# !/usr/bin/env python
# Created by "Thieu" at 07:44, 08/04/2020 ----------%
# Email: nguyenthieu2102@gmail.com %
# Github: https://github.com/thieu1995 %
# --------------------------------------------------%
import numpy as np
from copy import deepcopy
from mealpy.optimizer import Optimizer
[docs]class ImprovedBSO(Optimizer):
"""
My improved version of: Brain Storm Optimization (BSO)
Notes
~~~~~
+ Remove some probability parameters, and some useless equations.
+ Add Levy-flight technique for more robust
Hyper-parameters should fine tuned in approximate range to get faster convergen toward the global optimum:
+ m_clusters (int): [3, 10], number of clusters (m in the paper)
+ p1 (float): 25% percent
+ p2 (float): 50% percent changed by its own (local search), 50% percent changed by outside (global search)
+ p3 (float): 75% percent develop the old idea, 25% invented new idea based on levy-flight
+ p4 (float): [0.4, 0.6], Need more weights on the centers instead of the random position
Examples
~~~~~~~~
>>> import numpy as np
>>> from mealpy.human_based.BSO import ImprovedBSO
>>>
>>> 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
>>> m_clusters = 5
>>> p1 = 0.25
>>> p2 = 0.5
>>> p3 = 0.75
>>> p4 = 0.6
>>> model = ImprovedBSO(problem_dict1, epoch, pop_size, m_clusters, p1, p2, p3, p4)
>>> best_position, best_fitness = model.solve()
>>> print(f"Solution: {best_position}, Fitness: {best_fitness}")
"""
def __init__(self, problem, epoch=10000, pop_size=100,
m_clusters=5, p1=0.25, p2=0.5, p3=0.75, p4=0.5, **kwargs):
"""
Args:
problem (dict): The problem dictionary
epoch (int): maximum number of iterations, default = 10000
pop_size (int): number of population size, default = 100
m_clusters (int): number of clusters (m in the paper)
p1 (float): 25% percent
p2 (float): 50% percent changed by its own (local search), 50% percent changed by outside (global search)
p3 (float): 75% percent develop the old idea, 25% invented new idea based on levy-flight
p4 (float): Need more weights on the centers instead of the random position
"""
super().__init__(problem, kwargs)
self.nfe_per_epoch = pop_size
self.sort_flag = False
self.epoch = epoch
self.pop_size = pop_size
self.m_clusters = m_clusters
self.p1 = p1
self.p2 = p2
self.p3 = p3
self.p4 = p4
self.m_solution = int(self.pop_size / self.m_clusters)
self.pop_group, self.centers = None, None
def _find_cluster(self, pop_group):
centers = []
for i in range(0, self.m_clusters):
_, local_best = self.get_global_best_solution(pop_group[i])
centers.append(deepcopy(local_best))
return centers
def _make_group(self, pop):
pop_group = []
for idx in range(0, self.m_clusters):
pop_group.append(deepcopy(pop[idx * self.m_solution:(idx + 1) * self.m_solution]))
return pop_group
[docs] def initialization(self):
self.pop = self.create_population(self.pop_size)
self.pop_group = self._make_group(self.pop)
self.centers = self._find_cluster(self.pop_group)
_, self.g_best = self.get_global_best_solution(self.pop)
[docs] def evolve(self, epoch):
"""
The main operations (equations) of algorithm. Inherit from Optimizer class
Args:
epoch (int): The current iteration
"""
epxilon = 1 - 1 * (epoch + 1) / self.epoch # 1. Changed here, no need: k
if np.random.uniform() < self.p1: # p_5a
idx = np.random.randint(0, self.m_clusters)
solution_new = self.create_solution()
self.centers[idx] = solution_new
pop_group = deepcopy(self.pop_group)
for i in range(0, self.pop_size): # Generate new individuals
cluster_id = int(i / self.m_solution)
location_id = int(i % self.m_solution)
if np.random.uniform() < self.p2: # p_6b
if np.random.uniform() < self.p3:
pos_new = self.centers[cluster_id][self.ID_POS] + epxilon * np.random.uniform()
else: # 2. Using levy flight here
levy_step = self.get_levy_flight_step(beta=1.0, multiplier=0.001, case=-1)
pos_new = self.pop_group[cluster_id][location_id][self.ID_POS] + np.random.normal(0, 1, self.problem.n_dims) * levy_step
else:
id1, id2 = np.random.choice(range(0, self.m_clusters), 2, replace=False)
if np.random.uniform() < self.p4:
pos_new = 0.5 * (self.centers[id1][self.ID_POS] + self.centers[id2][self.ID_POS]) + epxilon * np.random.uniform()
else:
rand_id1 = np.random.randint(0, self.m_solution)
rand_id2 = np.random.randint(0, self.m_solution)
pos_new = 0.5 * (self.pop_group[id1][rand_id1][self.ID_POS] + self.pop_group[id2][rand_id2][self.ID_POS]) + \
epxilon * np.random.uniform()
pos_new = self.amend_position(pos_new)
pop_group[cluster_id][location_id] = [pos_new, None]
pop_group = [self.update_fitness_population(group) for group in pop_group]
for idx in range(0, self.m_clusters):
self.pop_group[idx] = self.greedy_selection_population(self.pop_group[idx], pop_group[idx])
