1
0
mirror of https://github.com/aportelli/LatAnalyze.git synced 2024-11-10 00:45:36 +00:00

numerical derivative class with arbitrary improvement order and step auto-tuning

This commit is contained in:
Antonin Portelli 2014-09-26 18:40:52 +01:00
parent 8ccc529776
commit 0fdd76b19d
6 changed files with 325 additions and 0 deletions

View File

@ -16,6 +16,7 @@ endif
noinst_PROGRAMS = \
exCompiledDoubleFunction\
exDerivative \
exFit \
exIntegrator \
exMat \
@ -29,6 +30,10 @@ exCompiledDoubleFunction_SOURCES = exCompiledDoubleFunction.cpp
exCompiledDoubleFunction_CFLAGS = -g -O2
exCompiledDoubleFunction_LDFLAGS = -L../lib/.libs -lLatAnalyze
exDerivative_SOURCES = exDerivative.cpp
exDerivative_CFLAGS = -g -O2
exDerivative_LDFLAGS = -L../lib/.libs -lLatAnalyze
exFit_SOURCES = exFit.cpp
exFit_CFLAGS = -g -O2
exFit_LDFLAGS = -L../lib/.libs -lLatAnalyze

40
examples/exDerivative.cpp Normal file
View File

@ -0,0 +1,40 @@
#include <iostream>
#include <LatAnalyze/Derivative.hpp>
#include <LatAnalyze/CompiledFunction.hpp>
#include <LatAnalyze/Math.hpp>
using namespace std;
using namespace Latan;
int main(int argc, char *argv[])
{
string source;
Index maxOrder;
double x;
if (argc != 4)
{
cerr << "usage: " << argv[0] << " <function> <max order> <point>";
cerr << endl;
return EXIT_FAILURE;
}
source = argv[1];
maxOrder = strTo<Index>(argv[2]);
x = strTo<double>(argv[3]);
CompiledDoubleFunction f(1, source);
CentralDerivative df(f);
for (Index i = 1; i <= 4; ++i)
{
cout << "--- O(h^" << 2*i << ") derivative" << endl;
for (Index j = 0; j <= maxOrder; ++j)
{
df.setOrder(j, i);
cout << "d^" << j << "f(" << x << ")= " << df(x) << endl;
}
}
return 0;
}

185
lib/Derivative.cpp Normal file
View File

@ -0,0 +1,185 @@
/*
* Derivative.cpp, part of LatAnalyze 3
*
* Copyright (C) 2013 - 2014 Antonin Portelli
*
* LatAnalyze 3 is free software: you can redistribute it and/or modify
* it under the terms of the GNU General Public License as published by
* the Free Software Foundation, either version 3 of the License, or
* (at your option) any later version.
*
* LatAnalyze 3 is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
* GNU General Public License for more details.
*
* You should have received a copy of the GNU General Public License
* along with LatAnalyze 3. If not, see <http://www.gnu.org/licenses/>.
*/
#include <LatAnalyze/Derivative.hpp>
#include <LatAnalyze/includes.hpp>
#include <LatAnalyze/Math.hpp>
#include <gsl/gsl_deriv.h>
using namespace std;
using namespace Latan;
using namespace Math;
/******************************************************************************
* Derivative implementation *
******************************************************************************/
// constructor /////////////////////////////////////////////////////////////////
Derivative::Derivative(const DoubleFunction &f, const Index dir,
const double step)
: DoubleFunction(f.getNArg())
, f_(f)
, dir_(dir)
, step_(step)
, buffer_(new DVec(f.getNArg()))
{}
Derivative::Derivative(const DoubleFunction &f, const Index dir,
const Index order, const DVec point, const double step)
: Derivative(f, dir, step)
{
setOrderAndPoint(order, point);
}
// access //////////////////////////////////////////////////////////////////////
Index Derivative::getOrder(void) const
{
return order_;
}
Index Derivative::getNPoint(void) const
{
return point_.size();
}
double Derivative::getStep(void) const
{
return step_;
}
void Derivative::setOrderAndPoint(const Index order, const DVec point)
{
if (order >= point.size())
{
LATAN_ERROR(Size, "derivative order is superior or equal to the number of point");
}
order_ = order;
point_ = point;
coefficient_.resize(point.size());
makeCoefficients();
}
void Derivative::setStep(const double step)
{
step_ = step;
}
// coefficient generation //////////////////////////////////////////////////////
// from B. Fornberg, “Generation of finite difference formulas on arbitrarily
// spaced grids,” Math. Comp., vol. 51, no. 184, pp. 699706, 1988.
// http://dx.doi.org/10.1090/S0025-5718-1988-0935077-0
void Derivative::makeCoefficients(void)
{
double c[3];
const Index N = point_.size() - 1, M = order_;
DMat curr(M + 1, N + 1), prev(M + 1, N + 1);
curr.fill(0.);
prev.fill(0.);
prev(0, 0) = 1.;
c[0] = 1.;
for (Index n = 1; n <= N; ++n)
{
c[1] = 1.;
for (Index nu = 0; nu <= n - 1; ++nu)
{
c[2] = point_(n) - point_(nu);
c[1] *= c[2];
for (Index m = 0; m <= min(n, M); ++m)
{
curr(m, nu) = point_(n)*prev(m, nu);
if (m)
{
curr(m, nu) -= m*prev(m-1, nu);
}
curr(m, nu) /= c[2];
}
}
for (Index m = 0; m <= min(n, M); ++m)
{
curr(m, n) = -point_(n-1)*prev(m, n-1);
if (m)
{
curr(m, n) += m*prev(m-1, n-1);
}
curr(m, n) *= c[0]/c[1];
}
c[0] = c[1];
prev = curr;
}
coefficient_ = curr.row(M);
}
// function call ///////////////////////////////////////////////////////////////
double Derivative::operator()(const double *x) const
{
ConstMap<DVec> xMap(x, f_.getNArg());
double res = 0.;
*buffer_ = xMap;
FOR_VEC(point_, i)
{
(*buffer_)(dir_) = x[dir_] + point_(i)*step_;
res += coefficient_[i]*f_(*buffer_);
}
res /= pow(step_, order_);
return res;
}
/******************************************************************************
* CentralDerivative implementation *
******************************************************************************/
// constructor /////////////////////////////////////////////////////////////////
CentralDerivative::CentralDerivative(const DoubleFunction &f, const Index dir,
const Index order, const Index precOrder)
: Derivative(f, dir)
{
setOrder(order, precOrder);
}
// access //////////////////////////////////////////////////////////////////////
Index CentralDerivative::getPrecOrder(void) const
{
return precOrder_;
}
void CentralDerivative::setOrder(const Index order, const Index precOrder)
{
const Index nPoint = 2*(precOrder + (order - 1)/2) + 1;
DVec point(nPoint);
precOrder_ = precOrder;
FOR_VEC(point, i)
{
point(i) = static_cast<double>(i - (nPoint - 1)/2);
}
setOrderAndPoint(order, point);
tuneStep();
}
// step tuning /////////////////////////////////////////////////////////////////
// the rounding error should be O(N*epsilon/h^order)
//
void CentralDerivative::tuneStep(void)
{
const Index nPoint = getNPoint();
const double epsilon = numeric_limits<double>::epsilon();
const double step = pow(epsilon*nPoint, 1./(2.*precOrder_+getOrder()));
setStep(step);
}

