/// <summary>
/// <para>Unscented transform parameters optimization procedure.</para>
/// <para>The OptimizationMethod param determines the optimization method:</para>
/// <para>- OptimizationMethod.RandomShoot - parameters are randomly sampled and the best sample is chosen as optimal;
/// <para>- OptimizationMethod.NelderMeed - parameters are optimized with non-gradient Nelder-Meed method.</para>
/// <para>The UTOptimizationType type param determines the relation between the optimized variable and the unscented tranform params (see UTParams and its constructors for details). </para>
/// <para>- If type is UTOptimizationType.ImplicitAlpha, then the optimized variable is saclar [alpha0];</para>
/// <para>- If type is UTOptimizationType.ImplicitAlphaBetaKappa, then optimized variable is a vector [alpha, beta, kappa];</para>
/// <para>- If type is UTOptimizationType.Explicit, then then optimized variable is a vector [lambda, wm0, wc0, wi]. ///TODO it is not correct to define the parameters of the unsctnted transform arbitraty, they have to be interdependent, so that the mean and cov would be transformed correctly.</para>
/// </summary>
/// <param name="method">Unscented transform parameters optimization method</param>
/// <param name="type">Unscented transform parameters definition type</param>
/// <param name="Phi1">State transformation: a nonlinear function which determines the dynamics: x_{t+1} = Phi_1(x_t) + Phi_2(x_t) W_t</param>
/// <param name="Phi2">Noise multiplicator in the dynamics equation: x_{t+1} = Phi(x_t) + W_t</param>
/// <param name="Psi1">Observations transformation: a nonlinear function which determines the relation between the state and the observations: y_t = Psi_1(x_t) + Psi_2(x_t) Nu_t</param>
/// <param name="Psi2">Noise multiplicator in the observations equation: y_t = Psi_1(x_t) + Psi_2(x_t) Nu_t</param>
/// <param name="Mw">Mean of the noise in the dynamics equation </param>
/// <param name="Rw">Covariance matrix of the state disturbances</param>
/// <param name="Mnu">Mean of the noise in the obseration equation </param>
/// <param name="Rnu">Convariance matrix of the observation noise</param>
/// <param name="Crit">Criterion: a function which determines the quality of the unscented Kalman filter. Depends on the sample covariance of the estimation error on the last step: val = Crit(Cov(X_T-Xhat_T,X_T-Xhat_T)) </param>
/// <param name="T">The upper bound of the observation interval</param>
/// <param name="models">Discrete vector model samples</param>
/// <param name="xhat0">Initial condition</param>
/// <param name="DX0Hat">Initial condition covariance</param>
/// <param name="outputFolder">The results are saved to this folder in file "UT_optimization_{type}.txt"</param>
static (double, UTParams, UTParams) UTParmsOptimize(OptimizationMethod method, UTDefinitionType type,
Func <int, Vector <double>, Vector <double> > Phi1,
Func <int, Vector <double>, Matrix <double> > Phi2,
Func <int, Vector <double>, Vector <double> > Psi1,
Func <int, Vector <double>, Matrix <double> > Psi2,
Vector <double> Mw,
Matrix <double> Rw,
Vector <double> Mnu,
Matrix <double> Rnu,
Func <Matrix <double>, double> Crit,
int T,
DiscreteVectorModel[] models,
Vector <double> xhat0,
Matrix <double> DX0Hat,
string outputFolder)
{
(int n, Vector <double> lowerBound, Vector <double> upperBound, Vector <double> initialGuess, string filename) = DefineOptimizationParameters(type, xhat0, string.IsNullOrWhiteSpace(outputFolder) ? null : Path.Combine(outputFolder, "UT_optimization_{type}.txt"));
double min = double.MaxValue;
Vector <double> argmin = Exts.Stack(initialGuess, initialGuess);
switch (method)
{
case OptimizationMethod.RandomShoot:
var OptimumRandom = RandomOptimizer.Minimize((x) => CalculateSampleCriterion(Phi1, Phi2, Psi1, Psi2, Mw, Rw, Mnu, Rnu, Crit, x, T, models, xhat0, DX0Hat), Exts.Stack(lowerBound, lowerBound), Exts.Stack(upperBound, upperBound), 100, 100, filename);
min = OptimumRandom.min;
argmin = OptimumRandom.argmin;
break;
case OptimizationMethod.NelderMeed:
NelderMeadSimplex optimizer = new NelderMeadSimplex(1e-3, 100);
var objective = ObjectiveFunction.Value((x) => CalculateSampleCriterion(Phi1, Phi2, Psi1, Psi2, Mw, Rw, Mnu, Rnu, Crit, x, T, models, xhat0, DX0Hat));
try
{
var optimumNM = optimizer.FindMinimum(objective, Exts.Stack(initialGuess, initialGuess));
min = optimumNM.FunctionInfoAtMinimum.Value;
argmin = optimumNM.MinimizingPoint;
}
catch (Exception e)
{
