public ExpressionsDialog(ref TrainingDataSet DataSet)
{
InitializeComponent();
_DataSet = DataSet;
ExpressionsDataGrid.ItemsSource = _DataSet.Expression;
}
public TrainingDataSetsDialog(ref TrainingDataSet DataSet)
{
InitializeComponent();
_DataSet = DataSet;
TrainingDataGrid.ItemsSource = _DataSet.Training;
}
public void PerformEpochAndAssert()
{
var trainingData = TrainingDataSet.For(inputs, outputs);
var errorRate = trainingMethod.PerformSingleEpochProducingErrorRate(trainingData);
AssertResultOfTrainingEpoch(errorRate);
}
public BatchClassificationDialog(ref Klu klu, ref TrainingDataSet dataSet, ProcessOptions processOptions)
{
InitializeComponent();
_BackgroundWorker = new BackgroundWorker();
_BackgroundWorker.WorkerReportsProgress = true;
_BackgroundWorker.WorkerSupportsCancellation = true;
_BackgroundWorker.DoWork += new DoWorkEventHandler(DoWork);
_BackgroundWorker.RunWorkerCompleted += new RunWorkerCompletedEventHandler(RunWorkerCompleted);
_BackgroundWorker.ProgressChanged += new ProgressChangedEventHandler(ProgressChanged);
BrowseButton.IsEnabled = true;
CancelButton.IsEnabled = false;
ClassifyButton.IsEnabled = false;
_DataSet = dataSet;
_SelectedFiles = new ArrayList();
_ProcessOptions = processOptions;
// We want the images to be as less distorded as possible, so we disable all the overlays.
_ProcessOptions.DrawAnthropometricPoints = 0;
_ProcessOptions.DrawDetectionTime = 0;
_ProcessOptions.DrawFaceRectangle = 0;
_ProcessOptions.DrawSearchRectangles = 0;
_ProcessOptions.DrawFeaturePoints = 0;
_KLU = klu;
_FFP = new FaceFeaturePoints();
_TempBitmap = new System.Drawing.Bitmap(10, 10);
ExpressionsComboBox.ItemsSource = _DataSet.Expression;
ClassifyButton.Content = "Classify";
}
public ExpressionsDialog(ref TrainingDataSet DataSet)
{
InitializeComponent();
_DataSet = DataSet;
ExpressionsDataGrid.ItemsSource = _DataSet.Expression;
}
public BatchClassificationDialog(ref Klu klu, ref TrainingDataSet dataSet, ProcessOptions processOptions)
{
InitializeComponent();
_BackgroundWorker = new BackgroundWorker();
_BackgroundWorker.WorkerReportsProgress = true;
_BackgroundWorker.WorkerSupportsCancellation = true;
_BackgroundWorker.DoWork += new DoWorkEventHandler(DoWork);
_BackgroundWorker.RunWorkerCompleted += new RunWorkerCompletedEventHandler(RunWorkerCompleted);
_BackgroundWorker.ProgressChanged += new ProgressChangedEventHandler(ProgressChanged);
BrowseButton.IsEnabled = true;
CancelButton.IsEnabled = false;
ClassifyButton.IsEnabled = false;
_DataSet = dataSet;
_SelectedFiles = new ArrayList();
_ProcessOptions = processOptions;
// We want the images to be as less distorded as possible, so we disable all the overlays.
_ProcessOptions.DrawAnthropometricPoints = 0;
_ProcessOptions.DrawDetectionTime = 0;
_ProcessOptions.DrawFaceRectangle = 0;
_ProcessOptions.DrawSearchRectangles = 0;
_ProcessOptions.DrawFeaturePoints = 0;
_KLU = klu;
_FFP = new FaceFeaturePoints();
_TempBitmap = new System.Drawing.Bitmap(10, 10);
ExpressionsComboBox.ItemsSource = _DataSet.Expression;
ClassifyButton.Content = "Classify";
}
private void DisplayData_Click(object sender, RoutedEventArgs e)
{
// Get selected
if (this.TrainingDataList.SelectedItem == null)
{
MessageBox.Show("Please select a data set first!");
return;
}
// Reset View
this.TrainingDataSet.Clear();
this.Hypothesis.Clear();
// Assign training data
switch (this.TrainingDataList.SelectedIndex)
{
case 0:
this.CurrentTestData = TrainingData.CourseraTs;
break;
case 1:
this.CurrentTestData = TrainingData.TestDataTs;
break;
case 2:
this.CurrentTestData = TrainingData.ExcelTs;
break;
case 3:
this.CurrentTestData = TrainingData.OpenClassRoomTs;
break;
}
// Get training data
for (int pointIndex = 0; pointIndex < CurrentTestData.GetLength(0); pointIndex++)
{
Point point = new Point(CurrentTestData[pointIndex, 0], CurrentTestData[pointIndex, 1]);
TrainingDataSet.Add(new ObservablePoint(point.X, point.Y));
Debug.OutPut(point);
}
// Get iterations
int iterations;
try
{
iterations = int.Parse(this.IterationsText.Text);
}
catch (Exception)
{
MessageBox.Show("Please enter a correct iteration int!");
return;
}
// Run new thread with the gradient algorithm
ThreadPool.QueueUserWorkItem(o =>
