| 1 | /* -*- mode: c++; tab-width: 4; indent-tabs-mode: nil; c-basic-offset: 4 -*- */ |
| 2 | |
| 3 | /* |
| 4 | Copyright (C) 2015 Andres Hernandez |
| 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 | /*! \file fireflyalgorithm.hpp |
| 21 | \brief Implementation based on: |
| 22 | Yang, Xin-She (2009) Firefly Algorithm, Levy Flights and Global |
| 23 | Optimization. Research and Development in Intelligent Systems XXVI, pp 209-218. |
| 24 | http://arxiv.org/pdf/1003.1464.pdf |
| 25 | */ |
| 26 | |
| 27 | #ifndef quantlib_optimization_fireflyalgorithm_hpp |
| 28 | #define quantlib_optimization_fireflyalgorithm_hpp |
| 29 | |
| 30 | #include <ql/math/optimization/problem.hpp> |
| 31 | #include <ql/math/optimization/constraint.hpp> |
| 32 | #include <ql/experimental/math/isotropicrandomwalk.hpp> |
| 33 | #include <ql/experimental/math/levyflightdistribution.hpp> |
| 34 | #include <ql/math/randomnumbers/mt19937uniformrng.hpp> |
| 35 | #include <ql/math/randomnumbers/seedgenerator.hpp> |
| 36 | |
| 37 | #include <cmath> |
| 38 | #include <random> |
| 39 | |
| 40 | namespace QuantLib { |
| 41 | |
| 42 | /*! The main process is as follows: |
| 43 | M individuals are used to explore the N-dimensional parameter space: |
| 44 | \f$ X_{i}^k = (X_{i, 1}^k, X_{i, 2}^k, \ldots, X_{i, N}^k) \f$ is the kth-iteration |
| 45 | for the ith-individual. X is updated via the rule |
| 46 | \f[ |
| 47 | X_{i, j}^{k+1} = X_{i, j}^k + I(X^k)_{i,j} + RandomWalk_{i,j}^k |
| 48 | \f] |
| 49 | |
| 50 | The intensity function I(X) should be monotonic |
| 51 | The optimization stops either because the number of iterations has been reached |
| 52 | or because the stationary function value limit has been reached. |
| 53 | |
| 54 | The current implementation extends the normal Firefly Algorithm with a |
| 55 | differential evolution (DE) optimizer according to: |
| 56 | Afnizanfaizal Abdullah, et al. "A New Hybrid Firefly Algorithm for Complex and |
| 57 | Nonlinear Problem". Volume 151 of the series Advances in Intelligent and Soft |
| 58 | Computing pp 673-680, 2012. |
| 59 | http://link.springer.com/chapter/10.1007%2F978-3-642-28765-7_81 |
| 60 | |
| 61 | In effect this implementation provides a fully fledged DE global optimizer |
| 62 | as well. The Firefly Algorithm was easy to combine with DE because it already |
| 63 | contained a step where the current solutions are sorted. The population is |
| 64 | then divided into two subpopulations based on their order. The subpopulation |
| 65 | with the best results are updated via the firefly algorithm. The worse |
| 66 | subpopulation is updated via the DE operator: |
| 67 | \f[ |
| 68 | Y^{k+1} = X_{best}^k + F(X_{r1}^k - X_{r2}^k) |
| 69 | \f] |
| 70 | and |
| 71 | \f[ |
| 72 | X_{i,j}^{k+1} = Y_{i,j}^{k+1}\ \text{if} R_{i,j} <= C |
| 73 | \f] |
| 74 | \f[ |
| 75 | X_{i,j}^{k+1} = X_{i,j}^{k+1}\ \text{otherwise} |
| 76 | \f] |
| 77 | where C is the crossover constant, and R is a random uniformly distributed |
| 78 | number. |
| 79 | */ |
| 80 | class FireflyAlgorithm : public OptimizationMethod { |
| 81 | public: |
| 82 | class RandomWalk; |
| 83 | class Intensity; |
| 84 | FireflyAlgorithm(Size M, |
| 85 | ext::shared_ptr<Intensity> intensity, |
| 86 | ext::shared_ptr<RandomWalk> randomWalk, |
| 87 | Size Mde = 0, |
| 88 | Real mutationFactor = 1.0, |
| 89 | Real crossoverFactor = 0.5, |
| 90 | unsigned long seed = SeedGenerator::instance().get()); |