Console.WriteLine($"Optimizer faild, using the initail guess ({e.Message})");
argmin = Exts.Stack(initialGuess, initialGuess);
}
break;
default: // no optimization by default
break;
}
return(min, new UTParams(xhat0.Count, argmin.Take(n).ToArray()), new UTParams(xhat0.Count, argmin.Skip(n).Take(n).ToArray()));
}
static void Run(Options o, string[] args)
{
if (string.IsNullOrWhiteSpace(o.ScriptsFolder))
{
o.ScriptsFolder = "..\\..\\..\\OutputScripts\\";
}
Directory.CreateDirectory(o.OutputFolder);
using (System.IO.StreamWriter outputfile = new System.IO.StreamWriter(Path.Combine(o.OutputFolder, "parameters.txt"), true))
{
outputfile.WriteLine($"{DateTime.Now}\t{string.Join(" ", args)}");
outputfile.Close();
}
// original continuous model
// x_t = x_0 + \int_0^t Phi_1(x_t) dt + \int_0^t Phi_2(x_t) dW_t
// discrete model:
// x_{t+1} = Phi_1(x_t) + Phi_2(x_t) W_t
//observations
// y_t = Psi_1(x_t) + Psi_2(x_t) Nu_t
#region 3d model Ienkaran Arasaratnam, Simon Haykin, and Tom R. Hurd
//mpdel params
//double sigma1 = Math.Sqrt(0.2);
//double sigma2 = 7.0 * 1e-3;
//double sigma_r = 50;
//double sigma_th = 0.1;
//double sigma_ph = 0.1;
// starting point
//Vector<double> mEta = Exts.Vector(1000, 0, 2650, 150, 200, 0, 1.0);
//Matrix<double> dEta = Exts.Diag(1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0);
//RandomVector<Normal> NormalEta = new RandomVector<Normal>(mEta, dEta);
//Func<Vector<double>> X0;
////X0 = () => NormalEta.Sample();
//X0 = () => mEta;
// dynamics
//Func<Vector<double>, Vector<double>> Phi1 = (x) => Exts.Vector(x[1], -x[6] * x[3], x[3], x[6] * x[1], x[5], 0, 0);
//Func<Matrix<double>> Phi2 = () => Exts.Diag(1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0);
//Vector<double> mW = Exts.Vector(0, 0, 0, 0, 0, 0, 0);
//Matrix<double> dW = Exts.Diag(1e1, Math.Pow(sigma1, 2), 1e1, Math.Pow(sigma1, 2), 1e1, Math.Pow(sigma1, 2), Math.Pow(sigma2, 2));
//RandomVector<Normal> NormalW = new RandomVector<Normal>(mW, dW);
//Func<Vector<double>> W;
//W = () => NormalW.Sample();
// observations
//Func<Vector<double>, Vector<double>> Psi1 = (x) => Utils.cart2sphere(Exts.Vector(x[0], x[2], x[4]));
//Func<Matrix<double>> Psi2 = () => Exts.Diag(1.0, 1.0, 1.0);
//Vector<double> mNu = Exts.Vector(0, 0, 0);
//Matrix<double> dNu = Exts.Diag(Math.Pow(sigma_r, 2), Math.Pow(sigma_th, 2), Math.Pow(sigma_ph, 2));
////Vector<double> X_R2 = Exts.Vector(0, 0);
////Vector<double> mNu = Exts.Vector(0, 0, 0, 0);
////Matrix<double> dNu = Exts.Diag(Math.Pow(0.1 * Math.PI / 180, 2), Math.Pow(50, 2), Math.Pow(0.1 * Math.PI / 180, 2), Math.Pow(50, 2));
////Func<double, Vector<double>, Vector<double>> Psi1 = (s, x) => Exts.Stack(Utils.cart2pol(Exts.Vector(x[0], x[1]) - X_R1), Utils.cart2pol(Exts.Vector(x[0], x[1]) - X_R2));
////Func<double, Vector<double>, Matrix<double>> Psi2 = (s, x) => Exts.Diag(1.0, 1.0, 1.0, 1.0);
//RandomVector<Normal> NormalNu = new RandomVector<Normal>(mNu, dNu);
//Func<Vector<double>> Nu;
//Nu = () => NormalNu.Sample();
//double h_state = 0.01;
//double h_obs = 0.01;
//double T = 1.0 + h_state / 2;
#endregion
#region 2d model with acceleration
// model params
double Alpha_n = o.alpha; // 0.05;
double Beta_n = o.beta; //1.0;
double Gamma_n = o.gamma; //0.5;
// starting point
Vector <double> mEta = Exts.Vector(0, 25000, 400, 10 * Math.PI / 180, Gamma_n / Alpha_n);
Matrix <double> dEta = Exts.Diag(Math.Pow(o.DX0, 2), Math.Pow(o.DX0, 2), Math.Pow(115, 2), Math.Pow(15 * Math.PI / 180, 2), Math.Pow(Beta_n, 2) / 2 / Alpha_n);
RandomVector <Normal> NormalEta = new RandomVector <Normal>(mEta, dEta);
ContinuousUniform UniformEtaV = new ContinuousUniform(200, 600);
Func <Vector <double> > X0;
X0 = () =>
{
var x = NormalEta.Sample();
x[2] = UniformEtaV.Sample();
return(x);
};
// dynamics
Func <Vector <double>, Vector <double> > Phi1 = (x) => Exts.Vector(x[2] * Math.Cos(x[3]), x[2] * Math.Sin(x[3]), 0, x[4] / x[2], -Alpha_n * x[4] + Gamma_n);
Func <Matrix <double> > Phi2 = () => Exts.Diag(0, 0, 0, 0, 1.0);
Vector <double> mW = Exts.Vector(0, 0, 0, 0, 0);
Matrix <double> dW = Exts.Diag(0, 0, 0, 0, Math.Pow(Beta_n, 2));
Normal NormalW = new Normal(mW[4], Math.Sqrt(dW[4, 4]));
Func <Vector <double> > W;
W = () => Exts.Vector(0, 0, 0, 0, NormalW.Sample());
// observations
Vector <double> X_R1 = Exts.Vector(-10000, 10000); // first radar location
Vector <double> X_R2 = Exts.Vector(0, 0); // second radar location
Func <Vector <double>, Vector <double> > Psi1 = (x) => Exts.Stack(Utils.cart2pol(Exts.Vector(x[0], x[1]) - X_R1), Utils.cart2pol(Exts.Vector(x[0], x[1]) - X_R2));
