#!/usr/bin/env python
# Created by "Thieu" at 14:07, 02/03/2021 ----------%
# Email: nguyenthieu2102@gmail.com %
# Github: https://github.com/thieu1995 %
# --------------------------------------------------%
import numpy as np
from copy import deepcopy
from mealpy.optimizer import Optimizer
[docs]class BaseICA(Optimizer):
"""
The original version of: Imperialist Competitive Algorithm (ICA)
Links:
1. https://ieeexplore.ieee.org/document/4425083
Hyper-parameters should fine tuned in approximate range to get faster convergen toward the global optimum:
+ empire_count (int): [3, 10], Number of Empires (also Imperialists)
+ assimilation_coeff (float): [1.0, 3.0], Assimilation Coefficient (beta in the paper)
+ revolution_prob (float): [0.01, 0.1], Revolution Probability
+ revolution_rate (float): [0.05, 0.2], Revolution Rate (mu)
+ revolution_step_size (float): [0.05, 0.2], Revolution Step Size (sigma)
+ zeta (float): [0.05, 0.2], Colonies Coefficient in Total Objective Value of Empires
Examples
~~~~~~~~
>>> import numpy as np
>>> from mealpy.human_based.ICA import BaseICA
>>>
>>> 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
>>> empire_count = 5
>>> assimilation_coeff = 1.5
>>> revolution_prob = 0.05
>>> revolution_rate = 0.1
>>> revolution_step_size = 0.1
>>> zeta = 0.1
>>> model = BaseICA(problem_dict1, epoch, pop_size, empire_count, assimilation_coeff, revolution_prob, revolution_rate, revolution_step_size, zeta)
>>> best_position, best_fitness = model.solve()
>>> print(f"Solution: {best_position}, Fitness: {best_fitness}")
References
~~~~~~~~~~
[1] Atashpaz-Gargari, E. and Lucas, C., 2007, September. Imperialist competitive algorithm: an algorithm for
optimization inspired by imperialistic competition. In 2007 IEEE congress on evolutionary computation (pp. 4661-4667). Ieee.
"""
def __init__(self, problem, epoch=10000, pop_size=100, empire_count=5, assimilation_coeff=1.5,
revolution_prob=0.05, revolution_rate=0.1, revolution_step_size=0.1, zeta=0.1, **kwargs):
"""
Args:
problem (dict): The problem dictionary
epoch (int): maximum number of iterations, default = 10000
pop_size (int): number of population size (n: pop_size, m: clusters), default = 100
empire_count (int): Number of Empires (also Imperialists)
assimilation_coeff (float): Assimilation Coefficient (beta in the paper)
revolution_prob (float): Revolution Probability
revolution_rate (float): Revolution Rate (mu)
revolution_step_size (float): Revolution Step Size (sigma)
zeta (float): Colonies Coefficient in Total Objective Value of Empires
"""
super().__init__(problem, kwargs)
self.nfe_per_epoch = pop_size
self.sort_flag = True
self.epoch = epoch
self.pop_size = pop_size
self.empire_count = empire_count
self.assimilation_coeff = assimilation_coeff
self.revolution_prob = revolution_prob
self.revolution_rate = revolution_rate
self.revolution_step_size = revolution_step_size
self.zeta = zeta
self.pop_empires, self.pop_colonies, self.empires = None, None, None
self.n_revoluted_variables, self.idx_list_variables = None, None
[docs] def revolution_country(self, position, idx_list_variables, n_revoluted):
pos_new = position + self.revolution_step_size * np.random.normal(0, 1, self.problem.n_dims)
idx_list = np.random.choice(idx_list_variables, n_revoluted, replace=False)
position[idx_list] = pos_new[idx_list] # Change only those selected index
return position
[docs] def initialization(self):
pop = self.create_population(self.pop_size)
self.pop, self.g_best = self.get_global_best_solution(pop)
# Initialization
self.n_revoluted_variables = int(round(self.revolution_rate * self.problem.n_dims))
self.idx_list_variables = list(range(0, self.problem.n_dims))
# pop = Empires
colony_count = self.pop_size - self.empire_count
self.pop_empires = deepcopy(self.pop[:self.empire_count])
self.pop_colonies = deepcopy(self.pop[self.empire_count:])
