# !/usr/bin/env python
# Created by "Thieu" at 09:17, 09/11/2020 ----------%
# Email: nguyenthieu2102@gmail.com %
# Github: https://github.com/thieu1995 %
# --------------------------------------------------%
import numpy as np
from scipy.spatial.distance import cdist
from copy import deepcopy
from mealpy.optimizer import Optimizer
[docs]class BaseBRO(Optimizer):
"""
My changed version of: Battle Royale Optimization (BRO)
Notes
~~~~~
I change the flow of algorithm
Hyper-parameters should fine tuned in approximate range to get faster convergen toward the global optimum:
+ threshold (int): [2, 5], dead threshold, default=3
Examples
~~~~~~~~
>>> import numpy as np
>>> from mealpy.human_based.BRO import BaseBRO
>>>
>>> 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
>>> threshold = 3
>>> model = BaseBRO(problem_dict1, epoch, pop_size, threshold)
>>> best_position, best_fitness = model.solve()
>>> print(f"Solution: {best_position}, Fitness: {best_fitness}")
"""
ID_DAM = 2
def __init__(self, problem, epoch=10000, pop_size=100, threshold=3, **kwargs):
"""
Args:
problem (dict): The problem dictionary
epoch (int): maximum number of iterations, default = 10000
pop_size (int): number of population size, default = 100
threshold (int): dead threshold, default=3
"""
super().__init__(problem, kwargs)
self.nfe_per_epoch = pop_size
self.sort_flag = False
self.epoch = epoch
self.pop_size = pop_size
self.threshold = threshold
## Dynamic variable
shrink = np.ceil(np.log10(self.epoch))
self.dyn_delta = round(self.epoch / shrink)
[docs] def create_solution(self):
"""
To get the position, fitness wrapper, target and obj list
+ A[self.ID_POS] --> Return: position
+ A[self.ID_TAR] --> Return: [target, [obj1, obj2, ...]]
+ A[self.ID_TAR][self.ID_FIT] --> Return: target
+ A[self.ID_TAR][self.ID_OBJ] --> Return: [obj1, obj2, ...]
Returns:
list: wrapper of solution with format [position, [target, [obj1, obj2, ...]], damage]
"""
position = np.random.uniform(self.problem.lb, self.problem.ub)
position = self.amend_position(position)
fitness = self.get_fitness_position(position=position)
damage = 0
return [position, fitness, damage]
def __get_min_idx(self, data):
k_zero = np.count_nonzero(data == 0)
if k_zero == len(data):
return np.random.choice(range(0, k_zero))
## 1st: Partition sorting, not good solution here.
# return np.argpartition(data, k_zero)[k_zero]
## 2nd: Faster
return np.where(data == np.min(data[data != 0]))[0][0]
[docs] def find_idx_min_distance(self, target_pos=None, pop=None):
list_pos = np.array([pop[idx][self.ID_POS] for idx in range(0, self.pop_size)])
target_pos = np.reshape(target_pos, (1, -1))
dist_list = cdist(list_pos, target_pos, 'euclidean')
dist_list = np.reshape(dist_list, (-1))
return self.__get_min_idx(dist_list)
[docs] def evolve(self, epoch):
"""
The main operations (equations) of algorithm. Inherit from Optimizer class
Args:
epoch (int): The current iteration
"""
nfe_epoch = 0
for i in range(self.pop_size):
# Compare ith soldier with nearest one (jth)
j = self.find_idx_min_distance(self.pop[i][self.ID_POS], self.pop)
if self.compare_agent(self.pop[i], self.pop[j]):
## Update Winner based on global best solution
pos_new = self.pop[i][self.ID_POS] + np.random.uniform() * \
np.mean(np.array([self.pop[i][self.ID_POS], self.g_best[self.ID_POS]]), axis=0)
pos_new = self.amend_position(pos_new)
fit_new = self.get_fitness_position(pos_new)
dam_new = self.pop[i][self.ID_DAM] - 1 ## Substract damaged hurt -1 to go next battle
self.pop[i] = [pos_new, fit_new, dam_new]
## Update Loser
if self.pop[j][self.ID_DAM] < self.threshold: ## If loser not dead yet, move it based on general
pos_new = np.random.uniform() * (np.maximum(self.pop[j][self.ID_POS], self.g_best[self.ID_POS]) -
np.minimum(self.pop[j][self.ID_POS], self.g_best[self.ID_POS])) + \
np.maximum(self.pop[j][self.ID_POS], self.g_best[self.ID_POS])
dam_new = self.pop[j][self.ID_DAM] + 1
self.pop[j][self.ID_TAR] = self.get_fitness_position(self.pop[j][self.ID_POS])
else: ## Loser dead and respawn again
pos_new = np.random.uniform(self.problem.lb, self.problem.ub)
dam_new = 0
pos_new = self.amend_position(pos_new)
fit_new = self.get_fitness_position(pos_new)
self.pop[j] = [pos_new, fit_new, dam_new]
nfe_epoch += 2
else:
## Update Loser by following position of Winner
self.pop[i] = deepcopy(self.pop[j])
## Update Winner by following position of General to protect the King and General
pos_new = self.pop[j][self.ID_POS] + np.random.uniform() * (self.g_best[self.ID_POS] - self.pop[j][self.ID_POS])
pos_new = self.amend_position(pos_new)
fit_new = self.get_fitness_position(pos_new)
dam_new = 0
self.pop[j] = [pos_new, fit_new, dam_new]
nfe_epoch += 1
self.nfe_per_epoch = nfe_epoch
if epoch >= self.dyn_delta: # max_epoch = 1000 -> delta = 300, 450, >500,....
