt[label[i]]++;
}
if (displayOutput)
{
nn.RegisterOutput(string.Format("Error : {0}", error));
}
if (i % 100 == 0 && i != 0)
{
nn.Save("Temporary.xml", "Testing");
}
//nn.Save("Temp\\" + saveCnt++ + ".xml", "ForGraph");
if (i % 50 == 0 && i != 0)
{
for (int j = 0; j < 10; j++)
{
Console.WriteLine("{0} : {1}", j, wrongDetectedCnt[j]);
}
Console.WriteLine("**********************");
}
}
nn.Save("Testing.xml", "DigitRecognizer");
return(error);
}