cost_empires_list = np.array([solution[self.ID_TAR][self.ID_FIT] for solution in self.pop_empires])
cost_empires_list_normalized = cost_empires_list - (np.max(cost_empires_list) + np.min(cost_empires_list))
prob_empires_list = np.abs(cost_empires_list_normalized / np.sum(cost_empires_list_normalized))
# Randomly choose colonies to empires
self.empires = {}
idx_already_selected = []
for i in range(0, self.empire_count - 1):
self.empires[i] = []
n_colonies = int(round(prob_empires_list[i] * colony_count))
idx_list = np.random.choice(list(set(range(0, colony_count)) - set(idx_already_selected)), n_colonies, replace=False).tolist()
idx_already_selected += idx_list
for idx in idx_list:
self.empires[i].append(self.pop_colonies[idx])
idx_last = list(set(range(0, colony_count)) - set(idx_already_selected))
self.empires[self.empire_count - 1] = []
for idx in idx_last:
self.empires[self.empire_count - 1].append(self.pop_colonies[idx])
[docs] def evolve(self, epoch):
"""
The main operations (equations) of algorithm. Inherit from Optimizer class
Args:
epoch (int): The current iteration
"""
# Assimilation
for idx, colonies in self.empires.items():
for idx_colony, colony in enumerate(colonies):
pos_new = colony[self.ID_POS] + self.assimilation_coeff * \
np.random.uniform(0, 1, self.problem.n_dims) * (self.pop_empires[idx][self.ID_POS] - colony[self.ID_POS])
pos_new = self.amend_position(pos_new)
self.empires[idx][idx_colony][self.ID_POS] = pos_new
self.empires[idx] = self.update_fitness_population(self.empires[idx])
# empires[idx], g_best = self.update_global_best_solution(empires[idx], self.ID_MIN_PROB, g_best)
# Revolution
for idx, colonies in self.empires.items():
# Apply revolution to Imperialist
pos_new = self.revolution_country(self.pop_empires[idx][self.ID_POS], self.idx_list_variables, self.n_revoluted_variables)
self.pop_empires[idx][self.ID_POS] = self.amend_position(pos_new)
# Apply revolution to Colonies
for idx_colony, colony in enumerate(colonies):
if np.random.rand() < self.revolution_prob:
pos_new = self.revolution_country(colony[self.ID_POS], self.idx_list_variables, self.n_revoluted_variables)
self.empires[idx][idx_colony][self.ID_POS] = self.amend_position(pos_new)
self.empires[idx] = self.update_fitness_population(self.empires[idx])
self.pop_empires = self.update_fitness_population(self.pop_empires)
_, g_best = self.update_global_best_solution(self.pop_empires)
# Intra-Empire Competition
for idx, colonies in self.empires.items():
for idx_colony, colony in enumerate(colonies):
if self.compare_agent(colony, self.pop_empires[idx]):
self.empires[idx][idx_colony], self.pop_empires[idx] = deepcopy(self.pop_empires[idx]), deepcopy(colony)
# Update Total Objective Values of Empires
cost_empires_list = []
for idx, colonies in self.empires.items():
fit_list = np.array([solution[self.ID_TAR][self.ID_FIT] for solution in colonies])
fit_empire = self.pop_empires[idx][self.ID_TAR][self.ID_FIT] + self.zeta * np.mean(fit_list)
cost_empires_list.append(fit_empire)
cost_empires_list = np.array(cost_empires_list)
# Find possession probability of each empire based on its total power
cost_empires_list_normalized = cost_empires_list - (np.max(cost_empires_list) + np.min(cost_empires_list))
prob_empires_list = np.abs(cost_empires_list_normalized / np.sum(cost_empires_list_normalized)) # Vector P
uniform_list = np.random.uniform(0, 1, len(prob_empires_list)) # Vector R
vector_D = prob_empires_list - uniform_list
idx_empire = np.argmax(vector_D)
# Find the weakest empire and weakest colony inside it
idx_weakest_empire = np.argmax(cost_empires_list)
if len(self.empires[idx_weakest_empire]) > 0:
colonies_sorted, best, worst = self.get_special_solutions(self.empires[idx_weakest_empire])
self.empires[idx_empire].append(colonies_sorted.pop(-1))
else:
self.empires[idx_empire].append(self.pop_empires.pop(idx_weakest_empire))
self.pop = self.pop_empires + self.pop_colonies