/*Check that the returned gradient is equal to the numerical gradient*/
public void GradientMatchNumerical()
{
//
var v = Vector <double> .Build.Dense(3);
//
ScanMatching.rEBb = Matrix <double> .Build.Random(2, 50);
ScanMatching.rPNn = Matrix <double> .Build.Random(2, 15);
var hp = new HelperFunctions();
Func <double[], double> f = x => hp.ExtractCost(x);
var derivEst = new NumericalJacobian(5, 2);
var gradExpect = derivEst.Evaluate(f, v.ToArray());
(var ftmp, var gradActual, var Htmp) = ScanMatching.Cost(v);
// Not sure how to do assert array equal with precision.
for (int i = 0; i < gradExpect.Length; i++)
{
Assert.Equal(gradExpect[i], gradActual.ToArray()[i], 5);
}
}
/*The following scenarios are contrived such that the point cloud locations are mirrored and should produce the same cost*/
public void EqualCostScenarios()
{
// Initialise point clouds
double[] x1 = { 0, 0, 0, 1 };
double[] x2 = { 0, -0.5, 0, 0.5 };
Matrix <double> PointCloud1 = Matrix <double> .Build.DenseOfColumnMajor(2, 2, x1);
Matrix <double> PointCloud2 = Matrix <double> .Build.DenseOfColumnMajor(2, 2, x2);
ScanMatching.rEBb = PointCloud2;
ScanMatching.rPNn = PointCloud1;
// Generate two pose vectors (x,y,psi)
double[] vv1 = { -0.25, 0.5, -0.1 };
Vector <double> v1 = Vector <double> .Build.Dense(vv1);
double[] vv2 = { 0.25, 0.5, 0.1 };
Vector <double> v2 = Vector <double> .Build.DenseOfArray(vv2);
(double expected, var G1, var H1) = ScanMatching.Cost(v2);
(double actual, var G2, var H2) = ScanMatching.Cost(v1);
/*Precision is given as an integer defining the number of decimal places,
* Normal practice would be to select precision based on machine precision
* E.g., sqrt(eps) where eps is machine precision, sqrt(eps) = 8.5e-8 on my machine.
* Therefore, 8 decimal places will suffice
*/
Assert.Equal(expected, actual, 8);
}
public void CloserPointsImproveCost()
{
// Initialise point clouds
double[] x1 = { 0, 1 };
double[] x2 = { 0, 2 };
Matrix <double> PointCloud1 = Matrix <double> .Build.DenseOfColumnMajor(2, 1, x1);
Matrix <double> PointCloud2 = Matrix <double> .Build.DenseOfColumnMajor(2, 1, x2);
ScanMatching.rEBb = PointCloud1;
ScanMatching.rPNn = PointCloud2;
// Generate two pose vectors (x,y,psi)
Vector <double> v1 = Vector <double> .Build.Dense(3);//This should yield a higher cost than
double[] vv = { 0, 1, 0 };
Vector <double> v2 = Vector <double> .Build.DenseOfArray(vv);//this one
(double CostExpectLow, var G1, var H1) = ScanMatching.Cost(v2);
(double CostExpectHigh, var G2, var H2) = ScanMatching.Cost(v1);
// Since the points lie on top of one another when pose = {0 1 0} that should yield a lower cost
Assert.True(CostExpectLow < CostExpectHigh, "Expected closer points to yield a lower cost, something is fundamentally wrong with the cost function");
}
public double ExtractCost(double[] v)
{
var vv = Vector <double> .Build.DenseOfArray(v);
(var f, _, _) = ScanMatching.Cost(vv);
return(f);
}
public MainWindow() : base(Gtk.WindowType.Toplevel)
{
Build();
var plotView = new PlotView();
this.Add(plotView);
plotView.ShowAll();
// Choose file
Gtk.FileChooserDialog fcd = new Gtk.FileChooserDialog("Open File", null, Gtk.FileChooserAction.Open);
fcd.AddButton(Gtk.Stock.Cancel, Gtk.ResponseType.Cancel);
fcd.AddButton(Gtk.Stock.Open, Gtk.ResponseType.Ok);
fcd.DefaultResponse = Gtk.ResponseType.Ok;
fcd.SelectMultiple = false;
Gtk.ResponseType response = (Gtk.ResponseType)fcd.Run();
String DataFile = fcd.Filename;
if (response == ResponseType.Ok)
{
// Run scan Matching
ScanMatching.RunScanMatch(DataFile);
Console.WriteLine($"{DataFile} selected");
}
else
{
Console.WriteLine("no file selected");
}
fcd.Destroy();
// Run Plot
var model = new PlotModel {
Title = "Scan Matching"
};
var scatterSeries = new ScatterSeries {
MarkerType = MarkerType.Circle
};
var r = new Random(314);
for (int i = 0; i < D_Scan_Matching_GUI.ScanMatching.rPNn.ColumnCount; i++)
{
var x = D_Scan_Matching_GUI.ScanMatching.rPNn[0, i];
var y = D_Scan_Matching_GUI.ScanMatching.rPNn[1, i];
var size = 5;
var colorValue = 1500;
scatterSeries.Points.Add(new ScatterPoint(x, y, size, colorValue));
}
model.Series.Add(scatterSeries);
model.Axes.Add(new LinearColorAxis {
Position = AxisPosition.Right, Palette = OxyPalettes.Jet(200)
});
plotView.Model = model;
}
public Tuple <double, Vector <double> > ExtractCostAndGrad(Vector <double> v)
{
(var f, var g, _) = ScanMatching.Cost(v);
return(Tuple.Create(f, g));
}