devtools/eigen
1
0
Fork 0
mirror of https://gitlab.com/libeigen/eigen.git synced 2026-04-10 11:34:33 +08:00
Code Issues Packages Projects Releases Wiki Activity

define and use struct Parameters

Browse Source
This commit is contained in:
Thomas Capricelli
2009-08-25 21:50:01 +02:00
parent d13bcdc891
commit e465ea82e1
3 changed files with 187 additions and 147 deletions
Download Patch File Download Diff File Expand all files Collapse all files

68
unsupported/Eigen/src/NonLinear/HybridNonLinearSolver.h
View File

@@ -17,6 +17,16 @@ public:
UserAksed = 6
};
struct Parameters {
Parameters()
: factor(Scalar(100.))
, maxfev(1000)
, xtol(ei_sqrt(epsilon<Scalar>())) {}
Scalar factor;
int maxfev; // maximum number of function evaluation
Scalar xtol;
};
Status solve(
Matrix< Scalar, Dynamic, 1 > &x,
const Scalar tol = ei_sqrt(epsilon<Scalar>())
@@ -25,10 +35,8 @@ public:
Matrix< Scalar, Dynamic, 1 > &x,
int &nfev, int &njev,
Matrix< Scalar, Dynamic, 1 > &diag,
const int mode=1,
const int maxfev = 1000,
const Scalar factor = Scalar(100.),
const Scalar xtol = ei_sqrt(epsilon<Scalar>())
const Parameters &parameters,
const int mode=1
);
Status solveNumericalDiff(
@@ -39,12 +47,10 @@ public:
Matrix< Scalar, Dynamic, 1 > &x,
int &nfev,
Matrix< Scalar, Dynamic, 1 > &diag,
const Parameters &parameters,
const int mode=1,
int nb_of_subdiagonals = -1,
int nb_of_superdiagonals = -1,
const int maxfev = 2000,
const Scalar factor = Scalar(100.),
const Scalar xtol = ei_sqrt(epsilon<Scalar>()),
const Scalar epsfcn = Scalar(0.)
);
@@ -68,6 +74,7 @@ HybridNonLinearSolver<FunctorType,Scalar>::solve(
const int n = x.size();
int nfev=0, njev=0;
Matrix< Scalar, Dynamic, 1> diag;
Parameters parameters;
/* check the input parameters for errors. */
if (n <= 0 || tol < 0.) {
@@ -75,15 +82,15 @@ HybridNonLinearSolver<FunctorType,Scalar>::solve(
return ImproperInputParameters;
}
parameters.maxfev = 100*(n+1);
parameters.xtol = tol;
diag.setConstant(n, 1.);
return solve(
x,
nfev, njev,
diag,
2,
(n+1)*100,
100.,
tol
parameters,
2
);
}
@@ -96,10 +103,8 @@ HybridNonLinearSolver<FunctorType,Scalar>::solve(
int &nfev,
int &njev,
Matrix< Scalar, Dynamic, 1 > &diag,
const int mode,
const int maxfev,
const Scalar factor,
const Scalar xtol
const Parameters &parameters,
const int mode
)
{
const int n = x.size();
@@ -133,7 +138,7 @@ HybridNonLinearSolver<FunctorType,Scalar>::solve(
/* check the input parameters for errors. */
if (n <= 0 || xtol < 0. || maxfev <= 0 || factor <= 0. )
if (n <= 0 || parameters.xtol < 0. || parameters.maxfev <= 0 || parameters.factor <= 0. )
return RelativeErrorTooSmall;
if (mode == 2)
for (j = 0; j < n; ++j)
@@ -187,9 +192,9 @@ HybridNonLinearSolver<FunctorType,Scalar>::solve(
wa3 = diag.cwise() * x;
xnorm = wa3.stableNorm();
delta = factor * xnorm;
delta = parameters.factor * xnorm;
if (delta == 0.)
delta = factor;
delta = parameters.factor;
}
/* form (q transpose)*fvec and store in qtf. */
@@ -326,12 +331,12 @@ HybridNonLinearSolver<FunctorType,Scalar>::solve(
/* test for convergence. */
if (delta <= xtol * xnorm || fnorm == 0.)
if (delta <= parameters.xtol * xnorm || fnorm == 0.)