Func <Matrix <double> > Psi2 = () => Exts.Diag(1.0, 1.0, 1.0, 1.0);
Vector <double> mNu = Exts.Vector(0, 0, 0, 0);
Matrix <double> dNu = Exts.Diag(Math.Pow(0.1 * Math.PI / 180, 2), Math.Pow(50, 2), Math.Pow(0.1 * Math.PI / 180, 2), Math.Pow(50, 2));
RandomVector <Normal> NormalNu = new RandomVector <Normal>(mNu, dNu);
Func <Vector <double> > Nu;
Nu = () => NormalNu.Sample();
//discretization
double h_state = o.h_state; //0.01;
double h_obs = o.h_obs; //1.0;
double T = o.T + h_state / 2;
Func <int, Vector <double>, Vector <double> > Phi1_discr = (i, x) =>
{
Vector <double> xx = x;
for (double s = h_state; s < h_obs + h_state / 2; s += h_state)
{
xx += h_state * Phi1(xx);
}
return(xx);
};
Func <int, Vector <double>, Matrix <double> > Phi2_discr = (i, x) => Math.Sqrt(h_obs) * Phi2();
Func <int, Vector <double>, Vector <double> > Psi1_discr = (i, x) => Psi1(x);
Func <int, Vector <double>, Matrix <double> > Psi2_discr = (i, x) => Psi2();
// derivatives for the Extended Kalman Filter
Func <Vector <double>, Matrix <double> > dPhi = (x) => Matrix <double> .Build.Dense(5, 5, new double[5 * 5] {
0, 0, 0, 0, 0,
0, 0, 0, 0, 0,
Math.Cos(x[3]), Math.Sin(x[3]), 0, -x[4] / Math.Pow(x[2], 2), 0,
-x[2] * Math.Sin(x[3]), x[2] * Math.Cos(x[3]), 0, 0, 0,
0, 0, 0, 1.0 / x[2], -Alpha_n
}); // Column-major order
Func <Vector <double>, Matrix <double> > dPsi = (x) =>
{
var x1 = x - X_R1.Stack(Exts.Vector(0, 0, 0));
var r1r1 = x1[0] * x1[0] + x1[1] * x1[1];
var r1 = Math.Sqrt(r1r1);
var x2 = x - X_R2.Stack(Exts.Vector(0, 0, 0));
var r2r2 = x2[0] * x2[0] + x2[1] * x2[1];
var r2 = Math.Sqrt(r2r2);
return(Matrix <double> .Build.Dense(4, 5, new double[5 * 4] {
-x1[1] / r1r1, x1[0] / r1, -x2[1] / r2r2, x2[0] / r2,
x1[0] / r1r1, x1[1] / r1, x2[0] / r2r2, x2[1] / r2,
0, 0, 0, 0,
0, 0, 0, 0,
0, 0, 0, 0
})); // Column-major order
};
Func <int, Vector <double>, Matrix <double>, (Vector <double>, Matrix <double>)> Ricatti = (i, x, P) =>
{
Vector <double> xx = x;
Matrix <double> PP = P;
for (double s = h_state; s < h_obs + h_state / 2; s += h_state)
{
PP += h_state * (dPhi(xx) * PP + PP * dPhi(xx).Transpose() + Phi2() * dW * Phi2().Transpose());
xx += h_state * (Phi1(xx) + Phi2() * mW);
}
return(xx, PP);
};
//trivial estimate
Func <int, Vector <double>, Vector <double>, Matrix <double>, (Vector <double>, Matrix <double>)> DummyEstimate = (i, y, x, P) =>
{
Vector <double> xx = x;
Matrix <double> PP = P;
for (double s = h_state; s < h_obs + h_state / 2; s += h_state)
{
PP += h_state * (dPhi(xx) * PP + PP * dPhi(xx).Transpose() + Phi2() * dW * Phi2().Transpose());
xx += h_state * (Phi1(xx) + Phi2() * mW);
}
return((0.5 * (Utils.pol2cart(Exts.Vector(y[0], y[1])) + X_R1 + Utils.pol2cart(Exts.Vector(y[2], y[3])) + X_R2)).Stack(Exts.Vector(xx[2], xx[3], xx[4])), PP);
};
#endregion
int N = (int)(T / h_obs);
Func <DiscreteVectorModel> ModelGenerator = () =>
{
DiscreteVectorModel model = null;
int n = 0;
double h_tolerance = h_state / 2.0;
double t_nextobservation = h_obs;
Vector <double> State = X0();
Vector <double> Obs;
model = new DiscreteVectorModel(Phi1_discr, Phi2_discr, Psi1_discr, Psi2_discr, (i) => W(), (i) => Nu(), X0(), true);
//for (double s = 0; s < T; s += h_obs)
//{
// model.Step();
//}
for (double s = h_state; s < T; s += h_state)
{
if (s > 0)
{
State = State + h_state * Phi1(State) + Math.Sqrt(h_state) * Phi2() * W();
}
if (Math.Abs(s - t_nextobservation) < h_tolerance)
{
Obs = Psi1(State) + Psi2() * Nu();
t_nextobservation += h_obs;
n++;
model.Trajectory.Add(n, new Vector <double>[] { State, Obs });
}
}
return(model);
};
// filter params file names
string CMNFFileName = Path.Combine(o.OutputFolder, "cmnf.params");
if (!string.IsNullOrWhiteSpace(o.CMNFFileName))
{
CMNFFileName = o.CMNFFileName;
}
string BCMNFFileName = Path.Combine(o.OutputFolder, "bcmnf.params");
if (!string.IsNullOrWhiteSpace(o.BCMNFFileName))
{
BCMNFFileName = o.BCMNFFileName;
}
string UKFFileName = Path.Combine(o.OutputFolder, "ukf.params");
if (!string.IsNullOrWhiteSpace(o.UKFFileName))
{
UKFFileName = o.UKFFileName;
}
string UKFOptStepwiseNMFileName = Path.Combine(o.OutputFolder, "ukfoptstepwiseNM.params");
if (!string.IsNullOrWhiteSpace(o.UKFStepwiseNelderMeadFileName))
{
UKFOptStepwiseNMFileName = o.UKFStepwiseNelderMeadFileName;
}
string UKFOptIntegralNMFileName = Path.Combine(o.OutputFolder, "ukfoptintegralNM.params");
if (!string.IsNullOrWhiteSpace(o.UKFIntegralNelderMeadFileName))
{