| 91 | void startState(Problem &P, const EndCriteria &endCriteria); |
| 92 | EndCriteria::Type minimize(Problem& P, const EndCriteria& endCriteria) override; |
| 93 | |
| 94 | protected: |
| 95 | std::vector<Array> x_, xI_, xRW_; |
| 96 | std::vector<std::pair<Real, Size> > values_; |
| 97 | Array lX_, uX_; |
| 98 | Real mutation_, crossover_; |
| 99 | Size M_, N_, Mde_, Mfa_; |
| 100 | ext::shared_ptr<Intensity> intensity_; |
| 101 | ext::shared_ptr<RandomWalk> randomWalk_; |
| 102 | std::mt19937 generator_; |
| 103 | std::uniform_int_distribution<QuantLib::Size> distribution_; |
| 104 | MersenneTwisterUniformRng rng_; |
| 105 | }; |
| 106 | |
| 107 | //! Base intensity class |
| 108 | /*! Derived classes need to implement only intensityImpl |
| 109 | */ |
| 110 | class FireflyAlgorithm::Intensity { |
| 111 | friend class FireflyAlgorithm; |
| 112 | public: |
| 113 | virtual ~Intensity() = default; |
| 114 | //! find brightest firefly for each firefly |
| 115 | void findBrightest(); |
| 116 | protected: |
| 117 | Size Mfa_, N_; |
| 118 | const std::vector<Array> *x_; |
| 119 | const std::vector<std::pair<Real, Size> > *values_; |
| 120 | std::vector<Array> *xI_; |
| 121 | |
| 122 | virtual Real intensityImpl(Real valueX, Real valueY, Real distance) = 0; |
| 123 | inline Real distance(const Array& x, const Array& y) const { |
| 124 | Real d = 0.0; |
| 125 | for (Size i = 0; i < N_; i++) { |
| 126 | Real diff = x[i] - y[i]; |
| 127 | d += diff*diff; |
| 128 | } |
| 129 | return d; |
| 130 | } |
| 131 | |
| 132 | private: |
| 133 | void init(FireflyAlgorithm *fa) { |
| 134 | x_ = &fa->x_; |
| 135 | xI_ = &fa->xI_; |
| 136 | values_ = &fa->values_; |
| 137 | Mfa_ = fa->Mfa_; |
| 138 | N_ = fa->N_; |
| 139 | } |
| 140 | }; |
| 141 | |
| 142 | //! Exponential Intensity |
| 143 | /* Exponentially decreasing intensity |
| 144 | */ |
| 145 | class ExponentialIntensity : public FireflyAlgorithm::Intensity { |
| 146 | public: |
| 147 | ExponentialIntensity(Real beta0, Real betaMin, Real gamma) |
| 148 | : beta0_(beta0), betaMin_(betaMin), gamma_(gamma) {} |
| 149 | protected: |
| 150 | Real intensityImpl(Real valueX, Real valueY, Real d) override { |
| 151 | return (beta0_ - betaMin_) * std::exp(x: -gamma_ * d) + betaMin_; |
| 152 | } |
| 153 | Real beta0_, betaMin_, gamma_; |
| 154 | }; |
| 155 | |
| 156 | //! Inverse Square Intensity |
| 157 | /* Inverse law square |
| 158 | */ |
| 159 | class InverseLawSquareIntensity : public FireflyAlgorithm::Intensity { |
| 160 | public: |
| 161 | InverseLawSquareIntensity(Real beta0, Real betaMin) |
| 162 | : beta0_(beta0), betaMin_(betaMin) {} |
| 163 | protected: |
| 164 | Real intensityImpl(Real valueX, Real valueY, Real d) override { |
| 165 | return (beta0_ - betaMin_) / (d + QL_EPSILON) + betaMin_; |
| 166 | } |
| 167 | Real beta0_, betaMin_; |
| 168 | }; |
| 169 | |
| 170 | //! Base Random Walk class |
| 171 | /*! Derived classes need to implement only walkImpl |
| 172 | */ |
| 173 | class FireflyAlgorithm::RandomWalk { |
| 174 | friend class FireflyAlgorithm; |
| 175 | public: |
| 176 | virtual ~RandomWalk() = default; |
| 177 | //! perform random walk |
| 178 | void walk() { |
| 179 | for (Size i = 0; i < Mfa_; i++) { |
| 180 | walkImpl(xRW&: (*xRW_)[(*values_)[i].second]); |
| 181 | } |
| 182 | } |
| 183 | protected: |
| 184 | Size Mfa_, N_; |
| 185 | const std::vector<Array> *x_; |
| 186 | const std::vector<std::pair<Real, Size> > *values_; |
| 187 | std::vector<Array> *xRW_; |