return RelativeErrorTooSmall;
/* tests for termination and stringent tolerances. */
if (nfev >= maxfev)
if (nfev >= parameters.maxfev)
return TooManyFunctionEvaluation;
/* Computing MAX */
if (Scalar(.1) * std::max(Scalar(.1) * delta, pnorm) <= epsilon<Scalar>() * xnorm)
@@ -384,6 +389,7 @@ HybridNonLinearSolver<FunctorType,Scalar>::solveNumericalDiff(
const int n = x.size();
int nfev=0;
Matrix< Scalar, Dynamic, 1> diag;
Parameters parameters;
/* check the input parameters for errors. */
if (n <= 0 || tol < 0.) {
@@ -391,16 +397,18 @@ HybridNonLinearSolver<FunctorType,Scalar>::solveNumericalDiff(
return ImproperInputParameters;
}
parameters.maxfev = 200*(n+1);
parameters.xtol = tol;
diag.setConstant(n, 1.);
return solveNumericalDiff(
x,
nfev,
diag,
parameters,
2,
-1, -1,
(n+1)*200,
100.,
tol, Scalar(0.)
Scalar(0.)
);
}
@@ -411,12 +419,10 @@ HybridNonLinearSolver<FunctorType,Scalar>::solveNumericalDiff(
Matrix< Scalar, Dynamic, 1 > &x,
int &nfev,
Matrix< Scalar, Dynamic, 1 > &diag,
const Parameters &parameters,
const int mode,
int nb_of_subdiagonals,
int nb_of_superdiagonals,
const int maxfev,
const Scalar factor,
const Scalar xtol,
const Scalar epsfcn
)
{
@@ -454,7 +460,7 @@ HybridNonLinearSolver<FunctorType,Scalar>::solveNumericalDiff(
/* check the input parameters for errors. */
if (n <= 0 || xtol < 0. || maxfev <= 0 || nb_of_subdiagonals < 0 || nb_of_superdiagonals < 0 || factor <= 0. )
if (n <= 0 || parameters.xtol < 0. || parameters.maxfev <= 0 || nb_of_subdiagonals < 0 || nb_of_superdiagonals < 0 || parameters.factor <= 0. )
return RelativeErrorTooSmall;
if (mode == 2)
for (j = 0; j < n; ++j)
@@ -514,9 +520,9 @@ HybridNonLinearSolver<FunctorType,Scalar>::solveNumericalDiff(
wa3 = diag.cwise() * x;
xnorm = wa3.stableNorm();
delta = factor * xnorm;
delta = parameters.factor * xnorm;
if (delta == 0.)
delta = factor;
delta = parameters.factor;
}
/* form (q transpose)*fvec and store in qtf. */
@@ -653,12 +659,12 @@ HybridNonLinearSolver<FunctorType,Scalar>::solveNumericalDiff(
/* test for convergence. */
if (delta <= xtol * xnorm || fnorm == 0.)
if (delta <= parameters.xtol * xnorm || fnorm == 0.)
return RelativeErrorTooSmall;
/* tests for termination and stringent tolerances. */
if (nfev >= maxfev)
if (nfev >= parameters.maxfev)
return TooManyFunctionEvaluation;
/* Computing MAX */
if (Scalar(.1) * std::max(Scalar(.1) * delta, pnorm) <= epsilon<Scalar>() * xnorm)

139
unsupported/Eigen/src/NonLinear/LevenbergMarquardt.h
View File

@@ -1,5 +1,4 @@
template<typename FunctorType, typename Scalar=double>
class LevenbergMarquardt
{
@@ -21,6 +20,20 @@ public:
UserAsked = 9
};
struct Parameters {
Parameters()
: factor(Scalar(100.))
, maxfev(400)
, ftol(ei_sqrt(epsilon<Scalar>()))
, xtol(ei_sqrt(epsilon<Scalar>()))
, gtol(Scalar(0.)) { }
Scalar factor;
int maxfev; // maximum number of function evaluation
Scalar ftol;
Scalar xtol;
Scalar gtol;
};
Status minimize(
Matrix< Scalar, Dynamic, 1 > &x,
const Scalar tol = ei_sqrt(epsilon<Scalar>())
@@ -31,12 +44,8 @@ public:
int &nfev,
int &njev,
Matrix< Scalar, Dynamic, 1 > &diag,
const int mode=1,
const Scalar factor = Scalar(100.),
const int maxfev = 400,
const Scalar ftol = ei_sqrt(epsilon<Scalar>()),
const Scalar xtol = ei_sqrt(epsilon<Scalar>()),
const Scalar gtol = Scalar(0.)