92
lib/Derivative.hpp Normal file
View File

@ -0,0 +1,92 @@
/*
* Derivative.hpp, part of LatAnalyze 3
*
* Copyright (C) 2013 - 2014 Antonin Portelli
*
* LatAnalyze 3 is free software: you can redistribute it and/or modify
* it under the terms of the GNU General Public License as published by
* the Free Software Foundation, either version 3 of the License, or
* (at your option) any later version.
*
* LatAnalyze 3 is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
* GNU General Public License for more details.
*
* You should have received a copy of the GNU General Public License
* along with LatAnalyze 3. If not, see <http://www.gnu.org/licenses/>.
*/
#ifndef Latan_Derivative_hpp_
#define Latan_Derivative_hpp_
#include <LatAnalyze/Global.hpp>
#include <LatAnalyze/Function.hpp>
BEGIN_NAMESPACE
/******************************************************************************
* Derivative *
******************************************************************************/
class Derivative: public DoubleFunction
{
public:
static constexpr double defaultStep = 1.0e-2;
public:
// constructor
Derivative(const DoubleFunction &f, const Index dir, const Index order,
const DVec point, const double step = defaultStep);
// destructor
virtual ~Derivative(void) = default;
// access
Index getNPoint(void) const;
Index getOrder(void) const;
double getStep(void) const;
void setOrderAndPoint(const Index order, const DVec point);
void setStep(const double step);
// function call
using DoubleFunction::operator();
virtual double operator()(const double *x) const;
protected:
// constructor
Derivative(const DoubleFunction &f, const Index dir,
const double step = defaultStep);
private:
void makeCoefficients(void);
private:
const DoubleFunction &f_;
Index dir_, order_;
double step_;
DVec point_, coefficient_;
std::shared_ptr<DVec> buffer_;
};
class CentralDerivative: private Derivative
{
public:
static const Index defaultPrecOrder = 2;
public:
// constructor
CentralDerivative(const DoubleFunction &f, const Index dir = 0,
const Index order = 1,
const Index precOrder = defaultPrecOrder);
// destructor
virtual ~CentralDerivative(void) = default;
// access
using Derivative::getNPoint;
using Derivative::getStep;
using Derivative::getOrder;
Index getPrecOrder(void) const;
void setOrder(const Index order, const Index precOrder = defaultPrecOrder);
// function call
using Derivative::operator();
private:
// step tuning
void tuneStep(void);
private:
Index precOrder_;
};
END_NAMESPACE
#endif // Latan_Derivative_hpp_

View File

@ -31,6 +31,7 @@ libLatAnalyze_la_SOURCES = \
Chi2Function.cpp \
CompiledFunction.cpp \
CompiledModel.cpp \
Derivative.cpp \
Exceptions.cpp \
File.cpp \
FitInterface.cpp \
@ -62,6 +63,7 @@ libLatAnalyze_la_HEADERS = \
CompiledFunction.hpp \
CompiledModel.hpp \
Dataset.hpp \
Derivative.hpp \
Exceptions.hpp \
FitInterface.hpp \
Function.hpp \

View File

@ -26,6 +26,7 @@
#include <iomanip>
#include <iostream>
#include <iterator>
#include <limits>
#include <sstream>
#include <utility>
#include <cfloat>