{
MachineLearningCalculator.GradientDescent(this.CurrentTestData, 0, 0, this.Hypothesis, iterations);
});
}
public BackPropagationPerformanceComparisonContainer()
{
var neuralNetworkUnderTest =
NeuralNetwork
.For(NeuralNetworkContext.MaximumPrecision)
.WithInputLayer(neuronCount: 5, activationType: ActivationType.Sigmoid)
.WithHiddenLayer(neuronCount: 100, activationType: ActivationType.Sigmoid)
.WithHiddenLayer(neuronCount: 70, activationType: ActivationType.TanH)
.WithHiddenLayer(neuronCount: 40, activationType: ActivationType.TanH)
.WithHiddenLayer(neuronCount: 100, activationType: ActivationType.Sigmoid)
.WithOutputLayer(neuronCount: 2, activationType: ActivationType.Sigmoid)
.Build();
trainingData = new[] {
TrainingDataSet.For(new [] { 0.78, 0.99, 0.67, 0.72, 0.22 }, new [] { 0.12, 0.14 })
};
multiThreadedController = TrainingController.For(
BackPropagation
.WithConfiguration(neuralNetworkUnderTest, ParallelOptionsExtensions.UnrestrictedMultiThreadedOptions));
singleThreadedController = TrainingController.For(
BackPropagation
.WithConfiguration(neuralNetworkUnderTest, ParallelOptionsExtensions.SingleThreadedOptions));
}
public TrainingDataSetsDialog(ref TrainingDataSet DataSet)
{
InitializeComponent();
_DataSet = DataSet;
TrainingDataGrid.ItemsSource = _DataSet.Training;
}
/// <summary>
/// No conversion is done here, but a list of all Expressions is returned.
/// </summary>
/// <param name="value"></param>
/// <param name="targetType"></param>
/// <param name="parameter"></param>
/// <param name="culture"></param>
/// <returns>Expressions</returns>
public object Convert(object value, Type targetType, object parameter, CultureInfo culture)
{
try
{
_TAM = new TableAdapterManager();
_DataSet = new TrainingDataSet();
_TAM.ExpressionTableAdapter = new ExpressionTableAdapter();
_TAM.ExpressionTableAdapter.Fill(_DataSet.Expression);
return(_DataSet.Expression);
//TableAdapterManager _TAM = new TableAdapterManager();
//TrainingDataSet _DataSet = new TrainingDataSet();
//_TAM.ExpressionTableAdapter = new ExpressionTableAdapter();
//_TAM.ExpressionTableAdapter.Fill(_DataSet.Expression);
//ArrayList expressions = new ArrayList();
//for (int i=0; i<_DataSet.Expression.Count; i++)
//{
// expressions.Add(_DataSet.Expression[i].Expression);
//}
//return expressions;
}
catch (NullReferenceException)
{
return(null);
}
}
/// <summary>
/// No conversion is done here, but a list of all Expressions is returned.
/// </summary>
/// <param name="value"></param>
/// <param name="targetType"></param>
/// <param name="parameter"></param>
/// <param name="culture"></param>
/// <returns>Expressions</returns>
public object Convert(object value, Type targetType, object parameter, CultureInfo culture)
{
try
{
_TAM = new TableAdapterManager();
_DataSet = new TrainingDataSet();
_TAM.ExpressionTableAdapter = new ExpressionTableAdapter();
_TAM.ExpressionTableAdapter.Fill(_DataSet.Expression);
return _DataSet.Expression;
//TableAdapterManager _TAM = new TableAdapterManager();
//TrainingDataSet _DataSet = new TrainingDataSet();
//_TAM.ExpressionTableAdapter = new ExpressionTableAdapter();
//_TAM.ExpressionTableAdapter.Fill(_DataSet.Expression);
//ArrayList expressions = new ArrayList();
//for (int i=0; i<_DataSet.Expression.Count; i++)
//{
// expressions.Add(_DataSet.Expression[i].Expression);
//}
//return expressions;
}
catch (NullReferenceException)
{
return null;
}
}
public async Task <InvokeResult> UpdateTrainingDataSetManager(TrainingDataSet set, EntityHeader org, EntityHeader user)
{
await AuthorizeAsync(set, AuthorizeResult.AuthorizeActions.Update, user, org);
var result = Validator.Validate(set, Actions.Update);
await _repo.UpdateTrainingDataSetsAsync(set);
return(result.ToInvokeResult());
}
public async Task <InvokeResult> AddTrainingDataSetManager(TrainingDataSet dataSet, EntityHeader org, EntityHeader user)
{
ValidationCheck(dataSet, Actions.Create);
await AuthorizeAsync(dataSet, AuthorizeResult.AuthorizeActions.Create, user, org);
await _repo.AddTrainingDataSetsAsync(dataSet);
return(InvokeResult.Success);
}
public static TrainingDataSet ToTrainingData(this BankTransaction bankTransaction)
{
var inputs = bankTransaction.ToNetworkInputData();
