newmatnl.cc
Go to the documentation of this file.
00001 //$$ newmatnl.cpp         Non-linear optimisation
00002 
00003 // Copyright (C) 1993,4,5,6: R B Davies
00004 
00005 
00006 #define WANT_MATH
00007 #define WANT_STREAM
00008 
00009 #include "newmatap.h"
00010 #include "newmatnl.h"
00011 
00012 #ifdef use_namespace
00013 namespace NEWMAT {
00014 #endif
00015 
00016 
00017 
00018 void FindMaximum2::Fit(ColumnVector& Theta, int n_it)
00019 {
00020    Tracer tr("FindMaximum2::Fit");
00021    enum State {Start, Restart, Continue, Interpolate, Extrapolate,
00022       Fail, Convergence};
00023    State TheState = Start;
00024    Real z,w,x,x2,g,l1,l2,l3,d1,d2=0,d3;
00025    ColumnVector Theta1, Theta2, Theta3;
00026    int np = Theta.Nrows();
00027    ColumnVector H1(np), H3, HP(np), K, K1(np);
00028    bool oorg, conv;
00029    int counter = 0;
00030    Theta1 = Theta; HP = 0.0; g = 0.0;
00031 
00032    // This is really a set of gotos and labels, but they do not work
00033    // correctly in AT&T C++ and Sun 4.01 C++.
00034 
00035    for(;;)
00036    {
00037       switch (TheState)
00038       {
00039       case Start:
00040          tr.ReName("FindMaximum2::Fit/Start");
00041          Value(Theta1, true, l1, oorg);
00042          if (oorg) Throw(ProgramException("invalid starting value\n"));
00043 
00044       case Restart:
00045          tr.ReName("FindMaximum2::Fit/ReStart");
00046          conv = NextPoint(H1, d1);
00047          if (conv) { TheState = Convergence; break; }
00048          if (counter++ > n_it) { TheState = Fail; break; }
00049 
00050          z = 1.0 / sqrt(d1);
00051          H3 = H1 * z; K = (H3 - HP) * g; HP = H3;
00052          g = 0.0;                     // de-activate to use curved projection
00053          if (g==0.0) K1 = 0.0; else K1 = K * 0.2 + K1 * 0.6;
00054          // (K - K1) * alpha + K1 * (1 - alpha)
00055          //     = K * alpha + K1 * (1 - 2 * alpha)
00056          K = K1 * d1; g = z;
00057 
00058       case Continue:
00059          tr.ReName("FindMaximum2::Fit/Continue");
00060          Theta2 = Theta1 + H1 + K;
00061          Value(Theta2, false, l2, oorg);
00062          if (counter++ > n_it) { TheState = Fail; break; }
00063          if (oorg)
00064          {
00065             H1 *= 0.5; K *= 0.25; d1 *= 0.5; g *= 2.0;
00066             TheState =  Continue; break;
00067          }
00068          d2 = LastDerivative(H1 + K * 2.0);
00069 
00070       case Interpolate:
00071          tr.ReName("FindMaximum2::Fit/Interpolate");
00072          z = d1 + d2 - 3.0 * (l2 - l1);
00073          w = z * z - d1 * d2;
00074          if (w < 0.0) { TheState = Extrapolate; break; }
00075          w = z + sqrt(w);
00076          if (1.5 * w + d1 < 0.0)
00077             { TheState = Extrapolate; break; }
00078          if (d2 > 0.0 && l2 > l1 && w > 0.0)
00079             { TheState = Extrapolate; break; }
00080          x = d1 / (w + d1); x2 = x * x; g /= x;
00081          Theta3 = Theta1 + H1 * x + K * x2;
00082          Value(Theta3, true, l3, oorg);
00083          if (counter++ > n_it) { TheState = Fail; break; }
