|
1
|
|
|
import random as rnd |
|
2
|
|
|
import copy |
|
3
|
|
|
from NiaPy.benchmarks.utility import Utility |
|
4
|
|
|
|
|
5
|
|
|
__all__ = ['DifferentialEvolutionAlgorithm'] |
|
6
|
|
|
|
|
7
|
|
|
|
|
8
|
|
View Code Duplication |
class SolutionDE(object): |
|
|
|
|
|
|
9
|
|
|
|
|
10
|
|
|
def __init__(self, D, LB, UB): |
|
11
|
|
|
self.D = D |
|
12
|
|
|
self.LB = LB |
|
13
|
|
|
self.UB = UB |
|
14
|
|
|
|
|
15
|
|
|
self.Solution = [] |
|
16
|
|
|
self.Fitness = float('inf') |
|
17
|
|
|
self.generateSolution() |
|
18
|
|
|
|
|
19
|
|
|
def generateSolution(self): |
|
20
|
|
|
self.Solution = [self.LB + (self.UB - self.LB) * rnd.random() |
|
21
|
|
|
for _i in range(self.D)] |
|
22
|
|
|
|
|
23
|
|
|
def evaluate(self): |
|
24
|
|
|
self.Fitness = SolutionDE.FuncEval(self.D, self.Solution) |
|
25
|
|
|
|
|
26
|
|
|
def repair(self): |
|
27
|
|
|
for i in range(self.D): |
|
28
|
|
|
if self.Solution[i] > self.UB: |
|
29
|
|
|
self.Solution[i] = self.UB |
|
30
|
|
|
if self.Solution[i] < self.LB: |
|
31
|
|
|
self.Solution[i] = self.LB |
|
32
|
|
|
|
|
33
|
|
|
def __eq__(self, other): |
|
34
|
|
|
return self.Solution == other.Solution and self.Fitness == other.Fitness |
|
35
|
|
|
|
|
36
|
|
|
|
|
37
|
|
|
class DifferentialEvolutionAlgorithm(object): |
|
38
|
|
|
r"""Implementation of Differential evolution algorithm. |
|
39
|
|
|
|
|
40
|
|
|
**Algorithm:** Differential evolution algorithm |
|
41
|
|
|
|
|
42
|
|
|
**Date:** 2018 |
|
43
|
|
|
|
|
44
|
|
|
**Author:** Uros Mlakar |
|
45
|
|
|
|
|
46
|
|
|
**License:** MIT |
|
47
|
|
|
|
|
48
|
|
|
**Reference paper:** |
|
49
|
|
|
Storn, Rainer, and Kenneth Price. "Differential evolution - a simple and |
|
50
|
|
|
efficient heuristic for global optimization over continuous spaces." |
|
51
|
|
|
Journal of global optimization 11.4 (1997): 341-359. |
|
52
|
|
|
""" |
|
53
|
|
|
|
|
54
|
|
View Code Duplication |
def __init__(self, D, NP, nFES, F, CR, benchmark): |
|
|
|
|
|
|
55
|
|
|
r"""**__init__(self, D, NP, nFES, F, CR, benchmark)**. |
|
56
|
|
|
|
|
57
|
|
|
Arguments: |
|
58
|
|
|
D {integer} -- dimension of problem |
|
59
|
|
|
|
|
60
|
|
|
NP {integer} -- population size |
|
61
|
|
|
|
|
62
|
|
|
nFES {integer} -- number of function evaluations |
|
63
|
|
|
|
|
64
|
|
|
F {decimal} -- scaling factor |
|
65
|
|
|
|
|
66
|
|
|
CR {decimal} -- crossover rate |
|
67
|
|
|
|
|
68
|
|
|
benchmark {object} -- benchmark implementation object |
|
69
|
|
|
|
|
70
|
|
|
Raises: |
|
71
|
|
|
TypeError -- Raised when given benchmark function which does not exists. |
|
72
|
|
|
|
|
73
|
|
|
""" |
|
74
|
|
|
|
|
75
|
|
|
self.benchmark = Utility().get_benchmark(benchmark) |
|
76
|
|
|
self.D = D # dimension of problem |
|
77
|
|
|
self.Np = NP # population size |
|
78
|
|
|
self.nFES = nFES # number of function evaluations |
|
79
|
|
|