var outputs = Enumerable.Repeat(0.0, 2).ToList();
outputs[Convert.ToInt32(bankTransaction.Class)] = 1;
return(TrainingDataSet.For(inputs.ToArray(), outputs.ToArray()));
}
public static TrainingDataSet TrainingDataFromRow(IrisDataRow row)
{
var species = Enumerable.Repeat(0.0, 3).ToArray();
species[Convert.ToInt32(row.Species) - 1] = 1.0;
return(TrainingDataSet.For(
row.PredictionDataPoints,
species.ToArray()));
}
private static List <TrainingDataSet> GetMyTrainingData()
{
// 2 tableaux
// inputs : double[22][64]
// outputs : double[22][3]
string filename = BASEFOLDER + "Base.txt";
string[] lines = File.ReadAllLines(filename); // lines va contenir 44 lines
int nbSamples = lines.Length / 2; // 22 exemples
double[][] inputs = new double[nbSamples][];
double[][] outputs = new double[nbSamples][];
for (int i = 0; i < nbSamples; i++)
{
inputs[i] = new double[64]; // inputs[i] est de type double[]
outputs[i] = new double[3];
// lines[2*i+0] contient la string "0000010101001001001010101010"
// lines[2*i+1] contient la string "C"
int k = 0;
foreach (char c in lines[2 * i])
{
if (c == '0')
{
inputs[i][k] = 0.0;
}
if (c == '1')
{
inputs[i][k] = 1.0;
}
k++;
}
if (lines[2 * i + 1] == "A")
{
outputs[i][0] = 1.0; outputs[i][1] = 0.0; outputs[i][2] = 0.0;
}
if (lines[2 * i + 1] == "B")
{
outputs[i][0] = 0.0; outputs[i][1] = 1.0; outputs[i][2] = 0.0;
}
if (lines[2 * i + 1] == "C")
{
outputs[i][0] = 0.0; outputs[i][1] = 0.0; outputs[i][2] = 1.0;
}
}
return(inputs.Select((input, i) => TrainingDataSet.For(input, outputs[i])).ToList());
}
private static List <TrainingDataSet> GetXorTrainingData()
{
var inputs = new[]
{
new[] { 0.0, 0.0 }, new[] { 0.0, 1.0 }, new[] { 1.0, 0.0 }, new[] { 1.0, 1.0 }
};
var outputs = new[]
{
new[] { 0.0 }, new[] { 1.0 }, new[] { 1.0 }, new[] { 0.0 }
};
return(inputs.Select((input, i) => TrainingDataSet.For(input, outputs[i])).ToList());
}
public object Convert(object value, Type targetType, object parameter, CultureInfo culture)
{
try
{
TableAdapterManager tam = new TableAdapterManager();
TrainingDataSet dataSet = new TrainingDataSet();
tam.ExpressionTableAdapter = new ExpressionTableAdapter();
tam.ExpressionTableAdapter.Fill(dataSet.Expression);
int eoid = System.Convert.ToInt32(value);
if (!DBNull.Value.Equals(eoid))
{
TrainingDataSet.ExpressionRow row = dataSet.Expression.FindByExpressionOID(eoid);
if (!DBNull.Value.Equals(row) && !DBNull.Value.Equals(row.Thumbnail))
{
byte[] imageData = row.Thumbnail;
if (imageData != null && !DBNull.Value.Equals(imageData))
{
byte[] data = (byte[])value;
MemoryStream ms = new MemoryStream(data);
BitmapImage image = new BitmapImage();
image.BeginInit();
image.StreamSource = ms;
image.EndInit();
return(image);
}
}
}
return(null);
}
catch (NullReferenceException)
{
return(null);
}
catch (System.Data.StrongTypingException)
{
return(null);
}
}
public object Convert(object value, Type targetType, object parameter, CultureInfo culture)
{
try
{
TableAdapterManager tam = new TableAdapterManager();
TrainingDataSet dataSet = new TrainingDataSet();
tam.ExpressionTableAdapter = new ExpressionTableAdapter();
tam.ExpressionTableAdapter.Fill(dataSet.Expression);
return(dataSet.Expression.FindByExpressionOID(System.Convert.ToInt32(value)).Expression);
}
catch (NullReferenceException)
{
return(null);
}
}
public object Convert(object value, Type targetType, object parameter, CultureInfo culture)
{
try
{
TableAdapterManager tam = new TableAdapterManager();
TrainingDataSet dataSet = new TrainingDataSet();
tam.ExpressionTableAdapter = new ExpressionTableAdapter();
tam.ExpressionTableAdapter.Fill(dataSet.Expression);
int eoid = System.Convert.ToInt32(value);
if (!DBNull.Value.Equals(eoid))
{
TrainingDataSet.ExpressionRow row = dataSet.Expression.FindByExpressionOID(eoid);
if (!DBNull.Value.Equals(row) && !DBNull.Value.Equals(row.Thumbnail))
{
byte[] imageData = row.Thumbnail;
if (imageData != null && !DBNull.Value.Equals(imageData))
{
byte[] data = (byte[])value;
MemoryStream ms = new MemoryStream(data);
BitmapImage image = new BitmapImage();
image.BeginInit();
image.StreamSource = ms;
image.EndInit();
return image;
}
}
}
return null;
}
catch (NullReferenceException)