UKFOptIntegralNMFileName = o.UKFIntegralNelderMeadFileName;
}
string UKFOptStepwiseRandFileName = Path.Combine(o.OutputFolder, "ukfoptstepwiserand.params");
if (!string.IsNullOrWhiteSpace(o.UKFStepwiseRandomShootFileName))
{
UKFOptStepwiseRandFileName = o.UKFStepwiseRandomShootFileName;
}
string UKFOptIntegralRandFileName = Path.Combine(o.OutputFolder, "ukfoptintegralrand.params");
if (!string.IsNullOrWhiteSpace(o.UKFIntegralRandomShootFileName))
{
UKFOptIntegralRandFileName = o.UKFIntegralRandomShootFileName;
}
// filters
List <(FilterType, string)> filters = new List <(FilterType, string)>();
if (o.CMNF)
{
filters.Add((FilterType.CMNF, CMNFFileName));
}
if (o.BCMNF)
{
filters.Add((FilterType.BCMNF, BCMNFFileName));
}
if (o.MCMNF)
{
filters.Add((FilterType.MCMNF, string.Empty));
}
if (o.UKF)
{
filters.Add((FilterType.UKFNoOptimization, UKFFileName));
}
if (o.UKFStepwiseNelderMead)
{
filters.Add((FilterType.UKFStepwise, UKFOptStepwiseNMFileName));
}
if (o.UKFIntegralNelderMead)
{
filters.Add((FilterType.UKFIntegral, UKFOptIntegralNMFileName));
}
if (o.UKFStepwiseRandomShoot)
{
filters.Add((FilterType.UKFStepwiseRandomShoot, UKFOptStepwiseRandFileName));
}
if (o.UKFIntegralRandomShoot)
{
filters.Add((FilterType.UKFIntegralRandomShoot, UKFOptIntegralRandFileName));
}
if (o.EKF)
{
filters.Add((FilterType.EKF, string.Empty));
}
if (o.Dummy)
{
filters.Add((FilterType.Dummy, string.Empty));
}
// test environment
TestEnvironmentVector testEnv = new TestEnvironmentVector()
{
TestName = "Target tracking",
TestFileName = "TargetTracking",
Phi1 = Phi1_discr,
Phi2 = Phi2_discr,
Psi1 = Psi1_discr,
Psi2 = Psi2_discr,
dPhi = (i, x) => dPhi(x),
dPsi = (i, x) => dPsi(x),
Xi = (i, x) => Phi1_discr(i, x) + Phi2_discr(i, x) * mW,
//Zeta = (i, x, y, k) => (y - Psi1_discr(i, x) - Psi2_discr(i, x) * mNu).Stack(Utils.pol2cart(Exts.Vector(y[0], y[1]))+X_R1).Stack(Utils.pol2cart(Exts.Vector(y[2], y[3]))+X_R2),
Zeta = (i, x, y, k) => (y - Psi1_discr(i, x) - Psi2_discr(i, x) * mNu),
Alpha = (i, x) => Phi1_discr(i, x) + Phi2_discr(i, x) * mW,
Gamma = (i, x, y) => (y).Stack(Utils.pol2cart(Exts.Vector(y[0], y[1])) + X_R1).Stack(Utils.pol2cart(Exts.Vector(y[2], y[3])) + X_R2),
//Gamma = (i, x, y) => y - Psi1_discr(i, x) - Psi2_discr(i, x) * mNu,
nMCMNF = o.MCMNFTrainCount,
W = (i) => W(),
Nu = (i) => Nu(),
DW = dW,
DNu = dNu,
X0 = () => X0(),
X0Hat = mEta,
DX0Hat = dEta,
Predict = Ricatti,
DummyEstimate = DummyEstimate,
ModelGenerator = ModelGenerator
};
if (o.Bulk)
{
testEnv.GenerateBundleSamples(o.T, o.TrainCount, o.OutputFolder);
}
else
{
testEnv.Initialize(o.T, o.TrainCount, o.OutputFolder, filters, o.Save, o.Load);
if (o.Sift)
{
testEnv.Sifter = (x) => Math.Sqrt(x[0] * x[0] + x[1] * x[1]) > o.SiftBound;
}
if (o.Aggregate)
{
testEnv.Aggregate(o.OutputFolder, o.OutputFolder, !o.NoBin, !o.NoText);
}
if (!o.Skip)
{
if (o.SamplesCount == 0)
{
testEnv.GenerateBundles(o.BundleCount, o.TestCount, o.OutputFolder, o.Parallel, o.ParallelismDegree, !o.NoBin, !o.NoText);
if (!o.NoPython)
{
testEnv.RunScript(Path.Combine(o.ScriptsFolder, "estimate_statistics.py"), o.OutputFolder);
}
}
else
{
if (o.SamplesCount == 1)
{
testEnv.GenerateOne(o.OutputFolder);
if (!o.NoPython)
{
testEnv.RunScript(Path.Combine(o.ScriptsFolder, "estimate_sample.py"), o.OutputFolder);
testEnv.RunScript(Path.Combine(o.ScriptsFolder, "trajectory.py"), o.OutputFolder);
}
}
else
{
for (int i = 0; i < o.SamplesCount; i++)
{
testEnv.GenerateOne(o.OutputFolder, i);
}
}
}
}
}
}
/// <summary>
/// <para>Unscented transform parameters stepwize optimization procedure.</para>
/// <para>The OptimizationMethod param determines the optimization method:</para>
/// <para>- OptimizationMethod.RandomShoot - parameters are randomly sampled and the best sample is chosen as optimal;
/// <para>- OptimizationMethod.NelderMeed - parameters are optimized with non-gradient Nelder-Meed method.</para>
/// <para>The UTOptimizationType type param determines the relation between the optimized variable and the unscented tranform params (see UTParams and its constructors for details). </para>
/// <para>- If type is UTOptimizationType.ImplicitAlpha, then the optimized variable is saclar [alpha0];</para>
/// <para>- If type is UTOptimizationType.ImplicitAlphaBetaKappa, then optimized variable is a vector [alpha, beta, kappa];</para>
/// <para>- If type is UTOptimizationType.Explicit, then then optimized variable is a vector [lambda, wm0, wc0, wi]. ///TODO it is not correct to define the parameters of the unsctnted transform arbitraty, they have to be interdependent, so that the mean and cov would be transformed correctly.</para>