pos_list = np.array([self.pop[idx][self.ID_POS] for idx in range(0, self.pop_size)])
pos_std = np.std(pos_list, axis=0)
lb = self.g_best[self.ID_POS] - pos_std
ub = self.g_best[self.ID_POS] + pos_std
self.problem.lb = np.clip(lb, self.problem.lb, self.problem.ub)
self.problem.ub = np.clip(ub, self.problem.lb, self.problem.ub)
self.dyn_delta += np.round(self.dyn_delta / 2)
[docs]class OriginalBRO(BaseBRO):
"""
The original version of: Battle Royale Optimization (BRO)
Links:
1. https://doi.org/10.1007/s00521-020-05004-4
Hyper-parameters should fine tuned in approximate range to get faster convergen toward the global optimum:
+ threshold (int): [2, 5], dead threshold, default=3
Examples
~~~~~~~~
>>> import numpy as np
>>> from mealpy.human_based.BRO import BaseBRO
>>>
>>> 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
>>> threshold = 3
>>> model = BaseBRO(problem_dict1, epoch, pop_size, threshold)
>>> best_position, best_fitness = model.solve()
>>> print(f"Solution: {best_position}, Fitness: {best_fitness}")
References
~~~~~~~~~~
[1] Rahkar Farshi, T., 2021. Battle royale optimization algorithm. Neural Computing and Applications, 33(4), pp.1139-1157.
"""
def __init__(self, problem, epoch=10000, pop_size=100, threshold=3, **kwargs):
"""
Args:
problem (dict): The problem dictionary
epoch (int): maximum number of iterations, default = 10000
pop_size (int): number of population size, default = 100
threshold (int): dead threshold, default=3
"""
super().__init__(problem, epoch, pop_size, threshold, **kwargs)
self.nfe_per_epoch = pop_size
self.sort_flag = False
[docs] def evolve(self, epoch):
"""
The main operations (equations) of algorithm. Inherit from Optimizer class
Args:
epoch (int): The current iteration
"""
for i in range(self.pop_size):
# Compare ith soldier with nearest one (jth)
j = self.find_idx_min_distance(self.pop[i][self.ID_POS], self.pop)
dam, vic = i, j ## This error in the algorithm's flow in the paper, But in the matlab code, he changed.
if self.compare_agent(self.pop[i], self.pop[j]):
dam, vic = j, i ## The mistake also here in the paper.
if self.pop[dam][self.ID_DAM] < self.threshold:
pos_new = np.random.uniform(0, 1, self.problem.n_dims) * \
(np.maximum(self.pop[dam][self.ID_POS], self.g_best[self.ID_POS]) -
np.minimum(self.pop[dam][self.ID_POS], self.g_best[self.ID_POS])) + \
np.maximum(self.pop[dam][self.ID_POS], self.g_best[self.ID_POS])
self.pop[dam][self.ID_POS] = self.amend_position(pos_new)
self.pop[dam][self.ID_TAR] = self.get_fitness_position(self.pop[dam][self.ID_POS])
self.pop[dam][self.ID_DAM] += 1
self.pop[vic][self.ID_DAM] = 0
else:
self.pop[dam] = self.create_solution()
if epoch >= self.dyn_delta:
pos_list = np.array([self.pop[idx][self.ID_POS] for idx in range(0, self.pop_size)])
pos_std = np.std(pos_list, axis=0)
lb = self.g_best[self.ID_POS] - pos_std
ub = self.g_best[self.ID_POS] + pos_std
self.problem.lb = np.clip(lb, self.problem.lb, self.problem.ub)
self.problem.ub = np.clip(ub, self.problem.lb, self.problem.ub)
self.dyn_delta += round(self.dyn_delta / 2)