{
return null;
}
catch (System.Data.StrongTypingException)
{
return null;
}
}
public object Convert(object value, Type targetType, object parameter, CultureInfo culture)
{
try
{
TableAdapterManager tam = new TableAdapterManager();
TrainingDataSet dataSet = new TrainingDataSet();
tam.ExpressionTableAdapter = new ExpressionTableAdapter();
tam.ExpressionTableAdapter.Fill(dataSet.Expression);
return dataSet.Expression.FindByExpressionOID(System.Convert.ToInt32(value)).Expression;
}
catch (NullReferenceException)
{
return null;
}
}
public async Task Train()
{
await Task.Run(new Action(() =>
{
Console.WriteLine();
for (int i = 0; i < 50; i++)
{
Console.WriteLine("Run Epoch {0}", i);
Network.IsPreTraining = true;
DoEpoch(TrainingDataSet.Take(500));
Network.IsPreTraining = false;
DoEpoch(TrainingDataSet);
}
Console.WriteLine("Complete.");
}
));
}
public TrainingDialog(ref TrainingDataSet dataSet)
{
InitializeComponent();
_BackgroundWorker = new BackgroundWorker();
_BackgroundWorker.WorkerReportsProgress = true;
_BackgroundWorker.WorkerSupportsCancellation = true;
_BackgroundWorker.DoWork += new DoWorkEventHandler(DoWork);
_BackgroundWorker.RunWorkerCompleted += new RunWorkerCompletedEventHandler(RunWorkerCompleted);
_BackgroundWorker.ProgressChanged += new ProgressChangedEventHandler(ProgressChanged);
_DataSet = dataSet;
_Chart = new Chart();
windowsFormsHost1.Child = _Chart;
_Series = 0;
StopButton.IsEnabled = false;
_Network = new ActivationNetwork(
new SigmoidFunction(),
_NumInputNeurons,
9,
_NumOutputNeurons
);
#region Initialize ANN stuff
_ANN = new ANN();
_ANN.NumLayers = 3;
_ANN.SetNumNeurons(0, _NumInputNeurons);
_ANN.SetNumNeurons(1, 9);
_ANN.SetNumNeurons(2, _NumOutputNeurons);
// Bind DataGrid to ANN-DataSet now
_DataSetANN = new DataSet("HiddenLayer");
_DataSetANN.Tables.Add("HiddenLayerTable");
uint tmp = 9;
_DataSetANN.Tables[0].Columns.Add("Neurons", tmp.GetType());
_DataSetANN.Tables[0].Rows.Add(tmp);
HiddenLayerDataGrid.DataContext = _DataSetANN.Tables[0];
#endregion
}
public TrainingDialog(ref TrainingDataSet dataSet)
{
InitializeComponent();
_BackgroundWorker = new BackgroundWorker();
_BackgroundWorker.WorkerReportsProgress = true;
_BackgroundWorker.WorkerSupportsCancellation = true;
_BackgroundWorker.DoWork += new DoWorkEventHandler(DoWork);
_BackgroundWorker.RunWorkerCompleted += new RunWorkerCompletedEventHandler(RunWorkerCompleted);
_BackgroundWorker.ProgressChanged += new ProgressChangedEventHandler(ProgressChanged);
_DataSet = dataSet;
_Chart = new Chart();
windowsFormsHost1.Child = _Chart;
_Series = 0;
StopButton.IsEnabled = false;
_Network = new ActivationNetwork(
new SigmoidFunction(),
_NumInputNeurons,
9,
_NumOutputNeurons
);
#region Initialize ANN stuff
_ANN = new ANN();
_ANN.NumLayers = 3;
_ANN.SetNumNeurons(0, _NumInputNeurons);
_ANN.SetNumNeurons(1, 9);
_ANN.SetNumNeurons(2, _NumOutputNeurons);
// Bind DataGrid to ANN-DataSet now
_DataSetANN = new DataSet("HiddenLayer");
_DataSetANN.Tables.Add("HiddenLayerTable");
uint tmp = 9;
_DataSetANN.Tables[0].Columns.Add("Neurons", tmp.GetType());
_DataSetANN.Tables[0].Rows.Add(tmp);
HiddenLayerDataGrid.DataContext = _DataSetANN.Tables[0];
#endregion
}
public BatchClassification(ref Klu klu, ref TrainingDataSet dataSet, ProcessOptions processOptions)
{
InitializeComponent();
_DataSet = dataSet;
_SelectedFiles = new ArrayList();
_ProcessOptions = processOptions;
// We want the images to be as less distorded as possible, so we disable all the overlays.
_ProcessOptions.DrawAnthropometricPoints = 0;
_ProcessOptions.DrawDetectionTime = 0;
_ProcessOptions.DrawFaceRectangle = 0;
_ProcessOptions.DrawSearchRectangles = 0;
_ProcessOptions.DrawFeaturePoints = 0;
_KLU = klu;
_FFP = new FaceFeaturePoints();
_TempBitmap = new System.Drawing.Bitmap(10, 10);
ExpressionsComboBox.ItemsSource = _DataSet.Expression;
ClassifyButton.Content = "Classify";
}
public BatchClassification(ref Klu klu, ref TrainingDataSet dataSet, ProcessOptions processOptions)
{
InitializeComponent();
_DataSet = dataSet;
_SelectedFiles = new ArrayList();
_ProcessOptions = processOptions;
// We want the images to be as less distorded as possible, so we disable all the overlays.