00084          if (oorg)
00085          {
00086             if (x <= 1.0)
00087                { x *= 0.5; x2 = x*x; g *= 2.0; d1 *= x; H1 *= x; K *= x2; }
00088             else
00089             {
00090                x = 0.5 * (x-1.0); x2 = x*x; Theta1 = Theta2;
00091                H1 = (H1 + K * 2.0) * x;
00092                K *= x2; g = 0.0; d1 = x * d2; l1 = l2;
00093             }
00094             TheState = Continue; break;
00095          }
00096 
00097          if (l3 >= l1 && l3 >= l2)
00098             { Theta1 = Theta3; l1 = l3; TheState =  Restart; break; }
00099 
00100          d3 = LastDerivative(H1 + K * 2.0);
00101          if (l1 > l2)
00102             { H1 *= x; K *= x2; Theta2 = Theta3; d1 *= x; d2 = d3*x; }
00103          else
00104          {
00105             Theta1 = Theta2; Theta2 = Theta3;
00106             x -= 1.0; x2 = x*x; g = 0.0; H1 = (H1 + K * 2.0) * x;
00107             K *= x2; l1 = l2; l2 = l3; d1 = x*d2; d2 = x*d3;
00108             if (d1 <= 0.0) { TheState = Start; break; }
00109          }
00110          TheState =  Interpolate; break;
00111 
00112       case Extrapolate:
00113          tr.ReName("FindMaximum2::Fit/Extrapolate");
00114          Theta1 = Theta2; g = 0.0; K *= 4.0; H1 = (H1 * 2.0 + K);
00115          d1 = 2.0 * d2; l1 = l2;
00116          TheState = Continue; break;
00117 
00118       case Fail:
00119          Throw(ConvergenceException(Theta));
00120 
00121       case Convergence:
00122          Theta = Theta1; return;
00123       }
00124    }
00125 }
00126 
00127 
00128 
00129 void NonLinearLeastSquares::Value
00130    (const ColumnVector& Parameters, bool, Real& v, bool& oorg)
00131 {
00132    Tracer tr("NonLinearLeastSquares::Value");
00133    Y.ReSize(n_obs); X.ReSize(n_obs,n_param);
00134    // put the fitted values in Y, the derivatives in X.
00135    Pred.Set(Parameters);
00136    if (!Pred.IsValid()) { oorg=true; return; }
00137    for (int i=1; i<=n_obs; i++)
00138    {
00139       Y(i) = Pred(i);
00140       X.Row(i) = Pred.Derivatives();
00141    }
00142    if (!Pred.IsValid()) { oorg=true; return; }  // check afterwards as well
00143    Y = *DataPointer - Y; Real ssq = Y.SumSquare();
00144    errorvar =  ssq / (n_obs - n_param);
00145    cout << "\n" << setw(15) << setprecision(10) << " " << errorvar;
00146    Derivs = Y.t() * X;          // get the derivative and stash it
00147    oorg = false; v = -0.5 * ssq;
00148 }
00149 
00150 bool NonLinearLeastSquares::NextPoint(ColumnVector& Adj, Real& test)
00151 {
00152    Tracer tr("NonLinearLeastSquares::NextPoint");
00153    QRZ(X, U); QRZ(X, Y, M);     // do the QR decomposition
00154    test = M.SumSquare();
00155    cout << " " << setw(15) << setprecision(10)
00156       << test << " " << Y.SumSquare() / (n_obs - n_param);
00157    Adj = U.i() * M;
00158    if (test < errorvar * criterion) return true;
00159    else return false;
00160 }
00161 
00162 Real NonLinearLeastSquares::LastDerivative(const ColumnVector& H)
00163 { return (Derivs * H).AsScalar(); }
00164 
00165 void NonLinearLeastSquares::Fit(const ColumnVector& Data,
