public void ShouldTrainForXor()
{
var brain = new NeuralNetwork(2, new int[3] {
5, 5, 5
}, 1, ActivationFunction.LogSigmoid);
var deepLearning = new DeepLearning(brain, 0.00001);
var inputs = new Matrix[4];
inputs[0] = new Matrix(2, 1);
inputs[0][0, 0] = 0;
inputs[0][1, 0] = 0;
inputs[1] = new Matrix(2, 1);
inputs[1][0, 0] = 1;
inputs[1][1, 0] = 0;
inputs[2] = new Matrix(2, 1);
inputs[2][0, 0] = 0;
inputs[2][1, 0] = 1;
inputs[3] = new Matrix(2, 1);
inputs[3][0, 0] = 1;
inputs[3][1, 0] = 1;
var targets = new Matrix[4];
targets[0] = new Matrix(1, 1);
targets[0][0, 0] = 0;
targets[1] = new Matrix(1, 1);
targets[1][0, 0] = 1;
targets[2] = new Matrix(1, 1);
targets[2][0, 0] = 1;
targets[3] = new Matrix(1, 1);
targets[3][0, 0] = 0;
var rand = new Random();
for (int i = 0; i < 10000; i++)
{
var t = rand.Next(4);
deepLearning.Train(inputs[t], targets[t]);
}
var i0 = brain.Prediction(inputs[0]); // epected 0
var i2 = brain.Prediction(inputs[1]); // expected 1
var i3 = brain.Prediction(inputs[2]); // expected 1
var i4 = brain.Prediction(inputs[3]); //expected 0
Assert.IsFalse(false);
}
private void ImageCallBack_2(IntPtr m_BufForDriver, ref MyCamera.MV_FRAME_OUT_INFO_EX pFrameInfo, IntPtr pUser)
{
MyCamera.MVCC_INTVALUE stParam = new MyCamera.MVCC_INTVALUE();
int nRet = cameraArr1[0].MV_CC_GetIntValue_NET("PayloadSize", ref stParam);
if (MyCamera.MV_OK != nRet)
{
MessageBox.Show("Get PayloadSize failed", "Error", MessageBoxButtons.OK, MessageBoxIcon.Error);
return;
}
UInt32 nPayloadSize = stParam.nCurValue;
if (nPayloadSize > m_nBufSizeForDriver)
{
if (m_BufForDriver != IntPtr.Zero)
{
Marshal.Release(m_BufForDriver);
}
m_nBufSizeForDriver = nPayloadSize;
m_BufForDriver = Marshal.AllocHGlobal((Int32)m_nBufSizeForDriver);
}
if (m_BufForDriver == IntPtr.Zero)
{
MessageBox.Show("采集失败，请重新连接设备", "Error", MessageBoxButtons.OK, MessageBoxIcon.Error);
return;
}
// 将海康数据类型转为Mat
nRet = cameraArr1[0].MV_CC_GetOneFrameTimeout_NET(m_BufForDriver, nPayloadSize, ref pFrameInfo, 1000); // m_BufForDriver为图像数据接收指针
//pTemp = m_BufForDriver;
byte[] byteImage = new byte[pFrameInfo.nHeight * pFrameInfo.nWidth];
Marshal.Copy(m_BufForDriver, byteImage, 0, pFrameInfo.nHeight * pFrameInfo.nWidth);
Mat matImage = new Mat(pFrameInfo.nHeight, pFrameInfo.nWidth, MatType.CV_8UC1, byteImage);
// 单通道图像转为三通道
Mat matImageNew = new Mat();
Cv2.CvtColor(matImage, matImageNew, ColorConversionCodes.GRAY2RGB);
Bitmap bitmap = matImageNew.ToBitmap(); // Mat转为Bitmap
// 是否进行推理
DeepLearning deepLearning = new DeepLearning();
if (isInference2)
{
bitmap = deepLearning.Inference(model2, modelType, bitmap);
}
if (pictureBox2.InvokeRequired) // 当一个控件的InvokeRequired属性值为真时，说明有一个创建它以外的线程想访问它
{
UpdateUI update = delegate { pictureBox2.Image = bitmap; };
pictureBox2.BeginInvoke(update);
}
else
{
pictureBox2.Image = bitmap;
}
}
static void Main(string[] args)
{
try
{
ConsoleCopy.Create();
var sw = Stopwatch.StartNew();
DeepLearning.Test();
sw.Stop();
Console.WriteLine($"done! {sw.Elapsed}");
}
catch (Exception e)
{
Console.WriteLine(e);
}
Console.Read();
}
public void ShouldTrainForMnist()
{
// training data
Matrix[] trainingSet;
Matrix[] targetTraining;
GetDataFromMnsit(@".\Mnist dataSet\mnistyTrain.csv", out trainingSet, out targetTraining);
Matrix[] testSet;
Matrix[] testTraining;
GetDataFromMnsit(@".\Mnist dataSet\mnistTest.csv", out testSet, out testTraining);
var brain = new NeuralNetwork(trainingSet[0].Rows, new int[1] {
60
}, 10, ActivationFunction.LogSigmoid);
// brain = NeuralNetwork.Import(@"c:\", "Mnisty.json");
var deepLearning = new DeepLearning(brain, 0.001);
var MaxAccurcy = calculateAccurcy(testSet, testTraining, brain);
System.Diagnostics.Debug.WriteLine("initial accurcy: " + MaxAccurcy);
for (int j = 0; j < 10000; j++)
{
deepLearning.BulkTraining(trainingSet, targetTraining);
var Currentaccurcy = calculateAccurcy(testSet, testTraining, brain);
if (Currentaccurcy > MaxAccurcy)
{
MaxAccurcy = Currentaccurcy;
brain.Export(@"c:\", "Mnisty.json");
}
System.Diagnostics.Debug.WriteLine("cycle " + j + " current " + Currentaccurcy + " max accurcy " + MaxAccurcy);