const Parameters &parameters,
const int mode=1
);
Status minimizeNumericalDiff(
@@ -48,12 +57,8 @@ public:
Matrix< Scalar, Dynamic, 1 > &x,
int &nfev,
Matrix< Scalar, Dynamic, 1 > &diag,
const Parameters &parameters,
const int mode=1,
const Scalar factor = Scalar(100.),
const int maxfev = 400,
const Scalar ftol = ei_sqrt(epsilon<Scalar>()),
const Scalar xtol = ei_sqrt(epsilon<Scalar>()),
const Scalar gtol = Scalar(0.),
const Scalar epsfcn = Scalar(0.)
);
@@ -67,12 +72,8 @@ public:
int &nfev,
int &njev,
Matrix< Scalar, Dynamic, 1 > &diag,
const int mode=1,
const Scalar factor = Scalar(100.),
const int maxfev = 400,
const Scalar ftol = ei_sqrt(epsilon<Scalar>()),
const Scalar xtol = ei_sqrt(epsilon<Scalar>()),
const Scalar gtol = Scalar(0.)
const Parameters &parameters,
const int mode=1
);
Matrix< Scalar, Dynamic, 1 > fvec;
@@ -96,6 +97,7 @@ LevenbergMarquardt<FunctorType,Scalar>::minimize(
Matrix< Scalar, Dynamic, Dynamic > fjac(m, n);
Matrix< Scalar, Dynamic, 1> diag, qtf;
VectorXi ipvt;
Parameters parameters;
/* check the input parameters for errors. */
if (n <= 0 || m < n || tol < 0.) {
@@ -103,14 +105,16 @@ LevenbergMarquardt<FunctorType,Scalar>::minimize(
return ImproperInputParameters;
}
parameters.ftol = tol;
parameters.xtol = tol;
parameters.maxfev = 100*(n+1);
return minimize(
x,
nfev, njev,
diag,
1,
100.,
(n+1)*100,
tol, tol, Scalar(0.)
parameters,
1
);
}
@@ -122,12 +126,8 @@ LevenbergMarquardt<FunctorType,Scalar>::minimize(
int &nfev,
int &njev,
Matrix< Scalar, Dynamic, 1 > &diag,
const int mode,
const Scalar factor,
const int maxfev,
const Scalar ftol,
const Scalar xtol,
const Scalar gtol
const Parameters &parameters,
const int mode
)
{
const int n = x.size();
@@ -156,7 +156,7 @@ LevenbergMarquardt<FunctorType,Scalar>::minimize(
/* check the input parameters for errors. */
if (n <= 0 || m < n || ftol < 0. || xtol < 0. || gtol < 0. || maxfev <= 0 || factor <= 0.)
if (n <= 0 || m < n || parameters.ftol < 0. || parameters.xtol < 0. || parameters.gtol < 0. || parameters.maxfev <= 0 || parameters.factor <= 0.)
return RelativeErrorTooSmall;
if (mode == 2)
@@ -208,9 +208,9 @@ LevenbergMarquardt<FunctorType,Scalar>::minimize(
wa3 = diag.cwise() * x;
xnorm = wa3.stableNorm();
delta = factor * xnorm;
delta = parameters.factor * xnorm;
if (delta == 0.)
delta = factor;
delta = parameters.factor;
}
/* form (q transpose)*fvec and store the first n components in */
@@ -247,7 +247,7 @@ LevenbergMarquardt<FunctorType,Scalar>::minimize(
/* test for convergence of the gradient norm. */
if (gnorm <= gtol)
if (gnorm <= parameters.gtol)
return CosinusTooSmall;
/* rescale if necessary. */
@@ -341,16 +341,16 @@ LevenbergMarquardt<FunctorType,Scalar>::minimize(
/* tests for convergence. */
if (ei_abs(actred) <= ftol && prered <= ftol && Scalar(.5) * ratio <= 1. && delta <= xtol * xnorm)
if (ei_abs(actred) <= parameters.ftol && prered <= parameters.ftol && Scalar(.5) * ratio <= 1. && delta <= parameters.xtol * xnorm)
return RelativeErrorAndReductionTooSmall;
if (ei_abs(actred) <= ftol && prered <= ftol && Scalar(.5) * ratio <= 1.)
