public void Test_Poly4()
{
double[] param = new double[2];
param[0] = 1;
param[1] = 1E10;
double[] ys = new double[10];
int info = 0;
do
{
NLFit.LevenbergMarquardtFit(new NLFit.LMFunction(this.poly4), param, ys, 1E-10, ref info);
} while (info == 5);
Assert.AreEqual(1.08222543840634, param[0], 1E-4, "Fit parameter 0 should be 1.08222543840634 in this model");
Assert.AreEqual(86441011642265, param[1], 1E4, "Fit parameter 1 should be 86441011642265 in this model");
double[] covar = new double[2 * 2];
double chisqr;
NLFit.ComputeCovariances(new NLFit.LMFunction(this.poly4), param, 10, 2, covar, out chisqr);
// Expected error for parameter1 is: 0.115944838291204
// Expected error for parameter2 is: 14249106469708
Assert.AreEqual(0.115944838291204, Math.Sqrt(covar[0]), 1E-5, "Square root of covariance of parameter0 should be 0.115944838291204");
Assert.AreEqual(14249106469708, Math.Sqrt(covar[3]), 142491064, "Square root of covariance of parameter1 should be 14249106469708");
// Origin calculates a dependence value of 0.92115 out here
double dependency = (covar[1] * covar[1]) / (covar[0] * covar[3]);
Assert.AreEqual(0.92115, dependency, 1E-4, "Dependency should be around 0.92115");
}
public void Test_Poly2()
{
double[] param = new double[2];
param[0] = 0.1;
param[1] = 1.01;
double[] ys = new double[9];
int info = 0;
do
{
NLFit.LevenbergMarquardtFit(new NLFit.LMFunction(this.poly2), param, ys, 1E-10, ref info);
} while (info == 5);
Assert.AreEqual(0.1111111, param[0], 1E-4, "Fit parameter 0 should be 0.11111 in this model");
Assert.AreEqual(1, param[1], 1E-4, "Fit parameter 1 should be 1 in this model");
double[] covar = new double[2 * 2];
double chisqr;
NLFit.ComputeCovariances(new NLFit.LMFunction(this.poly2), param, 9, 2, covar, out chisqr);
Assert.AreEqual(0.670194, covar[0], 0.001);
Assert.AreEqual(-0.10582, covar[1], 0.001);
Assert.AreEqual(-0.10582, covar[2], 0.001);
Assert.AreEqual(0.021164, covar[3], 0.001);
}
public void ZeroElements()
{
arr[0] = 1;
double r = NLFit.enorm(0, arr);
NUnit.Framework.Assert.AreEqual(0.0, r, 0);
}
public void FitFunction7malCos3xplus1(int numberOfYs, int numberOfParameter, double[] parameter, double[] ys, ref int info)
{
Assert.IsTrue(numberOfParameter == 3, "NumberOfParameter must be 3 in this test.");
Assert.IsTrue(numberOfYs == 3, "NumberOfYs must be 3 in this test.");
Assert.IsTrue(parameter.Length == 3, "Length of parameter array must be 3 in this test.");
Assert.IsTrue(ys.Length == 3, "Length of ys array must be 3 in this test.");
// we use the simplest model y=a+bx in the range -10 ..10 with step size of 1 to
// calculate the delta
double a = parameter[0];
double b = parameter[1];
double c = parameter[2];
double sum_se = 0;
double x;
for (x = -10; x <= 10; x += 1)
{
sum_se += NLFit.sqr((a * Math.Cos(b * x + c)) - (7 * Math.Cos(3 * x + 1)));
}
ys[0] = sum_se;
ys[1] = 0;
ys[2] = 0;
}
public void Fit1()
{
int info = 0;
double[] differences = new double[NumberOfData];
NLFit.LevenbergMarquardtFit(new NLFit.LMFunction(EvaluateFitDifferences), _cachedVaryingParameters, differences, 1E-10, ref info);
_resultingCovariances = new double[_cachedVaryingParameters.Length * _cachedVaryingParameters.Length];
NLFit.ComputeCovariances(new NLFit.LMFunction(EvaluateFitDifferences), _cachedVaryingParameters, NumberOfData, _cachedVaryingParameters.Length, _resultingCovariances, out _resultingSumChiSquare, out _resultingSigmaSquare);