_ProcessOptions.DrawAnthropometricPoints = 0;
_ProcessOptions.DrawDetectionTime = 0;
_ProcessOptions.DrawFaceRectangle = 0;
_ProcessOptions.DrawSearchRectangles = 0;
_ProcessOptions.DrawFeaturePoints = 0;
_KLU = klu;
_FFP = new FaceFeaturePoints();
_TempBitmap = new System.Drawing.Bitmap(10, 10);
ExpressionsComboBox.ItemsSource = _DataSet.Expression;
ClassifyButton.Content = "Classify";
}
public void GenerateTrainingAndTestSets()
{
TrainingSet = new TrainingDataSet(_dataSet, NumberOfInputs);
TestSet = new TestDataSet(_dataSet, NumberOfInputs);
}
public static void AddTrainingData(ref TrainingDataSet dataSet, ref FaceFeaturePoints ffp, int expressionOID, ref System.Drawing.Bitmap picture)
{
// Convert facial coordinates to relative face rectangle coordinates.
// We do this because we are not interested in absolute coordinates but in quality of the face.
int x = ffp.FaceRectangle.X;
int y = ffp.FaceRectangle.Y;
int w = ffp.FaceRectangle.Width;
int h = ffp.FaceRectangle.Height;
// Calculate the eye distance (in relative coordinates
float dx = i2r(ffp.RightEye.EyeCenter.X, x, w) - i2r(ffp.LeftEye.EyeCenter.X, x, w);
float dy = i2r(ffp.RightEye.EyeCenter.Y, y, h) - i2r(ffp.LeftEye.EyeCenter.Y, y, h);
//float dx = i2r(ffp.RightEye.EyeCenter.X - ffp.LeftEye.EyeCenter.X, x, w);
//float dy = i2r(ffp.RightEye.EyeCenter.Y - ffp.LeftEye.EyeCenter.Y, y, h);
float eyeDist = (float)Math.Sqrt(dx * dx + dy * dy);
// Construct thumbnail
const int thumbnailWidth = 50;
const int thumbnailHeight = 50;
System.Drawing.Rectangle faceRect = new System.Drawing.Rectangle(x, y, w, h);
System.Drawing.Bitmap faceImg = picture.Clone(faceRect, picture.PixelFormat);
System.Drawing.Image.GetThumbnailImageAbort tc = new System.Drawing.Image.GetThumbnailImageAbort(ThumbnailCallback);
System.Drawing.Image thumbnail = faceImg.GetThumbnailImage(thumbnailWidth, thumbnailHeight, tc, IntPtr.Zero);
System.IO.MemoryStream ms = new System.IO.MemoryStream();
thumbnail.Save(ms, System.Drawing.Imaging.ImageFormat.Png);
TrainingDataSet.TrainingRow row = dataSet.Training.NewTrainingRow();
row.ExpressionOID = expressionOID;
row.LipCornerLeftX = i2r(ffp.Mouth.LipCornerLeft.X, x, w);
row.LipCornerLeftY = i2r(ffp.Mouth.LipCornerLeft.Y, y, h);
row.LipCornerRightX = i2r(ffp.Mouth.LipCornerRight.X, x, w);
row.LipCornerRightY = i2r(ffp.Mouth.LipCornerRight.Y, y, h);
row.LipUpLeftX = i2r(ffp.Mouth.LipUpLeft.X, x, w);
row.LipUpLeftY = i2r(ffp.Mouth.LipUpLeft.Y, y, h);
row.LipUpCenterX = i2r(ffp.Mouth.LipUpCenter.X, x, w);
row.LipUpCenterY = i2r(ffp.Mouth.LipUpCenter.Y, y, h);
row.LipUpRightX = i2r(ffp.Mouth.LipUpRight.X, x, w);
row.LipUpRightY = i2r(ffp.Mouth.LipUpRight.Y, y, h);
row.LipUpRightX = i2r(ffp.Mouth.LipBottomLeft.X, x, w);
row.LipUpRightY = i2r(ffp.Mouth.LipBottomLeft.Y, y, h);
row.LipBottomCenterX = i2r(ffp.Mouth.LipBottomCenter.X, x, w);
row.LipBottomCenterY = i2r(ffp.Mouth.LipBottomCenter.Y, y, h);
row.LipBottomRightX = i2r(ffp.Mouth.LipBottomRight.X, x, w);
row.LipBottomRightY = i2r(ffp.Mouth.LipBottomRight.Y, y, h);
row.LipBottomLeftX = i2r(ffp.Mouth.LipBottomLeft.X, x, w);
row.LipBottomLeftY = i2r(ffp.Mouth.LipBottomLeft.Y, y, h);
row.EyeDistance = eyeDist;
row.LeftEyeCenterX = i2r(ffp.LeftEye.EyeCenter.X, x, w);
row.LeftEyeCenterY = i2r(ffp.LeftEye.EyeCenter.Y, y, h);
row.LeftLidBottomX = i2r(ffp.LeftEye.LidBottomCenter.X, x, w);
row.LeftLidBottomY = i2r(ffp.LeftEye.LidBottomCenter.Y, y, h);
row.LeftLidCornerLeftX = i2r(ffp.LeftEye.LidCornerLeft.X, x, w);
row.LeftLidCornerLeftY = i2r(ffp.LeftEye.LidCornerLeft.Y, y, h);
row.LeftLidCornerRightX = i2r(ffp.LeftEye.LidCornerRight.X, x, w);
row.LeftLidCornerRightY = i2r(ffp.LeftEye.LidCornerRight.Y, y, h);
row.LeftLidUpX = i2r(ffp.LeftEye.LidUpCenter.X, x, w);
row.LeftLidUpY = i2r(ffp.LeftEye.LidUpCenter.Y, y, h);
row.RightEyeCenterX = i2r(ffp.RightEye.EyeCenter.X, x, w);
row.RightEyeCenterY = i2r(ffp.RightEye.EyeCenter.Y, y, h);
row.RightLidBottomX = i2r(ffp.RightEye.LidBottomCenter.X, x, w);
row.RightLidBottomY = i2r(ffp.RightEye.LidBottomCenter.Y, y, h);
row.RightLidCornerLeftX = i2r(ffp.RightEye.LidCornerLeft.X, x, w);
row.RightLidCornerLeftY = i2r(ffp.RightEye.LidCornerLeft.Y, y, h);
row.RightLidCornerRightX = i2r(ffp.RightEye.LidCornerRight.X, x, w);
row.RightLidCornerRightY = i2r(ffp.RightEye.LidCornerRight.Y, y, h);
row.RightLidUpX = i2r(ffp.RightEye.LidUpCenter.X, x, w);
row.RightLidUpY = i2r(ffp.RightEye.LidUpCenter.Y, y, h);
row.MouthCenterX = i2r(ffp.Mouth.MouthCenter.Y, y, h);
row.MouthCenterY = i2r(ffp.Mouth.MouthCenter.Y, y, h);
row.Thumbnail = ms.ToArray();
dataSet.Training.AddTrainingRow(row);
}
/// <summary>
/// The main entry point for this window.