/// </summary>
/// <param name="method">Unscented transform parameters optimization method</param>
/// <param name="type">Unscented transform parameters definition type</param>
/// <param name="Phi1">State transformation: a nonlinear function which determines the dynamics: x_{t+1} = Phi_1(x_t) + Phi_2(x_t) W_t</param>
/// <param name="Phi2">Noise multiplicator in the dynamics equation: x_{t+1} = Phi(x_t) + W_t</param>
/// <param name="Psi1">Observations transformation: a nonlinear function which determines the relation between the state and the observations: y_t = Psi_1(x_t) + Psi_2(x_t) Nu_t</param>
/// <param name="Psi2">Noise multiplicator in the observations equation: y_t = Psi_1(x_t) + Psi_2(x_t) Nu_t</param>
/// <param name="Mw">Mean of the noise in the dynamics equation </param>
/// <param name="Rw">Covariance matrix of the state disturbances</param>
/// <param name="Mnu">Mean of the noise in the obseration equation </param>
/// <param name="Rnu">Convariance matrix of the observation noise</param>
/// <param name="Crit">Criterion: a function which determines the quality of the unscented Kalman filter. Depends on the sample covariance of the estimation error on the last step: val = Crit(Cov(X_T-Xhat_T,X_T-Xhat_T)) </param>
/// <param name="T">The upper bound of the observation interval</param>
/// <param name="models">Discrete vector model samples</param>
/// <param name="xhat0">Initial condition</param>
/// <param name="DX0Hat">Initial condition covariance</param>
/// <param name="outputFolder">The results are saved to this folder in file "UT_optimization_{type}.txt"</param>
static (double, UTParams[], UTParams[]) UTParmsOptimizeStepwise(OptimizationMethod method, UTDefinitionType type,
Func <int, Vector <double>, Vector <double> > Phi1,
Func <int, Vector <double>, Matrix <double> > Phi2,
Func <int, Vector <double>, Vector <double> > Psi1,
Func <int, Vector <double>, Matrix <double> > Psi2,
Vector <double> Mw,
Matrix <double> Rw,
Vector <double> Mnu,
Matrix <double> Rnu,
Func <Matrix <double>, double> Crit,
int T,
DiscreteVectorModel[] models,
Vector <double> xhat0,
Matrix <double> DX0Hat,
string outputFolder)
{
UTParams[] pForecast = new UTParams[T];
UTParams[] pCorrect = new UTParams[T];
(int n, Vector <double> lowerBound, Vector <double> upperBound, Vector <double> initialGuess, string filename) = DefineOptimizationParameters(type, xhat0, string.IsNullOrWhiteSpace(outputFolder) ? null : Path.Combine(outputFolder, "UT_stepwise_ptimization_{type}.txt"));
Vector <double>[] xHatU = models.Select(x => xhat0).ToArray();
Matrix <double>[] PHatU = models.Select(x => DX0Hat).ToArray();
double min = double.MaxValue;
Console.WriteLine($"UKF estimate parameters start");
DateTime start = DateTime.Now;
for (int t = 1; t < T; t++)
//Parallel.For(0, T, new ParallelOptions() { MaxDegreeOfParallelism = System.Environment.ProcessorCount }, t =>
{
DateTime startiteration = DateTime.Now;
min = double.MaxValue;
Vector <double> argmin = initialGuess;
switch (method)
{
case OptimizationMethod.RandomShoot:
var OptimumRandom = RandomOptimizer.Minimize((x) => CalculateSampleStepwiseCriterion(Phi1, Phi2, Psi1, Psi2, Mw, Rw, Mnu, Rnu, Crit, x, t, models, xHatU, PHatU), Exts.Stack(lowerBound, lowerBound), Exts.Stack(upperBound, upperBound), 100, 100, filename);
min = OptimumRandom.min;
argmin = OptimumRandom.argmin;
break;
case OptimizationMethod.NelderMeed:
NelderMeadSimplex optimizer = new NelderMeadSimplex(1e-3, 100);
var objective = ObjectiveFunction.Value((x) => CalculateSampleStepwiseCriterion(Phi1, Phi2, Psi1, Psi2, Mw, Rw, Mnu, Rnu, Crit, x, t, models, xHatU, PHatU));
try
{
var optimumNM = optimizer.FindMinimum(objective, Exts.Stack(initialGuess, initialGuess));
min = optimumNM.FunctionInfoAtMinimum.Value;
argmin = optimumNM.MinimizingPoint;
}
catch (Exception e)
{
Console.WriteLine($"Optimizer faild, using the initail guess ({e.Message})");
argmin = Exts.Stack(initialGuess, initialGuess);
}
break;
}
pForecast[t] = new UTParams(xhat0.Count, argmin.Take(n).ToArray());
pCorrect[t] = new UTParams(xhat0.Count, argmin.Skip(n).Take(n).ToArray());
for (int i = 0; i < models.Count(); i++)
{
(xHatU[i], PHatU[i]) = Step(Phi1, Phi2, Psi1, Psi2, Mw, Rw, Mnu, Rnu, pForecast[t], pCorrect[t], t, models[i].Trajectory[t][1], xHatU[i], PHatU[i]);
}