}
public void NormalAndSmallElements()
{
// we use the law 3^2 + 4^2 = 5^2
arr[0] = 400E-20;;
for (int i = 1; i < 90001; i++)
{
arr[i] = 1E-20;
}
double r = NLFit.enorm(90001, arr);
NUnit.Framework.Assert.AreEqual(500E-20, r, 0);
}
public void Test_2plus5xMod()
{
double[] param = new double[2];
param[0] = 1;
param[1] = 1;
double[] ys = new double[1];
int info = 0;
NLFit.LevenbergMarquardtFit(new NLFit.LMFunction(this.FitFunction2plus5xMod), param, ys, 1E-10, ref info);
Assert.AreEqual(0, info, "Info should be 0 due to inappropriate length of ys in this model");
}
public void OneElement()
{
for (int i = 0; i < arr.Length; i++)
{
arr[i] = (i + 1);
}
for (int i = 0; i < arr.Length; i++)
{
double r = NLFit.enorm(1, arr, i);
NUnit.Framework.Assert.AreEqual((double)(i + 1), r, 0);
}
}
public void Test_2plus5x()
{
double[] param = new double[2];
param[0] = 1;
param[1] = 1;
double[] ys = new double[2];
int info = 0;
NLFit.LevenbergMarquardtFit(new NLFit.LMFunction(this.FitFunction2plus5x), param, ys, 1E-10, ref info);
Assert.AreEqual(2, param[0], 1E-5, "Fit parameter 0 should be 2 in this model");
Assert.AreEqual(5, param[1], 1E-5, "Fit parameter 1 should be 5 in this model");
}
public void Test_Poly5()
{
double[] param = new double[3];
param[0] = 1;
param[1] = 1;
param[2] = 1;
double[] ys = new double[10];
int info = 0;
do
{
NLFit.LevenbergMarquardtFit(new NLFit.LMFunction(this.poly5), param, ys, 1E-10, ref info);
} while (info == 5);
Assert.AreEqual(1.13158682895444, param[0], 1E-4, "Fit parameter 0 should be 1.13158682895444 in this model");
Assert.AreEqual(0.00823465185064109, param[1], 1E-6, "Fit parameter 1 should be 0.00823465185064109 in this model");
Assert.AreEqual(-1.14393644797462, param[2], 1E-4, "Fit parameter 1 should be -1.14393644797462 in this model");
double[] covar = new double[3 * 3];
double chisqr;
NLFit.ComputeCovariances(new NLFit.LMFunction(this.poly5), param, 10, 3, covar, out chisqr);
// Expected error for parameter1 is: 0.115944838291204
// Expected error for parameter2 is: 14249106469708
Assert.AreEqual(0.19511, Math.Sqrt(covar[0]), 1e-4, "Square root of covariance of parameter0 should be 0.19511");
Assert.AreEqual(0.00197, Math.Sqrt(covar[3 * 1 + 1]), 1e-5, "Square root of covariance of parameter1 should be 0.00197");
Assert.AreEqual(3.50947, Math.Sqrt(covar[3 * 2 + 2]), 1e-4, "Square root of covariance of parameter1 should be 3.50947");
double dep;
double dep01 = (covar[3 * 0 + 1] * covar[3 * 1 + 0]) / (covar[3 * 0 + 0] * covar[3 * 1 + 1]);
double dep02 = (covar[3 * 0 + 2] * covar[3 * 2 + 0]) / (covar[3 * 0 + 0] * covar[3 * 2 + 2]);
dep = 1 - (1 - dep01) * (1 - dep02);
Assert.AreEqual(0.96865, dep, 1E-2, "Dependency should be around 0.97");
double dep10 = (covar[3 * 1 + 0] * covar[3 * 0 + 1]) / (covar[3 * 1 + 1] * covar[3 * 0 + 0]);
double dep12 = (covar[3 * 1 + 2] * covar[3 * 2 + 1]) / (covar[3 * 1 + 1] * covar[3 * 2 + 2]);
dep = 1 - (1 - dep10) * (1 - dep12);
Assert.AreEqual(0.95335, dep, 1E-2, "Dependency should be around 0.95");
double dep20 = (covar[3 * 2 + 0] * covar[3 * 0 + 2]) / (covar[3 * 2 + 2] * covar[3 * 0 + 0]);
double dep21 = (covar[3 * 2 + 1] * covar[3 * 1 + 2]) / (covar[3 * 2 + 2] * covar[3 * 1 + 1]);
dep = 1 - (1 - dep20) * (1 - dep21);
Assert.AreEqual(0.755, dep, 1E-2, "Dependency should be around 0.755");
}
public void Test_7malCos3xplus1()
{
double[] param = new double[3];