}
Assert.IsFalse(false);
}
public void ShouldExportImportSameBrain()
{
Matrix[] testSet;
Matrix[] testTraining;
GetDataFromMnsit(@".\Mnist dataSet\mnistTest.csv", out testSet, out testTraining);
var brain = new NeuralNetwork(testSet[0].Rows, new int[1] {
60
}, 10, ActivationFunction.LogSigmoid);
brain.Export(@"c:\", "testttt");
var deepLearning = new DeepLearning(brain, 0.001);
var acc1 = calculateAccurcy(testSet, testTraining, brain);
var tmp = NeuralNetwork.Import(@"c:\", "testttt");
var acc2 = calculateAccurcy(testSet, testTraining, tmp);
Assert.AreEqual(acc1, acc2);
}
private void OnImageGrabbed_2(Object sender, ImageGrabbedEventArgs e)
{
try
{
IGrabResult grabResult = e.GrabResult;
if (grabResult.IsValid)
{
// 四通道RGBA
Bitmap bitmap = new Bitmap(grabResult.Width, grabResult.Height, PixelFormat.Format32bppRgb);
// 锁定位图的位
BitmapData bmpData = bitmap.LockBits(new Rectangle(0, 0, bitmap.Width, bitmap.Height), ImageLockMode.ReadWrite, bitmap.PixelFormat);
// 将指针放置到位图的缓冲区
converter.OutputPixelFormat = PixelType.BGRA8packed;
IntPtr ptrBmp = bmpData.Scan0;
converter.Convert(ptrBmp, bmpData.Stride * bitmap.Height, grabResult);
bitmap.UnlockBits(bmpData);
DeepLearning deepLearning = new DeepLearning();
if (isInference2)
{
bitmap = deepLearning.Inference(model2, modelType, bitmap);
}
if (pictureBox2.InvokeRequired) // 当一个控件的InvokeRequired属性值为真时，说明有一个创建它以外的线程想访问它
{
UpdateUI update = delegate { pictureBox2.Image = bitmap; };
pictureBox2.BeginInvoke(update);
}
else
{
pictureBox2.Image = bitmap;
}
}
}
catch (Exception ex)
{
MessageBox.Show(string.Format("CAM:{0}！\n{1}", "CAM2", ex.Message));
}
finally
{
e.DisposeGrabResultIfClone();
}
}
static void Main(string[] args)
{
int nInput = 2;
int nOutput = 1;
int neuronsPerLayers = 10;
int nLayers = 3;
DeepLearning slp = new DeepLearning(nInput, neuronsPerLayers, nOutput, nLayers);
//SingleLayerPerceptron slp = new SingleLayerPerceptron(nInput, neuronsPerLayers, nOutput);
var nPoints = 10000;
for (int i = 0; i < nPoints; i++)
{
Random random = new Random(Guid.NewGuid().GetHashCode());
int value1 = random.Next() % 2;
random = new Random(Guid.NewGuid().GetHashCode());
int value2 = random.Next() % 2;
random = new Random(Guid.NewGuid().GetHashCode());
int value3 = random.Next() % 2;
//int xor = XOR(value1, value2);
int correctValue = ComplexeBoolMethod(value1, value2, value3);
slp.Train(new List <double>()
{
value1, value2
}, new List <double>()
{
correctValue
});
var percent = Math.Round(i * 100.0 / nPoints, 4);
//if (percent == (int)percent)
//{
// Console.Clear();
// Console.WriteLine(percent);
//}
}
double n = 0;
int nTestCase = 200;
for (int i = 0; i < nTestCase; i++)
{
Random random = new Random(Guid.NewGuid().GetHashCode());
int value1 = random.Next() % 2;
random = new Random(Guid.NewGuid().GetHashCode());
int value2 = random.Next() % 2;
random = new Random(Guid.NewGuid().GetHashCode());
int value3 = random.Next() % 2;
//int xor = XOR(value1, value2);
int correctValue = ComplexeBoolMethod(value1, value2, value3);
var guess = slp.Predict(new List <double>()
{
value1, value2
});
double result = guess[0] > 0.5 ? 1 : 0;
bool correct = result == correctValue;
Console.WriteLine("correct value : " + correctValue);
Console.WriteLine("guess : " + guess[0]);
Console.WriteLine("--");
if (correct)
{
n++;
}
}
var path = "C:\\Users\\Boudart Loic\\source\\repos\\MachineLearning\\MachineLearning\\brain.json";
slp.Save(path);
slp.Load(path);
//to debug
{
Random random = new Random(Guid.NewGuid().GetHashCode());
int value1 = random.Next() % 2;
random = new Random(Guid.NewGuid().GetHashCode());
int value2 = random.Next() % 2;
int xor = XOR(value1, value2);
slp.Train(new List <double>()
{
value1, value2
}, new List <double>()
{
xor
});
//if (percent == (int)percent)
//{
// Console.Clear();
// Console.WriteLine(percent);
//}
}
Console.WriteLine(n / nTestCase);
Console.Read();
}
public static void Learning(int from, int to)
{
DeepLearning.PrepareData($@"E:\projects\MonkeyOthello\tests\", from, to);
DeepLearning.TrainAll(from, to);
DeepLearning.TestAll(from, to);
}