| 188 | Array *lX_, *uX_; |
| 189 | |
| 190 | virtual void walkImpl(Array & xRW) = 0; |
| 191 | virtual void init(FireflyAlgorithm *fa) { |
| 192 | x_ = &fa->x_; |
| 193 | xRW_ = &fa->xRW_; |
| 194 | values_ = &fa->values_; |
| 195 | Mfa_ = fa->Mfa_; |
| 196 | N_ = fa->N_; |
| 197 | lX_ = &fa->lX_; |
| 198 | uX_ = &fa->uX_; |
| 199 | } |
| 200 | }; |
| 201 | |
| 202 | //! Distribution Walk |
| 203 | /* Random walk given by distribution template parameter. The |
| 204 | distribution must be compatible with std::mt19937. |
| 205 | */ |
| 206 | template <class Distribution> |
| 207 | class DistributionRandomWalk : public FireflyAlgorithm::RandomWalk { |
| 208 | public: |
| 209 | explicit DistributionRandomWalk(Distribution dist, |
| 210 | Real delta = 0.9, |
| 211 | unsigned long seed = SeedGenerator::instance().get()) |
| 212 | : walkRandom_(std::mt19937(seed), std::move(dist), 1, Array(1, 1.0), seed), |
| 213 | delta_(delta) {} |
| 214 | protected: |
| 215 | void walkImpl(Array& xRW) override { |
| 216 | walkRandom_.nextReal(&xRW[0]); |
| 217 | xRW *= delta_; |
| 218 | } |
| 219 | void init(FireflyAlgorithm* fa) override { |
| 220 | FireflyAlgorithm::RandomWalk::init(fa); |
| 221 | walkRandom_.setDimension(N_, *lX_, *uX_); |
| 222 | } |
| 223 | IsotropicRandomWalk<Distribution, std::mt19937> walkRandom_; |
| 224 | Real delta_; |
| 225 | }; |
| 226 | |
| 227 | //! Gaussian Walk |
| 228 | /* Gaussian random walk |
| 229 | */ |
| 230 | class GaussianWalk : public DistributionRandomWalk<std::normal_distribution<QuantLib::Real>> { |
| 231 | public: |
| 232 | explicit GaussianWalk(Real sigma, |
| 233 | Real delta = 0.9, |
| 234 | unsigned long seed = SeedGenerator::instance().get()) |
| 235 | : DistributionRandomWalk<std::normal_distribution<QuantLib::Real>>( |
| 236 | std::normal_distribution<QuantLib::Real>(0.0, sigma), delta, seed){} |
| 237 | }; |
| 238 | |
| 239 | //! Levy Flight Random Walk |
| 240 | /* Levy flight random walk |
| 241 | */ |
| 242 | class LevyFlightWalk : public DistributionRandomWalk<LevyFlightDistribution> { |
| 243 | public: |
| 244 | explicit LevyFlightWalk(Real alpha, |
| 245 | Real xm = 0.5, |
| 246 | Real delta = 0.9, |
| 247 | unsigned long seed = SeedGenerator::instance().get()) |
| 248 | : DistributionRandomWalk<LevyFlightDistribution>( |
| 249 | LevyFlightDistribution(xm, alpha), delta, seed) {} |
| 250 | }; |
| 251 | |
| 252 | //! Decreasing Random Walk |
| 253 | /* Gaussian random walk, but size of step decreases with each iteration step |
| 254 | */ |
| 255 | class DecreasingGaussianWalk : public GaussianWalk { |
| 256 | public: |
| 257 | explicit DecreasingGaussianWalk( |
| 258 | Real sigma, |
| 259 | Real delta = 0.9, |
| 260 | unsigned long seed = SeedGenerator::instance().get()) |
| 261 | : GaussianWalk(sigma, delta, seed), delta0_(delta) {} |
| 262 | protected: |
| 263 | void walkImpl(Array& xRW) override { |
| 264 | iteration_++; |
| 265 | if (iteration_ > Mfa_) { |
| 266 | //Every time all the fireflies have been processed |
| 267 | //multiply delta by itself |
| 268 | iteration_ = 0; |
| 269 | delta_ *= delta_; |
| 270 | } |
| 271 | GaussianWalk::walkImpl(xRW); |
| 272 | } |
| 273 | void init(FireflyAlgorithm* fa) override { |
| 274 | GaussianWalk::init(fa); |
| 275 | iteration_ = 0; |
| 276 | delta_ = delta0_; |
| 277 | } |
| 278 | |
| 279 | private: |
| 280 | Real delta0_; |
| 281 | Size iteration_; |
| 282 | }; |
| 283 | } |
| 284 | |
| 285 | #endif |
| 286 | |