-
Notifications
You must be signed in to change notification settings - Fork 27
/
gl25.py
154 lines (140 loc) · 8.13 KB
/
gl25.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
import numpy as np
from pypop7.optimizers.ga.ga import GA
class GL25(GA):
"""Global and Local genetic algorithm (GL25).
.. note:: `25` means that 25 percentage of function evaluations (or runtime) are first used for *global* search
while the remaining 75 percentage are then used for *local* search.
Parameters
----------
problem : dict
problem arguments with the following common settings (`keys`):
* 'fitness_function' - objective function to be **minimized** (`func`),
* 'ndim_problem' - number of dimensionality (`int`),
* 'upper_boundary' - upper boundary of search range (`array_like`),
* 'lower_boundary' - lower boundary of search range (`array_like`).
options : dict
optimizer options with the following common settings (`keys`):
* 'max_function_evaluations' - maximum of function evaluations (`int`, default: `np.inf`),
* 'max_runtime' - maximal runtime to be allowed (`float`, default: `np.inf`),
* 'seed_rng' - seed for random number generation needed to be *explicitly* set (`int`);
and with the following particular settings (`keys`):
* 'alpha' - global step-size for crossover (`float`, default: `0.8`),
* 'n_female_global' - number of female at global search stage (`int`, default: `200`),
* 'n_male_global' - number of male at global search stage (`int`, default: `400`),
* 'n_female_local' - number of female at local search stage (`int`, default: `5`),
* 'n_male_local' - number of male at local search stage (`int`, default: `100`),
* 'p_global' - percentage of global search stage (`float`, default: `0.25`),.
Examples
--------
Use the optimizer to minimize the well-known test function
`Rosenbrock <http://en.wikipedia.org/wiki/Rosenbrock_function>`_:
.. code-block:: python
:linenos:
>>> import numpy
>>> from pypop7.benchmarks.base_functions import rosenbrock # function to be minimized
>>> from pypop7.optimizers.ga.gl25 import GL25
>>> problem = {'fitness_function': rosenbrock, # define problem arguments
... 'ndim_problem': 2,
... 'lower_boundary': -5*numpy.ones((2,)),
... 'upper_boundary': 5*numpy.ones((2,))}
>>> options = {'max_function_evaluations': 5000, # set optimizer options
... 'seed_rng': 2022}
>>> gl25 = GL25(problem, options) # initialize the optimizer class
>>> results = gl25.optimize() # run the optimization process
>>> # return the number of function evaluations and best-so-far fitness
>>> print(f"GL25: {results['n_function_evaluations']}, {results['best_so_far_y']}")
GL25: 5000, 1.0505276479694516e-05
For its correctness checking of coding, refer to `this code-based repeatability report
<https://tinyurl.com/ytzffmbc>`_ for more details.
Attributes
----------
alpha : `float`
global step-size for crossover.
n_female_global : `int`
number of female at global search stage.
n_female_local : `int`
number of female at local search stage.
n_individuals : `int`
population size.
n_male_global : `int`
number of male at global search stage.
n_male_local : `int`
number of male at local search stage.
p_global : `float`
percentage of global search stage.
References
----------
García-Martínez, C., Lozano, M., Herrera, F., Molina, D. and Sánchez, A.M., 2008.
Global and local real-coded genetic algorithms based on parent-centric crossover operators.
European Journal of Operational Research, 185(3), pp.1088-1113.
https://www.sciencedirect.com/science/article/abs/pii/S0377221706006308
"""
def __init__(self, problem, options):
GA.__init__(self, problem, options)
self.alpha = options.get('alpha', 0.8)
assert self.alpha > 0.0
self.p_global = options.get('p_global', 0.25) # percentage of global search stage
assert 0.0 <= self.p_global <= 1.0
self.n_female_global = options.get('n_female_global', 200) # number of female at global search stage
assert self.n_female_global > 0
self.n_male_global = options.get('n_male_global', 400) # number of male at global search stage
assert self.n_male_global > 0
self.n_female_local = options.get('n_female_local', 5) # number of female at local search stage
assert self.n_female_local > 0
self.n_male_local = options.get('n_male_local', 100) # number of male at local search stage
assert self.n_male_local > 0
self.n_individuals = int(np.maximum(self.n_male_global, self.n_male_local))
self._assortative_mating = 5
self._n_selected = np.zeros((self.n_individuals,)) # number of individuals selected as female
# set maximum of function evaluations and runtime for global search stage
self._max_fe_global = self.p_global*self.max_function_evaluations
self._max_runtime_global = self.p_global*self.max_runtime
def initialize(self, args=None):
x = self.rng_initialization.uniform(self.initial_lower_boundary, self.initial_upper_boundary,
size=(self.n_individuals, self.ndim_problem)) # population
y = np.empty((self.n_individuals,)) # fitness
for i in range(self.n_individuals):
if self._check_terminations():
break
y[i] = self._evaluate_fitness(x[i], args)
return x, y
def iterate(self, x=None, y=None, n_female=None, n_male=None, args=None):
order = np.argsort(y)
x_male, female = x[order[range(n_male)]], order[range(n_female)]
x_female, _n_selected = x[female], self._n_selected[female]
# use the uniform fertility selection (UFS) as female selection mechanism
female = np.argmin(_n_selected)
_n_selected[female] += 1
self._n_selected[order[range(n_female)]] = _n_selected
female = x_female[female]
# use negative assortative mating (NAM) as male selection mechanism
distances = np.empty((self._assortative_mating,))
male = self.rng_optimization.choice(n_male, size=self._assortative_mating, replace=False)
for i, m in enumerate(male):
distances[i] = np.linalg.norm(female - x_male[m])
male = x_male[male[np.argmax(distances)]]
# use the parent-centric BLX crossover operator
interval = np.abs(female - male)
l, u = female - interval*self.alpha, female + interval*self.alpha
xx = self.rng_optimization.uniform(np.clip(l, self.lower_boundary, self.upper_boundary),
np.clip(u, self.lower_boundary, self.upper_boundary))
yy = self._evaluate_fitness(xx, args)
# use the replace worst (RW) strategy
if yy < y[order[-1]]:
x[order[-1]], y[order[-1]], self._n_selected[order[-1]] = xx, yy, 0
self._n_generations += 1
return x, yy
def optimize(self, fitness_function=None, args=None):
fitness, is_switch = GA.optimize(self, fitness_function), True
x, y = self.initialize(args)
yy = y # only for printing
while not self._check_terminations():
self._print_verbose_info(fitness, yy)
if self.n_function_evaluations >= self._max_fe_global or self.runtime >= self._max_runtime_global:
if is_switch: # local search
init, is_switch = range(np.maximum(self.n_female_local, self.n_male_local)), False
x, y, self._n_selected = x[init], y[init], self._n_selected[init]
x, yy = self.iterate(x, y, self.n_female_local, self.n_male_local, args)
else: # global search
x, yy = self.iterate(x, y, self.n_female_global, self.n_male_global, args)
return self._collect(fitness, yy)