# !/usr/bin/env python
# Created by "Thieu" at 22:07, 07/04/2020 ----------%
# Email: nguyenthieu2102@gmail.com %
# Github: https://github.com/thieu1995 %
# --------------------------------------------------%
import numpy as np
from copy import deepcopy
from mealpy.optimizer import Optimizer
[docs]class BaseFA(Optimizer):
"""
The original version of: Fireworks Algorithm (FA)
Links:
1. https://doi.org/10.1007/978-3-642-13495-1_44
Hyper-parameters should fine tuned in approximate range to get faster convergen toward the global optimum:
+ max_sparks (int): parameter controlling the total number of sparks generated by the pop_size fireworks, default=50
+ p_a (float): percent (const parameter), default=0.04
+ p_b (float): percent (const parameter), default=0.8
+ max_ea (int): maximum explosion amplitude, default=40
+ m_sparks (int): number of sparks generated in each explosion generation, default=5
Examples
~~~~~~~~
>>> import numpy as np
>>> from mealpy.swarm_based.FA import BaseFA
>>>
>>> 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
>>> max_sparks = 50
>>> p_a = 0.04
>>> p_b = 0.8
>>> max_ea = 40
>>> m_sparks = 5
>>> model = BaseFA(problem_dict1, epoch, pop_size, max_sparks, p_a, p_b, max_ea, m_sparks)
>>> best_position, best_fitness = model.solve()
>>> print(f"Solution: {best_position}, Fitness: {best_fitness}")
References
~~~~~~~~~~
[1] Tan, Y. and Zhu, Y., 2010, June. Fireworks algorithm for optimization. In International
conference in swarm intelligence (pp. 355-364). Springer, Berlin, Heidelberg.
"""
def __init__(self, problem, epoch=10000, pop_size=100,
max_sparks=50, p_a=0.04, p_b=0.8, max_ea=40, m_sparks=5, **kwargs):
"""
Args:
problem (dict): The problem dictionary
epoch (int): maximum number of iterations, default = 10000
pop_size (int): number of population size, default = 100
max_sparks (int): parameter controlling the total number of sparks generated by the pop_size fireworks, default=50
p_a (float): percent (const parameter), default=0.04
p_b (float): percent (const parameter), default=0.8
max_ea (int): maximum explosion amplitude, default=40
m_sparks (int): number of sparks generated in each explosion generation, default=5
"""
super().__init__(problem, kwargs)
self.nfe_per_epoch = pop_size
self.sort_flag = False
self.epoch = epoch
self.pop_size = pop_size
self.max_sparks = max_sparks
self.p_a = p_a
self.p_b = p_b
self.max_ea = max_ea
self.m_sparks = m_sparks
[docs] def evolve(self, epoch):
"""
The main operations (equations) of algorithm. Inherit from Optimizer class
Args:
epoch (int): The current iteration
"""
nfe_epoch = 0
fit_list = np.array([agent[self.ID_TAR][self.ID_FIT] for agent in self.pop])
fit_list = sorted(fit_list)
pop_new = []
for idx in range(0, self.pop_size):
si = self.max_sparks * (fit_list[-1] - self.pop[idx][self.ID_TAR][self.ID_FIT] + self.EPSILON) / \
(self.pop_size * fit_list[-1] - np.sum(fit_list) + self.EPSILON)
Ai = self.max_ea * (self.pop[idx][self.ID_TAR][self.ID_FIT] - fit_list[0] + self.EPSILON) / \
(np.sum(fit_list) - fit_list[0] + self.EPSILON)
if si < self.p_a * self.max_sparks:
si_ = int(round(self.p_a * self.max_sparks) + 1)
elif si > self.p_b * self.m_sparks:
si_ = int(round(self.p_b * self.max_sparks) + 1)
else:
si_ = int(round(si) + 1)
## Algorithm 1
pop_new = []
for j in range(0, si_):
pos_new = deepcopy(self.pop[idx][self.ID_POS])
list_idx = np.random.choice(range(0, self.problem.n_dims), round(np.random.uniform() * self.problem.n_dims), replace=False)
displacement = Ai * np.random.uniform(-1, 1)
pos_new[list_idx] = pos_new[list_idx] + displacement
pos_new = np.where(np.logical_or(pos_new < self.problem.lb, pos_new > self.problem.ub),
self.problem.lb + np.abs(pos_new) % (self.problem.ub - self.problem.lb), pos_new)
pos_new = self.amend_position(pos_new)
pop_new.append([pos_new, None])
nfe_epoch += 1
pop_new = self.update_fitness_population(pop_new)
for _ in range(0, self.m_sparks):
idx = np.random.randint(0, self.pop_size)
pos_new = deepcopy(self.pop[idx][self.ID_POS])
list_idx = np.random.choice(range(0, self.problem.n_dims), round(np.random.uniform() * self.problem.n_dims), replace=False)
pos_new[list_idx] = pos_new[list_idx] + np.random.normal(0, 1) # Gaussian
pos_new = np.where(np.logical_or(pos_new < self.problem.lb, pos_new > self.problem.ub), self.problem.lb + \
np.abs(pos_new) % (self.problem.ub - self.problem.lb), pos_new)
pos_new = self.amend_position(pos_new)
pop_new.append([pos_new, None])
nfe_epoch += 1
pop_new = self.update_fitness_population(pop_new)
## Update the global best
self.pop = self.get_sorted_strim_population(pop_new + self.pop, self.pop_size)