Console.WriteLine($"UKF estimate parameters for t={t}, done in {(DateTime.Now - startiteration).ToString(@"hh\:mm\:ss\.fff")}");
}
// });
DateTime finish = DateTime.Now;
Console.WriteLine($"UKF estimate parameters finished in {(finish - start).ToString(@"hh\:mm\:ss\.fff")}");
return(min, pForecast, pCorrect);
}
static void Run(Options o, string[] args)
{
if (string.IsNullOrWhiteSpace(o.OutputFolder))
{
o.OutputFolder = Settings.Default.OutputFolder;
}
if (string.IsNullOrWhiteSpace(o.PlotsFolder))
{
o.PlotsFolder = Settings.Default.LatexFolder;
}
if (string.IsNullOrWhiteSpace(o.ScriptsFolder))
{
o.ScriptsFolder = Settings.Default.ScriptsFolder;
}
if (string.IsNullOrWhiteSpace(o.TemplatesFolder))
{
o.TemplatesFolder = Settings.Default.LatexFolder;
}
if (new[] { "sphere", "polar", "polartwo" }.Contains(o.Model))
{
#region sphere
if (o.Model == "sphere")
{
int N = o.N;
Vector <double> mX = Exts.Vector(30, 40, 100); Matrix <double> KX = Exts.Diag(30 * 30, 30 * 30, 30 * 30);
Vector <double> mNu = Exts.Vector(0, 0, 0); Matrix <double> KNu = Exts.Diag(30 * 30, Math.Pow(5 * Math.PI / 180.0, 2.0), Math.Pow(5 * Math.PI / 180.0, 2.0));
Normal[] NormalX = new Normal[3] {
new Normal(mX[0], Math.Sqrt(KX[0, 0])), new Normal(mX[1], Math.Sqrt(KX[1, 1])), new Normal(mX[2], Math.Sqrt(KX[2, 2]))
};
Normal[] NormalNu = new Normal[3] {
new Normal(mNu[0], Math.Sqrt(KNu[0, 0])), new Normal(mNu[1], Math.Sqrt(KNu[1, 1])), new Normal(mNu[2], Math.Sqrt(KNu[2, 2]))
};;
TestEnvironmentStatic testSphere = new TestEnvironmentStatic
{
Phi = x => Utils.cart2sphere(x),
InvPhi = y => Utils.sphere2cart(y),
W = () => Exts.Vector(NormalX[0].Sample(), NormalX[1].Sample(), NormalX[2].Sample()),
Nu = () => Exts.Vector(NormalNu[0].Sample(), NormalNu[1].Sample(), NormalNu[2].Sample()),
MX = mX,
KX = KX,
KNu = KNu
};
testSphere.Initialize(N, o.OutputFolder);
Vector <double> mErr;
Matrix <double> KErr;
Matrix <double> KErrTh;
Vector <double> mErr_inv;
Matrix <double> KErr_inv;
Matrix <double> KErrTh_inv;
Vector <double> mErr_lin;
Matrix <double> KErr_lin;
Matrix <double> KErrTh_lin;
Vector <double> mErr_UT;
Matrix <double> KErr_UT;
Matrix <double> KErrTh_UT;
string fileName_alldata = Path.Combine(o.OutputFolder, "test_sphere_alldata.txt");
testSphere.GenerateBundle(N, out mErr, out KErr, out KErrTh, out mErr_inv, out KErr_inv, out KErrTh_inv, out mErr_lin, out KErr_lin, out KErrTh_lin, out mErr_UT, out KErr_UT, out KErrTh_UT, fileName_alldata);
string fileName = Path.Combine(o.OutputFolder, "test_sphere.txt");
using (System.IO.StreamWriter outputfile = new System.IO.StreamWriter(fileName))
{
//outputfile.WriteLine($"P = {P}");
outputfile.WriteLine($"mErr = {mErr}");
outputfile.WriteLine($"KErr = {KErr}");
outputfile.WriteLine($"KErrTh = {KErrTh}");
//outputfile.WriteLine($"P_inv = {P_inv}");
outputfile.WriteLine($"mErr_inv = {mErr_inv}");
outputfile.WriteLine($"KErr_inv = {KErr_inv}");
outputfile.WriteLine($"KErrTh_inv = {KErrTh_inv}");
//outputfile.WriteLine($"P_lin = {P_lin}");
outputfile.WriteLine($"mErr_lin = {mErr_lin}");
outputfile.WriteLine($"KErr_lin = {KErr_lin}");
outputfile.WriteLine($"KErrTh_lin = {KErrTh_lin}");
//outputfile.WriteLine($"P_UT = {P_UT}");
outputfile.WriteLine($"mErr_UT = {mErr_UT}");
outputfile.WriteLine($"KErr_UT = {KErr_UT}");
outputfile.WriteLine($"KErrTh_UT = {KErrTh_UT}");
outputfile.Close();
}
}
#endregion
#region polar
if (o.Model == "polar")
{
int N = o.N;
Vector <double> mX = Exts.Vector(300, 400); Matrix <double> KX = Exts.Diag(30 * 30, 30 * 30);
//Vector<double> mX = Exts.Vector(30000, 40000); Matrix<double> KX = Exts.Diag(100 * 100, 100 * 100);
//Vector<double> mX = Exts.Vector(30000, 40000); Matrix<double> KX = Exts.Diag(4500 * 4500, 4500 * 4500);
Vector <double> mNu = Exts.Vector(0, 0); Matrix <double> KNu = Exts.Diag(Math.Pow(5 * Math.PI / 180.0, 2.0), 30 * 30);
Normal[] NormalX = new Normal[2] {
new Normal(mX[0], Math.Sqrt(KX[0, 0])), new Normal(mX[1], Math.Sqrt(KX[1, 1]))
};
Normal[] NormalNu = new Normal[2] {
new Normal(mNu[0], Math.Sqrt(KNu[0, 0])), new Normal(mNu[1], Math.Sqrt(KNu[1, 1]))
};;
//Console.WriteLine(mX.ToLine());
TestEnvironmentStatic testPolar = new TestEnvironmentStatic
{
Phi = x => Utils.cart2pol(x),
InvPhi = y => Utils.pol2cart(y),
W = () => Exts.Vector(NormalX[0].Sample(), NormalX[1].Sample()),
Nu = () => Exts.Vector(NormalNu[0].Sample(), NormalNu[1].Sample()),
MX = mX,
KX = KX,
KNu = KNu
};
testPolar.Initialize(N, o.OutputFolder);
Vector <double> mErr;
Matrix <double> KErr;
Matrix <double> KErrTh;
Vector <double> mErr_inv;
