| 1 | /* -*- mode: c++; tab-width: 4; indent-tabs-mode: nil; c-basic-offset: 4 -*- */ |
| 2 | |
| 3 | /* |
| 4 | Copyright (C) 2001, 2002, 2003 Nicolas Di Césaré |
| 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 <ql/math/optimization/armijo.hpp> |
| 21 | #include <ql/math/optimization/method.hpp> |
| 22 | #include <ql/math/optimization/problem.hpp> |
| 23 | |
| 24 | namespace QuantLib { |
| 25 | |
| 26 | Real ArmijoLineSearch::operator()(Problem& P, |
| 27 | EndCriteria::Type& ecType, |
| 28 | const EndCriteria& endCriteria, |
| 29 | const Real t_ini) |
| 30 | { |
| 31 | //OptimizationMethod& method = P.method(); |
| 32 | Constraint& constraint = P.constraint(); |
| 33 | succeed_=true; |
| 34 | bool maxIter = false; |
| 35 | Real qtold, t = t_ini; |
| 36 | Size loopNumber = 0; |
| 37 | |
| 38 | Real q0 = P.functionValue(); |
| 39 | Real qp0 = P.gradientNormValue(); |
| 40 | |
| 41 | qt_ = q0; |
| 42 | qpt_ = (gradient_.empty()) ? qp0 : -DotProduct(v1: gradient_,v2: searchDirection_); |
| 43 | |
| 44 | // Initialize gradient |
| 45 | gradient_ = Array(P.currentValue().size()); |
| 46 | // Compute new point |
| 47 | xtd_ = P.currentValue(); |
| 48 | t = update(params&: xtd_, direction: searchDirection_, beta: t, constraint); |
| 49 | // Compute function value at the new point |
| 50 | qt_ = P.value (x: xtd_); |
| 51 | |
| 52 | // Enter in the loop if the criterion is not satisfied |
| 53 | if ((qt_-q0) > -alpha_*t*qpt_) { |
| 54 | do { |
| 55 | loopNumber++; |
| 56 | // Decrease step |
| 57 | t *= beta_; |
| 58 | // Store old value of the function |
| 59 | qtold = qt_; |
| 60 | // New point value |
| 61 | xtd_ = P.currentValue(); |
| 62 | t = update(params&: xtd_, direction: searchDirection_, beta: t, constraint); |
| 63 | |
| 64 | // Compute function value at the new point |
| 65 | qt_ = P.value (x: xtd_); |
| 66 | P.gradient (grad_f&: gradient_, x: xtd_); |
| 67 | // and it squared norm |
| 68 | maxIter = endCriteria.checkMaxIterations(iteration: loopNumber, ecType); |
| 69 | } while ( |
| 70 | (((qt_ - q0) > (-alpha_ * t * qpt_)) || |
| 71 | ((qtold - q0) <= (-alpha_ * t * qpt_ / beta_))) && |
| 72 | (!maxIter)); |
| 73 | } |
| 74 | |
| 75 | if (maxIter) |
| 76 | succeed_ = false; |
| 77 | |
| 78 | // Compute new gradient |
| 79 | P.gradient(grad_f&: gradient_, x: xtd_); |
| 80 | // and it squared norm |
| 81 | qpt_ = DotProduct(v1: gradient_, v2: gradient_); |
| 82 | |
| 83 | // Return new step value |
| 84 | return t; |
| 85 | } |
| 86 | |
| 87 | } |
| 88 | |