if (ei_abs(actred) <= parameters.ftol && prered <= parameters.ftol && Scalar(.5) * ratio <= 1.)
return RelativeReductionTooSmall;
if (delta <= xtol * xnorm)
if (delta <= parameters.xtol * xnorm)
return RelativeErrorTooSmall;
/* tests for termination and stringent tolerances. */
if (nfev >= maxfev)
if (nfev >= parameters.maxfev)
return TooManyFunctionEvaluation;
if (ei_abs(actred) <= epsilon<Scalar>() && prered <= epsilon<Scalar>() && Scalar(.5) * ratio <= 1.)
return FtolTooSmall;
@@ -379,6 +379,7 @@ LevenbergMarquardt<FunctorType,Scalar>::minimizeNumericalDiff(
Matrix< Scalar, Dynamic, Dynamic > fjac(m, n);
Matrix< Scalar, Dynamic, 1> diag, qtf;
VectorXi ipvt;
Parameters parameters;
/* check the input parameters for errors. */
if (n <= 0 || m < n || tol < 0.) {
@@ -386,14 +387,17 @@ LevenbergMarquardt<FunctorType,Scalar>::minimizeNumericalDiff(
return ImproperInputParameters;
}
parameters.ftol = tol;
parameters.xtol = tol;
parameters.maxfev = 200*(n+1);
return minimizeNumericalDiff(
x,
nfev,
diag,
parameters,
1,
100.,
(n+1)*200,
tol, tol, Scalar(0.), Scalar(0.)
Scalar(0.)
);
}
@@ -403,12 +407,8 @@ LevenbergMarquardt<FunctorType,Scalar>::minimizeNumericalDiff(
Matrix< Scalar, Dynamic, 1 > &x,
int &nfev,
Matrix< Scalar, Dynamic, 1 > &diag,
const Parameters &parameters,
const int mode,
const Scalar factor,
const int maxfev,
const Scalar ftol,
const Scalar xtol,
const Scalar gtol,
const Scalar epsfcn
)
{
@@ -437,7 +437,7 @@ LevenbergMarquardt<FunctorType,Scalar>::minimizeNumericalDiff(
/* check the input parameters for errors. */
if (n <= 0 || m < n || ftol < 0. || xtol < 0. || gtol < 0. || maxfev <= 0 || factor <= 0.)
if (n <= 0 || m < n || parameters.ftol < 0. || parameters.xtol < 0. || parameters.gtol < 0. || parameters.maxfev <= 0 || parameters.factor <= 0.)
return RelativeErrorTooSmall;
if (mode == 2)
for (j = 0; j < n; ++j)
@@ -488,9 +488,9 @@ LevenbergMarquardt<FunctorType,Scalar>::minimizeNumericalDiff(
wa3 = diag.cwise() * x;
xnorm = wa3.stableNorm();
delta = factor * xnorm;
delta = parameters.factor * xnorm;
if (delta == 0.)
delta = factor;
delta = parameters.factor;
}
/* form (q transpose)*fvec and store the first n components in */
@@ -527,7 +527,7 @@ LevenbergMarquardt<FunctorType,Scalar>::minimizeNumericalDiff(
/* test for convergence of the gradient norm. */
if (gnorm <= gtol)
if (gnorm <= parameters.gtol)
return CosinusTooSmall;
/* rescale if necessary. */
@@ -621,16 +621,16 @@ LevenbergMarquardt<FunctorType,Scalar>::minimizeNumericalDiff(
/* tests for convergence. */
if (ei_abs(actred) <= ftol && prered <= ftol && Scalar(.5) * ratio <= 1. && delta <= xtol * xnorm)
if (ei_abs(actred) <= parameters.ftol && prered <= parameters.ftol && Scalar(.5) * ratio <= 1. && delta <= parameters.xtol * xnorm)
return RelativeErrorAndReductionTooSmall;
if (ei_abs(actred) <= ftol && prered <= ftol && Scalar(.5) * ratio <= 1.)
if (ei_abs(actred) <= parameters.ftol && prered <= parameters.ftol && Scalar(.5) * ratio <= 1.)