# Needed to update the centers and population
self.centers = self._find_cluster(self.pop_group)
self.pop = []
for idx in range(0, self.m_clusters):
self.pop += self.pop_group[idx]
[docs]class BaseBSO(ImprovedBSO):
"""
The original version of: Brain Storm Optimization (BSO)
Links:
1. https://doi.org/10.1007/978-3-642-21515-5_36
Hyper-parameters should fine tuned in approximate range to get faster convergen toward the global optimum:
+ m_clusters (int): [3, 10], number of clusters (m in the paper)
+ p1 (float): [0.1, 0.5], probability
+ p2 (float): [0.5, 0.95], probability
+ p3 (float): [0.2, 0.8], probability
+ p4 (float): [0.2, 0.8], probability
+ slope (int): [10, 15, 20, 25], changing logsig() function's slope (k: in the paper)
+ miu (float): [0], mean of normal distribution (gaussian)
+ xichma (float): [1], standard deviation of normal distribution (gaussian)
Examples
~~~~~~~~
>>> import numpy as np
>>> from mealpy.human_based.BSO import BaseBSO
>>>
>>> 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
>>> m_clusters = 5
>>> p1 = 0.2
>>> p2 = 0.8
>>> p3 = 0.4
>>> p4 = 0.5
>>> slope = 20
>>> miu = 0
>>> xichma = 1
>>> model = BaseBSO(problem_dict1, epoch, pop_size, m_clusters, p1, p2, p3, p4, slope, miu, xichma)
>>> best_position, best_fitness = model.solve()
>>> print(f"Solution: {best_position}, Fitness: {best_fitness}")
References
~~~~~~~~~~
[1] Shi, Y., 2011, June. Brain storm optimization algorithm. In International
conference in swarm intelligence (pp. 303-309). Springer, Berlin, Heidelberg.
"""
def __init__(self, problem, epoch=10000, pop_size=100,
m_clusters=5, p1=0.2, p2=0.8, p3=0.4, p4=0.5, slope=20, miu=0, xichma=1, **kwargs):
"""
Args:
problem (dict): The problem dictionary
epoch (int): maximum number of iterations, default = 10000
pop_size (int): number of population size, default = 100
m_clusters (int): number of clusters (m in the paper)
p1 (float): probability
p2 (float): probability
p3 (float): probability
p4 (float): probability
slope (int): changing logsig() function's slope (k: in the paper)
miu (float): mean of normal distribution (gaussian)
xichma (float): standard deviation of normal distribution (gaussian)
"""
super().__init__(problem, epoch, pop_size, m_clusters, p1, p2, p3, p4, **kwargs)
self.slope = slope
self.miu = miu
self.xichma = xichma
[docs] def amend_position(self, position=None):
"""
If solution out of bound at dimension x, then it will re-arrange to random location in the range of domain
Args:
position: vector position (location) of the solution.
Returns:
Amended position
"""
return np.where(np.logical_and(self.problem.lb <= position, position <= self.problem.ub),
position, np.random.uniform(self.problem.lb, self.problem.ub))
[docs] def evolve(self, epoch):
"""
The main operations (equations) of algorithm. Inherit from Optimizer class
Args:
epoch (int): The current iteration
"""
x = (0.5 * self.epoch - (epoch + 1)) / self.slope
epxilon = np.random.uniform() * (1 / (1 + np.exp(-x)))
if np.random.rand() < self.p1: # p_5a
idx = np.random.randint(0, self.m_clusters)
solution_new = self.create_solution()
self.centers[idx] = solution_new
pop_group = deepcopy(self.pop_group)
for i in range(0, self.pop_size): # Generate new individuals
cluster_id = int(i / self.m_solution)
location_id = int(i % self.m_solution)
if np.random.uniform() < self.p2: # p_6b
if np.random.uniform() < self.p3: # p_6i
cluster_id = np.random.randint(0, self.m_clusters)
if np.random.uniform() < self.p3:
pos_new = self.centers[cluster_id][self.ID_POS] + epxilon * np.random.normal(self.miu, self.xichma)
else:
rand_idx = np.random.randint(0, self.m_solution)
pos_new = self.pop_group[cluster_id][rand_idx][self.ID_POS] + np.random.uniform()
else:
id1, id2 = np.random.choice(range(0, self.m_clusters), 2, replace=False)
if np.random.uniform() < self.p4:
pos_new = 0.5 * (self.centers[id1][self.ID_POS] + self.centers[id2][self.ID_POS]) + \
epxilon * np.random.normal(self.miu, self.xichma)
else:
rand_id1 = np.random.randint(0, self.m_solution)
rand_id2 = np.random.randint(0, self.m_solution)
pos_new = 0.5 * (self.pop_group[id1][rand_id1][self.ID_POS] + self.pop_group[id2][rand_id2][self.ID_POS]) + \
epxilon * np.random.normal(self.miu, self.xichma)
pos_new = self.amend_position(pos_new)
pop_group[cluster_id][location_id] = [pos_new, None]
pop_group = [self.update_fitness_population(group) for group in pop_group]
for idx in range(0, self.m_clusters):
self.pop_group[idx] = self.greedy_selection_population(self.pop_group[idx], pop_group[idx])
# Needed to update the centers and population
self.centers = self._find_cluster(self.pop_group)
self.pop = []
for idx in range(0, self.m_clusters):
self.pop += self.pop_group[idx]