Matrix <double> KErr_inv;
Matrix <double> KErrTh_inv;
Vector <double> mErr_lin;
Matrix <double> KErr_lin;
Matrix <double> KErrTh_lin;
Vector <double> mErr_UT;
Matrix <double> KErr_UT;
Matrix <double> KErrTh_UT;
string fileName_alldata = Path.Combine(o.OutputFolder, "test_polar_alldata.txt");
testPolar.GenerateBundle(N, out mErr, out KErr, out KErrTh, out mErr_inv, out KErr_inv, out KErrTh_inv, out mErr_lin, out KErr_lin, out KErrTh_lin, out mErr_UT, out KErr_UT, out KErrTh_UT, fileName_alldata);
string fileName = Path.Combine(o.OutputFolder, "test_polar.txt");
using (System.IO.StreamWriter outputfile = new System.IO.StreamWriter(fileName))
{
//outputfile.WriteLine($"P = {P}");
outputfile.WriteLine($"mErr = {mErr}");
outputfile.WriteLine($"KErr = {KErr}");
outputfile.WriteLine($"KErrTh = {KErrTh}");
//outputfile.WriteLine($"P_inv = {P_inv}");
outputfile.WriteLine($"mErr_inv = {mErr_inv}");
outputfile.WriteLine($"KErr_inv = {KErr_inv}");
outputfile.WriteLine($"KErrTh_inv = {KErrTh_inv}");
//outputfile.WriteLine($"P_lin = {P_lin}");
outputfile.WriteLine($"mErr_lin = {mErr_lin}");
outputfile.WriteLine($"KErr_lin = {KErr_lin}");
outputfile.WriteLine($"KErrTh_lin = {KErrTh_lin}");
//outputfile.WriteLine($"P_UT = {P_UT}");
outputfile.WriteLine($"mErr_UT = {mErr_UT}");
outputfile.WriteLine($"KErr_UT = {KErr_UT}");
outputfile.WriteLine($"KErrTh_UT = {KErrTh_UT}");
outputfile.Close();
}
}
#endregion
#region polartwo
if (o.Model == "polartwo")
{
int N = o.N;
Vector <double> secondpoint = Exts.Vector(-10000, 10000);
Vector <double> mX = Exts.Vector(30000, 40000); Matrix <double> KX = Exts.Diag(2000 * 2000, 2000 * 2000);
Vector <double> mNu = Exts.Vector(0, 0, 0, 0);
Matrix <double> KNu = Exts.Diag(Math.Pow(0.1 * Math.PI / 180.0, 2.0),
50 * 50,
Math.Pow(0.1 * Math.PI / 180.0, 2.0),
50 * 50);
Normal[] NormalX = new Normal[2] {
new Normal(mX[0], Math.Sqrt(KX[0, 0])), new Normal(mX[1], Math.Sqrt(KX[1, 1]))
};
Normal[] NormalNu = new Normal[4] {
new Normal(mNu[0], Math.Sqrt(KNu[0, 0])),
new Normal(mNu[1], Math.Sqrt(KNu[1, 1])),
new Normal(mNu[2], Math.Sqrt(KNu[2, 2])),
new Normal(mNu[3], Math.Sqrt(KNu[3, 3]))
};
//Console.WriteLine(mX.ToLine());
TestEnvironmentStatic testPolar = new TestEnvironmentStatic
{
Phi = x => Exts.Stack(Utils.cart2pol(x), Utils.cart2pol(x - secondpoint)),
InvPhi = y => Exts.Stack(Utils.pol2cart(Exts.Vector(y[0], y[1])), Utils.pol2cart(Exts.Vector(y[2], y[3])) + secondpoint),
W = () => Exts.Vector(NormalX[0].Sample(), NormalX[1].Sample()),
Nu = () => Exts.Vector(NormalNu[0].Sample(), NormalNu[1].Sample(), NormalNu[2].Sample(), NormalNu[3].Sample()),
//Nu = () => Exts.Vector(0,0,0,0),
MX = mX,
KX = KX,
KNu = KNu
};
testPolar.Initialize(N, o.OutputFolder);
Vector <double> mErr;
Matrix <double> KErr;
Matrix <double> KErrTh;
Vector <double> mErr_inv;
Matrix <double> KErr_inv;
Matrix <double> KErrTh_inv;
Vector <double> mErr_lin;
Matrix <double> KErr_lin;
Matrix <double> KErrTh_lin;
Vector <double> mErr_UT;
Matrix <double> KErr_UT;
Matrix <double> KErrTh_UT;
string fileName_alldata = Path.Combine(o.OutputFolder, "test_polartwo_alldata.txt");
testPolar.GenerateBundle(N, out mErr, out KErr, out KErrTh, out mErr_inv, out KErr_inv, out KErrTh_inv, out mErr_lin, out KErr_lin, out KErrTh_lin, out mErr_UT, out KErr_UT, out KErrTh_UT, fileName_alldata);
string fileName = Path.Combine(o.OutputFolder, "test_polartwo.txt");
using (System.IO.StreamWriter outputfile = new System.IO.StreamWriter(fileName))
{
//outputfile.WriteLine($"P = {P}");
outputfile.WriteLine($"mErr = {mErr}");
outputfile.WriteLine($"KErr = {KErr}");
outputfile.WriteLine($"KErrTh = {KErrTh}");
//outputfile.WriteLine($"P_inv = {P_inv}");
outputfile.WriteLine($"mErr_inv = {mErr_inv}");
outputfile.WriteLine($"KErr_inv = {KErr_inv}");
outputfile.WriteLine($"KErrTh_inv = {KErrTh_inv}");
//outputfile.WriteLine($"P_lin = {P_lin}");
outputfile.WriteLine($"mErr_lin = {mErr_lin}");
outputfile.WriteLine($"KErr_lin = {KErr_lin}");
outputfile.WriteLine($"KErrTh_lin = {KErrTh_lin}");
//outputfile.WriteLine($"P_UT = {P_UT}");
outputfile.WriteLine($"mErr_UT = {mErr_UT}");
outputfile.WriteLine($"KErr_UT = {KErr_UT}");
outputfile.WriteLine($"KErrTh_UT = {KErrTh_UT}");
outputfile.Close();
}
}
#endregion
}
else
{
TestEnvironmentVector testEnv = new TestEnvironmentVector();
if (o.Model == "cubic")
{
testEnv = new TestCubicSensorScalar(o.DW, o.DNu);
}
if (o.Model == "invprop-good")
{
testEnv = new TestInverseProportionGoodScalar(o.Bound, o.DW, o.DNu);
}