self.F = F # scaling factor |
|
80
|
|
|
self.CR = CR # crossover rate |
|
81
|
|
|
self.Lower = self.benchmark.Lower # lower bound |
|
82
|
|
|
self.Upper = self.benchmark.Upper # upper bound |
|
83
|
|
|
|
|
84
|
|
|
SolutionDE.FuncEval = staticmethod(self.benchmark.function()) |
|
85
|
|
|
self.FEs = 0 |
|
86
|
|
|
self.Done = False |
|
87
|
|
|
self.Population = [] |
|
88
|
|
|
self.bestSolution = SolutionDE(self.D, self.Lower, self.Upper) |
|
89
|
|
|
|
|
90
|
|
|
def evalPopulation(self): |
|
91
|
|
|
"""Evaluate population.""" |
|
92
|
|
|
|
|
93
|
|
|
for p in self.Population: |
|
94
|
|
|
p.evaluate() |
|
95
|
|
|
if p.Fitness < self.bestSolution.Fitness: |
|
96
|
|
|
self.bestSolution = copy.deepcopy(p) |
|
97
|
|
|
|
|
98
|
|
|
def initPopulation(self): |
|
99
|
|
|
"""Initialize population.""" |
|
100
|
|
|
|
|
101
|
|
|
for _i in range(self.Np): |
|
102
|
|
|
self.Population.append(SolutionDE(self.D, self.Lower, self.Upper)) |
|
103
|
|
|
|
|
104
|
|
|
def tryEval(self, v): |
|
105
|
|
|
"""Check evaluations.""" |
|
106
|
|
|
if self.FEs <= self.nFES: |
|
107
|
|
|
v.evaluate() |
|
108
|
|
|
self.FEs += 1 |
|
109
|
|
|
else: |
|
110
|
|
|
self.Done = True |
|
111
|
|
|
|
|
112
|
|
|
def generationStep(self, Population): |
|
113
|
|
|
"""Implement main generation step.""" |
|
114
|
|
|
|
|
115
|
|
View Code Duplication |
newPopulation = [] |
|
|
|
|
|
|
116
|
|
|
for i in range(self.Np): |
|
117
|
|
|
newSolution = SolutionDE(self.D, self.Lower, self.Upper) |
|
118
|
|
|
|
|
119
|
|
|
r = rnd.sample(range(0, self.Np), 3) |
|
120
|
|
|
while i in r: |
|
121
|
|
|
r = rnd.sample(range(0, self.Np), 3) |
|
122
|
|
|
jrand = int(rnd.random() * self.Np) |
|
123
|
|
|
|
|
124
|
|
|
for j in range(self.D): |
|
125
|
|
|
if rnd.random() < self.CR or j == jrand: |
|
126
|
|
|
newSolution.Solution[j] = Population[r[0]].Solution[j] + self.F * \ |
|
127
|
|
|
(Population[r[1]].Solution[j] - Population[r[2]].Solution[j]) |
|
128
|
|
|
else: |
|
129
|
|
|
newSolution.Solution[j] = Population[i].Solution[j] |
|
130
|
|
|
newSolution.repair() |
|
131
|
|
|
self.tryEval(newSolution) |
|
132
|
|
|
|
|
133
|
|
|
if newSolution.Fitness < self.bestSolution.Fitness: |
|
134
|
|
|
self.bestSolution = copy.deepcopy(newSolution) |
|
135
|
|
|
if newSolution.Fitness < self.Population[i].Fitness: |
|
136
|
|
|
newPopulation.append(newSolution) |
|
137
|
|
|
else: |
|
138
|
|
|
newPopulation.append(Population[i]) |
|
139
|
|
|
return newPopulation |
|
140
|
|
|
|
|
141
|
|
|
def run(self): |
|
142
|
|
|
"""Run.""" |
|
143
|
|
|
self.initPopulation() |
|
144
|
|
|
self.evalPopulation() |
|
145
|
|
|
self.FEs = self.Np |
|
146
|
|
|
while not self.Done: |
|
147
|
|
|
self.Population = self.generationStep(self.Population) |
|
148
|
|
|
return self.bestSolution.Fitness |
|
149
|
|
|
|
NiaOrg /
NiaPy
Loading history...