/// </summary>
public MainWindow()
{
InitializeComponent();
Console.WriteLine("Loc: " + System.Reflection.Assembly.GetExecutingAssembly().Location);
try
{
_CurrentImagePathIndex = 0;
_ImagePathArray = new ArrayList();
_ProcessOptions = new ProcessOptions();
_FFP = new FaceFeaturePoints();
_ProcessOptions.DoEyeProcessing = 1;
_ProcessOptions.DoMouthProcessing = 1;
_ProcessOptions.DrawAnthropometricPoints = 0;
_ProcessOptions.DrawSearchRectangles = 0;
_ProcessOptions.DrawFaceRectangle = 1;
_ProcessOptions.DrawDetectionTime = 1;
_ProcessOptions.DrawFeaturePoints = 1;
_ProcessOptions.DoVisualDebug = 0;
#region Intialize encapsulated OpenCV subsystem
_KLU = new Klu();
_TempBitmap = new System.Drawing.Bitmap(10, 10);
// Create a Timer with a Normal Priority
_CaptureTimer = new DispatcherTimer(DispatcherPriority.ApplicationIdle, this.Dispatcher);
// Set the callback to just show the time ticking away
// NOTE: We are using a control so this has to run on
// the UI thread
_CaptureTimer.Tick += new EventHandler(
delegate(object s, EventArgs a)
{
_KLU.ProcessCaptureImage(ref _ProcessOptions, ref _FFP);
// Ensure the image (bitmap) we are writing to has the correct dimensions
int width = 0, height = 0;
_KLU.GetLastProcessedImageDims(ref width, ref height);
if (_TempBitmap.Width != width || _TempBitmap.Height != height)
{
Console.WriteLine("Need to resize the _TempBitmap to " + width + "x" + height);
_TempBitmap.Dispose();
GC.Collect();
_TempBitmap = new System.Drawing.Bitmap(width, height);
}
_KLU.GetLastProcessedImage(ref _TempBitmap);
_KLU.SetWpfImageFromBitmap(ref image1, ref _TempBitmap);
//_KLU.SetImageBrushFromBitmap(ref imageBrush, ref _TempBitmap);
}
);
#endregion
#region "Connect" to database
_TAM = new TableAdapterManager();
_DataSet = new TrainingDataSet();
// Load data from SQL database and fill our DataSet
_TAM.ExpressionTableAdapter = new ExpressionTableAdapter();
_TAM.TrainingTableAdapter = new TrainingTableAdapter();
LoadData();
#endregion
}
catch (Exception ex)
{
MessageBox.Show(ex.Message, "Error!", MessageBoxButton.OK, MessageBoxImage.Error);
}
}
public Task <InvokeResult> AddTrainingDataSetAsync([FromBody] TrainingDataSet set)
{
return(this._mgr.AddTrainingDataSetManager(set, OrgEntityHeader, UserEntityHeader));
}
public Task <InvokeResult> UpdateTrainingDataSetAsync([FromBody] TrainingDataSet set)
{
this.SetUpdatedProperties(set);
return(this._mgr.UpdateTrainingDataSetManager(set, OrgEntityHeader, UserEntityHeader));
}
public abstract double PerformSingleEpochProducingErrorRate(TrainingDataSet trainingDataSet);
public void Start()
{
var dataSets = new List <TrainingDataSet>();
trainingLogger.Message($"{BELT}{DateTime.Now.TimeOfDay}{BELT}");
trainingLogger.Message("Opening connection to database");
using (var db = new GlobalTerrorismContext())
{
var sampleRecord = db.AttackRecords.First(); // assert database connection
trainingLogger.Message("Preparing");
var count = db.AttackRecords.Count();
var nulls = 0;
initializers = Columns.Select(x => new InputInitializer(x.PropertyInfo)).ToList();
trainingLogger.Message("Downloading data");
var attackRecords = db.AttackRecords.ToList();
trainingLogger.Message("Normalizing data");
for (int i = 0; i < count; i++)
{
foreach (var initializer in initializers)
{
initializer.UpdateLimits(attackRecords[i]);
}
//log.Progress($"Normalizing data: {(float)i / count * 100}%");
}
trainingLogger.Message("Creating data sets");
//log.Finish();
for (int i = 0; i < count; i++)
{
var inputs = initializers.Select(x => x.TryGetValue(attackRecords[i]));
if (inputs.Any(x => x == null))
{
nulls++;
continue;
}
var dataSet = new TrainingDataSet