param[0] = 1;
param[1] = 3.1;
param[2] = 1.1;
double[] ys = new double[3];
int info = 0;
NLFit.LevenbergMarquardtFit(new NLFit.LMFunction(this.FitFunction7malCos3xplus1), param, ys, 1E-10, ref info);
Assert.AreEqual(7, param[0], 1E-4, "Fit parameter 0 should be 7 in this model");
Assert.AreEqual(3, param[1], 1E-4, "Fit parameter 1 should be 3 in this model");
Assert.AreEqual(1, param[2], 1E-4, "Fit parameter 2 should be 1 in this model");
}
public void FitFunction7malCos3xplus1Mod(int numberOfYs, int numberOfParameter, double[] parameter, double[] ys, ref int info)
{
Assert.IsTrue(numberOfParameter == 3, "NumberOfParameter must be 3 in this test.");
Assert.IsTrue(numberOfYs == 21, "NumberOfYs must be 21 in this test.");
Assert.IsTrue(parameter.Length == 3, "Length of parameter array must be 3 in this test.");
Assert.IsTrue(ys.Length == 21, "Length of ys array must be 21 in this test.");
// we use the simplest model y=a+bx in the range -10 ..10 with step size of 1 to
// calculate the delta
double a = parameter[0];
double b = parameter[1];
double c = parameter[2];
for (int i = -10; i <= 10; i++)
{
ys[i + 10] = NLFit.sqr((a * Math.Cos(b * i + c)) - (7 * Math.Cos(3 * i + 1)));
}
}
public void Test_Poly3()
{
double[] param = new double[2];
param[0] = 1;
param[1] = 1;
double[] ys = new double[2];
int info = 0;
do
{
NLFit.LevenbergMarquardtFit(new NLFit.LMFunction(poly3), param, ys, 1E-10, ref info);
} while (info == 5);
Assert.AreEqual(0.1111111, param[0], 1E-4, "Fit parameter 0 should be 0.11111 in this model");
Assert.AreEqual(1, param[1], 1E-4, "Fit parameter 1 should be 1 in this model");
double[] covar = new double[2 * 2];
NLFit.ComputeCovariances(new NLFit.LMFunction(poly3), param, 2, 2, covar, out var chisqr, out var sigmaSquare);
}
public void FitFunction2plus5xMod(int numberOfYs, int numberOfParameter, double[] parameter, double[] ys, ref int info)
{
Assert.IsTrue(numberOfParameter == 2, "NumberOfParameter must be 2 in this test.");
Assert.IsTrue(numberOfYs == 1, "NumberOfYs must be 1 in this test.");
Assert.IsTrue(parameter.Length == 2, "Length of parameter array must be 2 in this test.");
Assert.IsTrue(ys.Length == 1, "Length of ys array must be 1 in this test.");
// we use the simplest model y=a+bx in the range -10 ..10 with step size of 1 to
// calculate the delta
double a = parameter[0];
double b = parameter[1];
double sum_se = 0;
double x;
for (x = -10; x <= 10; x += 1)
{
sum_se += NLFit.sqr((a + b * x) - (2 + 5 * x));
}
ys[0] = sum_se;
}
/// <summary>
/// Can only be used, if all fit functions provide the jacobian.
/// </summary>
public void Fit2Jac()
{
//int info = 0;
object workingmemory = null;
NonLinearFit2.LEVMAR_DER(
new NonLinearFit2.FitFunction(EvalulateFitValues),
new NonLinearFit2.JacobianFunction(EvaluateFitJacobian),
_cachedVaryingParameters,
_cachedDependentValues,
_cachedWeights,
100, // itmax
null, // opts,
null, // info,
ref workingmemory,
null, // covar,
null // arbitrary data
);
_resultingCovariances = new double[_cachedVaryingParameters.Length * _cachedVaryingParameters.Length];
NLFit.ComputeCovariances(new NLFit.LMFunction(EvaluateFitDifferences), _cachedVaryingParameters, NumberOfData, _cachedVaryingParameters.Length, _resultingCovariances, out _resultingSumChiSquare, out _resultingSigmaSquare);
}
static void Main(string[] args)
{
double[] arr = { 1E20, 1, 1E-20 };
double result = NLFit.enorm(3, arr);
}