if (o.Model == "invprop-bad")
{
testEnv = new TestInverseProportionBadScalar(o.Bound, o.DW, o.DNu);
}
if (o.Model == "logreg-simple")
{
testEnv = new TestLogisticModelScalar(o.Bound, o.DW, o.DNu);
}
if (o.Model == "logreg-zero")
{
testEnv = new TestLogisticModelZeroScalar(o.Bound, o.DW, o.DNu);
}
if (o.Model == "logreg-uniform")
{
testEnv = new TestLogisticModelUniformNoiseScalar();
}
if (o.Model == "samplereg")
{
testEnv = new TestSampledRegression(o.DNu);
}
if (o.Model == "switchingobs")
{
testEnv = new TestSwitchingObservations(o.DNu);
}
if (o.Model == "switchingobsident")
{
switch (o.IdentNumber)
{
case 1: testEnv = new AnotherTestSwitchingObservationsIdentification(o.DNu); break;
case 2: testEnv = new YetAnotherTestSwitchingObservationsIdentification(o.DNu); break;
case 3: testEnv = new HopefullyTheLastTestSwitchingObservationsIdentification(o.DNu); break;
default: testEnv = new TestSwitchingObservationsIdentification(o.DNu); break;
}
}
if (o.Model == "simpleident")
{
testEnv = new TestSimpleIdentification();
}
string CMNFFileName = Path.Combine(o.OutputFolder, "cmnf.params");
if (!string.IsNullOrWhiteSpace(o.CMNFFileName))
{
CMNFFileName = o.CMNFFileName;
}
string BCMNFFileName = Path.Combine(o.OutputFolder, "bcmnf.params");
if (!string.IsNullOrWhiteSpace(o.BCMNFFileName))
{
BCMNFFileName = o.BCMNFFileName;
}
string UKFFileName = Path.Combine(o.OutputFolder, "ukf.params");
if (!string.IsNullOrWhiteSpace(o.UKFFileName))
{
UKFFileName = o.UKFFileName;
}
string UKFOptStepwiseNMFileName = Path.Combine(o.OutputFolder, "ukfoptstepwiseNM.params");
if (!string.IsNullOrWhiteSpace(o.UKFStepwiseNelderMeadFileName))
{
UKFOptStepwiseNMFileName = o.UKFStepwiseNelderMeadFileName;
}
string UKFOptIntegralNMFileName = Path.Combine(o.OutputFolder, "ukfoptintegralNM.params");
if (!string.IsNullOrWhiteSpace(o.UKFIntegralNelderMeadFileName))
{
UKFOptIntegralNMFileName = o.UKFIntegralNelderMeadFileName;
}
string UKFOptStepwiseRandFileName = Path.Combine(o.OutputFolder, "ukfoptstepwiserand.params");
if (!string.IsNullOrWhiteSpace(o.UKFStepwiseRandomShootFileName))
{
UKFOptStepwiseRandFileName = o.UKFStepwiseRandomShootFileName;
}
string UKFOptIntegralRandFileName = Path.Combine(o.OutputFolder, "ukfoptintegralrand.params");
if (!string.IsNullOrWhiteSpace(o.UKFIntegralRandomShootFileName))
{
UKFOptIntegralRandFileName = o.UKFIntegralRandomShootFileName;
}
List <(FilterType, string)> filters = new List <(FilterType, string)>();
if (o.CMNF)
{
filters.Add((FilterType.CMNF, CMNFFileName));
}
if (o.BCMNF)
{
testEnv.Alpha = testEnv.Xi;
testEnv.Gamma = (t, x, y) => y;
filters.Add((FilterType.BCMNF, BCMNFFileName));
}
if (o.MCMNF)
{
testEnv.nMCMNF = o.MCMNFTrainCount;
filters.Add((FilterType.MCMNF, string.Empty));
}
if (o.UKF)
{
filters.Add((FilterType.UKFNoOptimization, UKFFileName));
}
if (o.UKFStepwiseNelderMead)
{
filters.Add((FilterType.UKFStepwise, UKFOptStepwiseNMFileName));
}
if (o.UKFIntegralNelderMead)
{
filters.Add((FilterType.UKFIntegral, UKFOptIntegralNMFileName));
}
if (o.UKFStepwiseRandomShoot)
{
filters.Add((FilterType.UKFStepwiseRandomShoot, UKFOptStepwiseRandFileName));
}
if (o.UKFIntegralRandomShoot)
{
filters.Add((FilterType.UKFIntegralRandomShoot, UKFOptIntegralRandFileName));
}
if (o.EKF)
{
filters.Add((FilterType.EKF, string.Empty));
}
if (o.Dummy)
{
filters.Add((FilterType.Dummy, string.Empty));
}
using (System.IO.StreamWriter outputfile = new System.IO.StreamWriter(Path.Combine(o.OutputFolder, "parameters.txt"), true))
{
outputfile.WriteLine($"{DateTime.Now}\t{string.Join(" ", args)}");
outputfile.Close();
}
if (o.Bulk)
{
testEnv.GenerateBundleSamples(o.T, o.TrainCount, o.OutputFolder);
}
else
{
testEnv.Initialize(o.T, o.TrainCount, o.OutputFolder, filters, o.Save, o.Load);
if (o.Aggregate)
{
testEnv.Aggregate(o.OutputFolder, o.OutputFolder, !o.NoBin, !o.NoText);
}
if (!o.Skip)
{
testEnv.GenerateBundles(o.BundleCount, o.TestCount, o.OutputFolder, o.Parallel, o.ParallelismDegree, !o.NoBin, !o.NoText);
//if (o.BundleCount > 1)
// testEnv.GenerateBundles(o.BundleCount, o.TestCount, o.OutputFolder, o.Parallel, o.ParallelismDegree);
//else
// testEnv.GenerateBundle(o.TestCount, o.OutputFolder);
if (o.SamplesCount == 1)
{
testEnv.GenerateOne(o.OutputFolder);
}
else
{
for (int i = 0; i < o.SamplesCount; i++)
{
testEnv.GenerateOne(o.OutputFolder, i);
}
}
//testEnv.GenerateReport(o.TemplatesFolder, o.PlotsFolder);
}
if (!o.NoPython)
{
testEnv.ProcessResults(o.OutputFolder, o.ScriptsFolder, o.PlotsFolder);
}
}
}
}