{
Inputs = inputs.Select(x => x.Value).ToArray(),
Outputs = CreateOutput(attackRecords[i].Fatalities)
};
//log.Progress($"Creating data sets: {(float)i / count * 100}%");
dataSets.Add(dataSet);
}
//log.Finish();
trainingLogger.Message($"{nulls} records skipped as they contains nulls.");
}
trainingLogger.Message("Training started");
network.Train(
trainingDataSets: dataSets,
learningRate: learningRate,
learningDataPercentage: learningDataRatio
);
trainingLogger.Message("Finished");
trainingLogger.Message($"Training accuracy: {network.Accuracy}");
TrainingCompleted?.Invoke(this, ProcessResult.Success);
}
public override double PerformSingleEpochProducingErrorRate(TrainingDataSet trainingDataSet)
{
neuralNetwork.PredictionFor(trainingDataSet.Inputs, parallelOptions);
return(BackPropagate(trainingDataSet.Outputs));
}
public static IEnumerable <TrainingDataSet> BuildTrainingDataForAllInputs(
IEnumerable <double[]> inputs,
IEnumerable <double[]> outputs)
{
return(inputs.Zip(outputs, (input, output) => TrainingDataSet.For(input, output)));
}
public static void AddTrainingData(ref TrainingDataSet dataSet, ref FaceFeaturePoints ffp, int expressionOID, ref System.Drawing.Bitmap picture)
{
// Convert facial coordinates to relative face rectangle coordinates.
// We do this because we are not interested in absolute coordinates but in quality of the face.
int x = ffp.FaceRectangle.X;
int y = ffp.FaceRectangle.Y;
int w = ffp.FaceRectangle.Width;
int h = ffp.FaceRectangle.Height;
// Calculate the eye distance (in relative coordinates
float dx = i2r(ffp.RightEye.EyeCenter.X, x, w) - i2r(ffp.LeftEye.EyeCenter.X, x, w);
float dy = i2r(ffp.RightEye.EyeCenter.Y, y, h) - i2r(ffp.LeftEye.EyeCenter.Y, y, h);
//float dx = i2r(ffp.RightEye.EyeCenter.X - ffp.LeftEye.EyeCenter.X, x, w);
//float dy = i2r(ffp.RightEye.EyeCenter.Y - ffp.LeftEye.EyeCenter.Y, y, h);
float eyeDist = (float)Math.Sqrt(dx * dx + dy * dy);
// Construct thumbnail
const int thumbnailWidth = 50;
const int thumbnailHeight = 50;
System.Drawing.Rectangle faceRect = new System.Drawing.Rectangle(x, y, w, h);
System.Drawing.Bitmap faceImg = picture.Clone(faceRect, picture.PixelFormat);
System.Drawing.Image.GetThumbnailImageAbort tc = new System.Drawing.Image.GetThumbnailImageAbort(ThumbnailCallback);
System.Drawing.Image thumbnail = faceImg.GetThumbnailImage(thumbnailWidth, thumbnailHeight, tc, IntPtr.Zero);
System.IO.MemoryStream ms = new System.IO.MemoryStream();
thumbnail.Save(ms, System.Drawing.Imaging.ImageFormat.Png);
TrainingDataSet.TrainingRow row = dataSet.Training.NewTrainingRow();
row.ExpressionOID = expressionOID;
row.LipCornerLeftX = i2r(ffp.Mouth.LipCornerLeft.X, x, w);
row.LipCornerLeftY = i2r(ffp.Mouth.LipCornerLeft.Y, y, h);
row.LipCornerRightX = i2r(ffp.Mouth.LipCornerRight.X, x, w);
row.LipCornerRightY = i2r(ffp.Mouth.LipCornerRight.Y, y, h);
row.LipUpLeftX = i2r(ffp.Mouth.LipUpLeft.X, x, w);
row.LipUpLeftY = i2r(ffp.Mouth.LipUpLeft.Y, y, h);
row.LipUpCenterX = i2r(ffp.Mouth.LipUpCenter.X, x, w);
row.LipUpCenterY = i2r(ffp.Mouth.LipUpCenter.Y, y, h);
row.LipUpRightX = i2r(ffp.Mouth.LipUpRight.X, x, w);
row.LipUpRightY = i2r(ffp.Mouth.LipUpRight.Y, y, h);
row.LipUpRightX = i2r(ffp.Mouth.LipBottomLeft.X, x, w);
row.LipUpRightY = i2r(ffp.Mouth.LipBottomLeft.Y, y, h);
row.LipBottomCenterX = i2r(ffp.Mouth.LipBottomCenter.X, x, w);
row.LipBottomCenterY = i2r(ffp.Mouth.LipBottomCenter.Y, y, h);
row.LipBottomRightX = i2r(ffp.Mouth.LipBottomRight.X, x, w);
row.LipBottomRightY = i2r(ffp.Mouth.LipBottomRight.Y, y, h);
row.LipBottomLeftX = i2r(ffp.Mouth.LipBottomLeft.X, x, w);
row.LipBottomLeftY = i2r(ffp.Mouth.LipBottomLeft.Y, y, h);
