Source code for mealpy.swarm_based.MSA

# !/usr/bin/env python
# Created by "Thieu" at 14:52, 17/03/2020 ----------%
#       Email: nguyenthieu2102@gmail.com            %
#       Github: https://github.com/thieu1995        %
# --------------------------------------------------%

import numpy as np
from math import gamma
from copy import deepcopy
from mealpy.optimizer import Optimizer


[docs]class BaseMSA(Optimizer): """ My changed version of: Moth Search Algorithm (MSA) Links: 1. https://www.mathworks.com/matlabcentral/fileexchange/59010-moth-search-ms-algorithm 2. https://doi.org/10.1007/s12293-016-0212-3 Notes ~~~~~ + The matlab version of original paper is not good (especially convergence chart) + I add Normal random number (Gaussian distribution) in each updating equation (Better performance) Hyper-parameters should fine tuned in approximate range to get faster convergen toward the global optimum: + n_best (int): [3, 10], how many of the best moths to keep from one generation to the next, default=5 + partition (float): [0.3, 0.8], The proportional of first partition, default=0.5 + max_step_size (float): [0.5, 2.0], Max step size used in Levy-flight technique, default=1.0 Examples ~~~~~~~~ >>> import numpy as np >>> from mealpy.swarm_based.MSA import BaseMSA >>> >>> def fitness_function(solution): >>> return np.sum(solution**2) >>> >>> problem_dict1 = { >>> "fit_func": fitness_function, >>> "lb": [-10, -15, -4, -2, -8], >>> "ub": [10, 15, 12, 8, 20], >>> "minmax": "min", >>> "verbose": True, >>> } >>> >>> epoch = 1000 >>> pop_size = 50 >>> n_best = 5 >>> partition = 0.5 >>> max_step_size = 1.0 >>> model = BaseMSA(problem_dict1, epoch, pop_size, n_best, partition, max_step_size) >>> best_position, best_fitness = model.solve() >>> print(f"Solution: {best_position}, Fitness: {best_fitness}") References ~~~~~~~~~~ [1] Wang, G.G., 2018. Moth search algorithm: a bio-inspired metaheuristic algorithm for global optimization problems. Memetic Computing, 10(2), pp.151-164. """ def __init__(self, problem, epoch=10000, pop_size=100, n_best=5, partition=0.5, max_step_size=1.0, **kwargs): """ Args: problem (dict): The problem dictionary epoch (int): maximum number of iterations, default = 10000 pop_size (int): number of population size, default = 100 n_best (int): how many of the best moths to keep from one generation to the next, default=5 partition (float): The proportional of first partition, default=0.5 max_step_size (float): Max step size used in Levy-flight technique, default=1.0 """ super().__init__(problem, kwargs) self.nfe_per_epoch = pop_size self.sort_flag = True self.epoch = epoch self.pop_size = pop_size self.n_best = n_best self.partition = partition self.max_step_size = max_step_size # np1 in paper self.n_moth1 = int(np.ceil(self.partition * self.pop_size)) # np2 in paper, we actually don't need this variable self.n_moth2 = self.pop_size - self.n_moth1 # you can change this ratio so as to get much better performance self.golden_ratio = (np.sqrt(5) - 1) / 2.0 def _levy_walk(self, iteration): beta = 1.5 # Eq. 2.23 sigma = (gamma(1 + beta) * np.sin(np.pi * (beta - 1) / 2) / (gamma(beta / 2) * (beta - 1) * 2 ** ((beta - 2) / 2))) ** (1 / (beta - 1)) u = np.random.uniform(self.problem.lb, self.problem.ub) * sigma v = np.random.uniform(self.problem.lb, self.problem.ub) step = u / np.abs(v) ** (1.0 / (beta - 1)) # Eq. 2.21 scale = self.max_step_size / (iteration + 1) delta_x = scale * step return delta_x
[docs] def evolve(self, epoch): """ The main operations (equations) of algorithm. Inherit from Optimizer class Args: epoch (int): The current iteration """ pop_best = deepcopy(self.pop[:self.n_best]) pop_new = [] for idx in range(0, self.pop_size): # Migration operator if idx < self.n_moth1: # scale = self.max_step_size / (epoch+1) # Smaller step for local walk pos_new = self.pop[idx][self.ID_POS] + np.random.normal() * self._levy_walk(epoch) else: # Flying in a straight line temp_case1 = self.pop[idx][self.ID_POS] + np.random.normal() * \ self.golden_ratio * (self.g_best[self.ID_POS] - self.pop[idx][self.ID_POS]) temp_case2 = self.pop[idx][self.ID_POS] + np.random.normal() * \ (1.0 / self.golden_ratio) * (self.g_best[self.ID_POS] - self.pop[idx][self.ID_POS]) pos_new = np.where(np.random.uniform(self.problem.n_dims) < 0.5, temp_case2, temp_case1) pos_new = self.amend_position(pos_new) pop_new.append([pos_new, None]) pop_new = self.update_fitness_population(pop_new) pop_new = self.greedy_selection_population(self.pop, pop_new) self.pop, _ = self.get_global_best_solution(pop_new) # Replace the worst with the previous generation's elites. for i in range(0, self.n_best): self.pop[-1 - i] = deepcopy(pop_best[i])