| 1 |
/* -*- mode: c++; tab-width: 4; indent-tabs-mode: nil; c-basic-offset: 4 -*- */ |
| 2 |
|
| 3 |
/* |
| 4 |
Copyright (C) 2009 Frédéric Degraeve |
| 5 |
|
| 6 |
This file is part of QuantLib, a free-software/open-source library |
| 7 |
for financial quantitative analysts and developers - http://quantlib.org/ |
| 8 |
|
| 9 |
QuantLib is free software: you can redistribute it and/or modify it |
| 10 |
under the terms of the QuantLib license. You should have received a |
| 11 |
copy of the license along with this program; if not, please email |
| 12 |
<quantlib-dev@lists.sf.net>. The license is also available online at |
| 13 |
<http://quantlib.org/license.shtml>. |
| 14 |
|
| 15 |
This program is distributed in the hope that it will be useful, but WITHOUT |
| 16 |
ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS |
| 17 |
FOR A PARTICULAR PURPOSE. See the license for more details. |
| 18 |
*/ |
| 19 |
|
| 20 |
#include "optimization/BFGS.hpp" |
| 21 |
#include "optimization/Problem.hpp" |
| 22 |
#include "optimization/LineSearch.hpp" |
| 23 |
|
| 24 |
namespace QuantLib { |
| 25 |
|
| 26 |
DynamicVector<RealType> BFGS::getUpdatedDirection(const Problem& P, |
| 27 |
RealType, |
| 28 |
const DynamicVector<RealType>& oldGradient) { |
| 29 |
if (inverseHessian_.getNRow() == 0) |
| 30 |
{ |
| 31 |
// first time in this update, we create needed structures |
| 32 |
inverseHessian_ = DynamicRectMatrix<RealType>(P.currentValue().size(), |
| 33 |
P.currentValue().size(), 0.); |
| 34 |
for (size_t i = 0; i < P.currentValue().size(); ++i) |
| 35 |
inverseHessian_(i,i) = 1.; |
| 36 |
} |
| 37 |
|
| 38 |
DynamicVector<RealType> diffGradient; |
| 39 |
DynamicVector<RealType> diffGradientWithHessianApplied(P.currentValue().size(), 0.); |
| 40 |
|
| 41 |
diffGradient = lineSearch_->lastGradient() - oldGradient; |
| 42 |
for (size_t i = 0; i < P.currentValue().size(); ++i) |
| 43 |
for (size_t j = 0; j < P.currentValue().size(); ++j) |
| 44 |
diffGradientWithHessianApplied[i] += inverseHessian_(i,j) * diffGradient[j]; |
| 45 |
|
| 46 |
double fac, fae, fad; |
| 47 |
double sumdg, sumxi; |
| 48 |
|
| 49 |
fac = fae = sumdg = sumxi = 0.; |
| 50 |
for (size_t i = 0; i < P.currentValue().size(); ++i) |
| 51 |
{ |
| 52 |
fac += diffGradient[i] * lineSearch_->searchDirection()[i]; |
| 53 |
fae += diffGradient[i] * diffGradientWithHessianApplied[i]; |
| 54 |
sumdg += std::pow(diffGradient[i], 2.); |
| 55 |
sumxi += std::pow(lineSearch_->searchDirection()[i], 2.); |
| 56 |
} |
| 57 |
|
| 58 |
if (fac > std::sqrt(1e-8 * sumdg * sumxi)) // skip update if fac not sufficiently positive |
| 59 |
{ |
| 60 |
fac = 1.0 / fac; |
| 61 |
fad = 1.0 / fae; |
| 62 |
|
| 63 |
for (size_t i = 0; i < P.currentValue().size(); ++i) |
| 64 |
diffGradient[i] = fac * lineSearch_->searchDirection()[i] - fad * diffGradientWithHessianApplied[i]; |
| 65 |
|
| 66 |
for (size_t i = 0; i < P.currentValue().size(); ++i) |
| 67 |
for (size_t j = 0; j < P.currentValue().size(); ++j) |
| 68 |
{ |
| 69 |
inverseHessian_(i,j) += fac * lineSearch_->searchDirection()[i] * lineSearch_->searchDirection()[j]; |
| 70 |
inverseHessian_(i,j) -= fad * diffGradientWithHessianApplied[i] * diffGradientWithHessianApplied[j]; |
| 71 |
inverseHessian_(i,j) += fae * diffGradient[i] * diffGradient[j]; |
| 72 |
} |
| 73 |
} |
| 74 |
//else |
| 75 |
// throw "BFGS: FAC not sufficiently positive"; |
| 76 |
|
| 77 |
|
| 78 |
DynamicVector<RealType> direction(P.currentValue().size()); |
| 79 |
for (size_t i = 0; i < P.currentValue().size(); ++i) |
| 80 |
{ |
| 81 |
direction[i] = 0.0; |
| 82 |
for (size_t j = 0; j < P.currentValue().size(); ++j) |
| 83 |
direction[i] -= inverseHessian_(i,j) * lineSearch_->lastGradient()[j]; |
| 84 |
} |
| 85 |
|
| 86 |
return direction; |
| 87 |
} |
| 88 |
|
| 89 |
} |