row.EyeDistance = eyeDist;
row.LeftEyeCenterX = i2r(ffp.LeftEye.EyeCenter.X, x, w);
row.LeftEyeCenterY = i2r(ffp.LeftEye.EyeCenter.Y, y, h);
row.LeftLidBottomX = i2r(ffp.LeftEye.LidBottomCenter.X, x, w);
row.LeftLidBottomY = i2r(ffp.LeftEye.LidBottomCenter.Y, y, h);
row.LeftLidCornerLeftX = i2r(ffp.LeftEye.LidCornerLeft.X, x, w);
row.LeftLidCornerLeftY = i2r(ffp.LeftEye.LidCornerLeft.Y, y, h);
row.LeftLidCornerRightX = i2r(ffp.LeftEye.LidCornerRight.X, x, w);
row.LeftLidCornerRightY = i2r(ffp.LeftEye.LidCornerRight.Y, y, h);
row.LeftLidUpX = i2r(ffp.LeftEye.LidUpCenter.X, x, w);
row.LeftLidUpY = i2r(ffp.LeftEye.LidUpCenter.Y, y, h);
row.RightEyeCenterX = i2r(ffp.RightEye.EyeCenter.X, x, w);
row.RightEyeCenterY = i2r(ffp.RightEye.EyeCenter.Y, y, h);
row.RightLidBottomX = i2r(ffp.RightEye.LidBottomCenter.X, x, w);
row.RightLidBottomY = i2r(ffp.RightEye.LidBottomCenter.Y, y, h);
row.RightLidCornerLeftX = i2r(ffp.RightEye.LidCornerLeft.X, x, w);
row.RightLidCornerLeftY = i2r(ffp.RightEye.LidCornerLeft.Y, y, h);
row.RightLidCornerRightX = i2r(ffp.RightEye.LidCornerRight.X, x, w);
row.RightLidCornerRightY = i2r(ffp.RightEye.LidCornerRight.Y, y, h);
row.RightLidUpX = i2r(ffp.RightEye.LidUpCenter.X, x, w);
row.RightLidUpY = i2r(ffp.RightEye.LidUpCenter.Y, y, h);
row.MouthCenterX = i2r(ffp.Mouth.MouthCenter.Y, y, h);
row.MouthCenterY = i2r(ffp.Mouth.MouthCenter.Y, y, h);
row.Thumbnail = ms.ToArray();
dataSet.Training.AddTrainingRow(row);
}
/// <summary>
/// The main entry point for this window.
/// </summary>
public MainWindow()
{
InitializeComponent();
Console.WriteLine("Loc: " + System.Reflection.Assembly.GetExecutingAssembly().Location);
try
{
_CurrentImagePathIndex = 0;
_ImagePathArray = new ArrayList();
_ProcessOptions = new ProcessOptions();
_FFP = new FaceFeaturePoints();
_ProcessOptions.DoEyeProcessing = 1;
_ProcessOptions.DoMouthProcessing = 1;
_ProcessOptions.DrawAnthropometricPoints = 0;
_ProcessOptions.DrawSearchRectangles = 0;
_ProcessOptions.DrawFaceRectangle = 1;
_ProcessOptions.DrawDetectionTime = 1;
_ProcessOptions.DrawFeaturePoints = 1;
_ProcessOptions.DoVisualDebug = 0;
#region Intialize encapsulated OpenCV subsystem
_KLU = new Klu();
_TempBitmap = new System.Drawing.Bitmap(10, 10);
// Create a Timer with a Normal Priority
_CaptureTimer = new DispatcherTimer(DispatcherPriority.ApplicationIdle, this.Dispatcher);
// Set the callback to just show the time ticking away
// NOTE: We are using a control so this has to run on
// the UI thread
_CaptureTimer.Tick += new EventHandler(
delegate(object s, EventArgs a)
{
_KLU.ProcessCaptureImage(ref _ProcessOptions, ref _FFP);
// Ensure the image (bitmap) we are writing to has the correct dimensions
int width = 0, height = 0;
_KLU.GetLastProcessedImageDims(ref width, ref height);
if (_TempBitmap.Width != width || _TempBitmap.Height != height)
{
Console.WriteLine("Need to resize the _TempBitmap to " + width + "x" + height);
_TempBitmap.Dispose();
GC.Collect();
_TempBitmap = new System.Drawing.Bitmap(width, height);
}
_KLU.GetLastProcessedImage(ref _TempBitmap);
_KLU.SetWpfImageFromBitmap(ref image1, ref _TempBitmap);
//_KLU.SetImageBrushFromBitmap(ref imageBrush, ref _TempBitmap);
}
);
#endregion
#region "Connect" to database
_TAM = new TableAdapterManager();
_DataSet = new TrainingDataSet();
// Load data from SQL database and fill our DataSet
_TAM.ExpressionTableAdapter = new ExpressionTableAdapter();
_TAM.TrainingTableAdapter = new TrainingTableAdapter();
LoadData();
#endregion
}
catch (Exception ex)
{
MessageBox.Show(ex.Message, "Error!", MessageBoxButton.OK, MessageBoxImage.Error);
}
}