return RelativeReductionTooSmall;
if (delta <= xtol * xnorm)
if (delta <= parameters.xtol * xnorm)
return RelativeErrorTooSmall;
/* tests for termination and stringent tolerances. */
if (nfev >= maxfev)
if (nfev >= parameters.maxfev)
return TooManyFunctionEvaluation;
if (ei_abs(actred) <= epsilon<Scalar>() && prered <= epsilon<Scalar>() && Scalar(.5) * ratio <= 1.)
return FtolTooSmall;
@@ -660,6 +660,7 @@ LevenbergMarquardt<FunctorType,Scalar>::minimizeOptimumStorage(
Matrix< Scalar, Dynamic, Dynamic > fjac(m, n);
Matrix< Scalar, Dynamic, 1> diag, qtf;
VectorXi ipvt;
Parameters parameters;
/* check the input parameters for errors. */
if (n <= 0 || m < n || tol < 0.) {
@@ -667,14 +668,16 @@ LevenbergMarquardt<FunctorType,Scalar>::minimizeOptimumStorage(
return ImproperInputParameters;
}
parameters.ftol = tol;
parameters.xtol = tol;
parameters.maxfev = 100*(n+1);
return minimizeOptimumStorage(
x,
nfev, njev,
diag,
1,
100.,
(n+1)*100,
tol, tol, Scalar(0.)
parameters,
1
);
}
@@ -685,12 +688,8 @@ LevenbergMarquardt<FunctorType,Scalar>::minimizeOptimumStorage(
int &nfev,
int &njev,
Matrix< Scalar, Dynamic, 1 > &diag,
const int mode,
const Scalar factor,
const int maxfev,
const Scalar ftol,
const Scalar xtol,
const Scalar gtol
const Parameters &parameters,
const int mode
)
{
const int n = x.size();
@@ -720,7 +719,7 @@ LevenbergMarquardt<FunctorType,Scalar>::minimizeOptimumStorage(
/* check the input parameters for errors. */
if (n <= 0 || m < n || ftol < 0. || xtol < 0. || gtol < 0. || maxfev <= 0 || factor <= 0.)
if (n <= 0 || m < n || parameters.ftol < 0. || parameters.xtol < 0. || parameters.gtol < 0. || parameters.maxfev <= 0 || parameters.factor <= 0.)
return RelativeErrorTooSmall;
if (mode == 2)
@@ -805,9 +804,9 @@ LevenbergMarquardt<FunctorType,Scalar>::minimizeOptimumStorage(
wa3 = diag.cwise() * x;
xnorm = wa3.stableNorm();
delta = factor * xnorm;
delta = parameters.factor * xnorm;
if (delta == 0.)
delta = factor;
delta = parameters.factor;
}
/* compute the norm of the scaled gradient. */
@@ -827,7 +826,7 @@ LevenbergMarquardt<FunctorType,Scalar>::minimizeOptimumStorage(
/* test for convergence of the gradient norm. */
if (gnorm <= gtol)
if (gnorm <= parameters.gtol)
return CosinusTooSmall;
/* rescale if necessary. */
@@ -921,16 +920,16 @@ LevenbergMarquardt<FunctorType,Scalar>::minimizeOptimumStorage(
/* tests for convergence. */
if (ei_abs(actred) <= ftol && prered <= ftol && Scalar(.5) * ratio <= 1. && delta <= xtol * xnorm)
if (ei_abs(actred) <= parameters.ftol && prered <= parameters.ftol && Scalar(.5) * ratio <= 1. && delta <= parameters.xtol * xnorm)
return RelativeErrorAndReductionTooSmall;
if (ei_abs(actred) <= ftol && prered <= ftol && Scalar(.5) * ratio <= 1.)
if (ei_abs(actred) <= parameters.ftol && prered <= parameters.ftol && Scalar(.5) * ratio <= 1.)
return RelativeReductionTooSmall;
if (delta <= xtol * xnorm)
if (delta <= parameters.xtol * xnorm)
return RelativeErrorTooSmall;
/* tests for termination and stringent tolerances. */
if (nfev >= maxfev)
if (nfev >= parameters.maxfev)
return TooManyFunctionEvaluation;
if (ei_abs(actred) <= epsilon<Scalar>() && prered <= epsilon<Scalar>() && Scalar(.5) * ratio <= 1.)
return FtolTooSmall;