00166    ColumnVector& Parameters)
00167 {
00168    Tracer tr("NonLinearLeastSquares::Fit");
00169    n_param = Parameters.Nrows(); n_obs = Data.Nrows();
00170    DataPointer = &Data;
00171    FindMaximum2::Fit(Parameters, Lim);
00172    cout << "\nConverged\n";
00173 }
00174 
00175 void NonLinearLeastSquares::MakeCovariance()
00176 {
00177    if (Covariance.Nrows()==0)
00178    {
00179       UpperTriangularMatrix UI = U.i();
00180       Covariance << UI * UI.t() * errorvar;
00181       SE << Covariance;                 // get diagonals
00182       for (int i = 1; i<=n_param; i++) SE(i) = sqrt(SE(i));
00183    }
00184 }
00185 
00186 void NonLinearLeastSquares::GetStandardErrors(ColumnVector& SEX)
00187    { MakeCovariance(); SEX = SE.AsColumn(); }
00188 
00189 void NonLinearLeastSquares::GetCorrelations(SymmetricMatrix& Corr)
00190    { MakeCovariance(); Corr << SE.i() * Covariance * SE.i(); }
00191 
00192 void NonLinearLeastSquares::GetHatDiagonal(DiagonalMatrix& Hat) const
00193 {
00194    Hat.ReSize(n_obs);
00195    for (int i = 1; i<=n_obs; i++) Hat(i) = X.Row(i).SumSquare();
00196 }
00197 
00198 
00199 // the MLE_D_FI routines
00200 
00201 void MLE_D_FI::Value
00202    (const ColumnVector& Parameters, bool wg, Real& v, bool& oorg)
00203 {
00204    Tracer tr("MLE_D_FI::Value");
00205    if (!LL.IsValid(Parameters,wg)) { oorg=true; return; }
00206    v = LL.LogLikelihood();
00207    if (!LL.IsValid()) { oorg=true; return; }     // check validity again
00208    cout << "\n" << setw(20) << setprecision(10) << v;
00209    oorg = false;
00210    Derivs = LL.Derivatives();                    // Get derivatives
00211 }
00212 
00213 bool MLE_D_FI::NextPoint(ColumnVector& Adj, Real& test)
00214 {
00215    Tracer tr("MLE_D_FI::NextPoint");
00216    SymmetricMatrix FI = LL.FI();
00217    LT = Cholesky(FI);
00218    ColumnVector Adj1 = LT.i() * Derivs;
00219    Adj = LT.t().i() * Adj1;
00220    test = SumSquare(Adj1);
00221    cout << "   " << setw(20) << setprecision(10) << test;
00222    return (test < Criterion);
00223 }
00224 
00225 Real MLE_D_FI::LastDerivative(const ColumnVector& H)
00226 { return (Derivs.t() * H).AsScalar(); }
00227 
00228 void MLE_D_FI::Fit(ColumnVector& Parameters)
00229 {
00230    Tracer tr("MLE_D_FI::Fit");
00231    FindMaximum2::Fit(Parameters,Lim);
00232    cout << "\nConverged\n";
00233 }
00234   
00235 void MLE_D_FI::MakeCovariance()
00236 {
00237    if (Covariance.Nrows()==0)
00238    {
00239       LowerTriangularMatrix LTI = LT.i();
00240       Covariance << LTI.t() * LTI;
00241       SE << Covariance;                // get diagonal
00242       int n = Covariance.Nrows();
00243       for (int i=1; i <= n; i++) SE(i) = sqrt(SE(i));
00244    }
00245 }
00246 
00247 void MLE_D_FI::GetStandardErrors(ColumnVector& SEX)
00248 { MakeCovariance(); SEX = SE.AsColumn(); }
00249    
00250 void MLE_D_FI::GetCorrelations(SymmetricMatrix& Corr)
00251 { MakeCovariance(); Corr << SE.i() * Covariance * SE.i(); }
00252 
00253 
00254 
00255 #ifdef use_namespace
00256 }
00257 #endif
00258 


rl_agent
Author(s): Todd Hester
autogenerated on Thu Jun 6 2019 22:00:13