|
1
|
|
|
# Author: Simon Blanke |
|
2
|
|
|
# Email: [email protected] |
|
3
|
|
|
# License: MIT License |
|
4
|
|
|
|
|
5
|
|
|
import copy |
|
6
|
|
|
import random |
|
7
|
|
|
|
|
8
|
|
|
import numpy as np |
|
9
|
|
|
|
|
10
|
|
|
from .base_population_optimizer import BasePopulationOptimizer |
|
11
|
|
|
from ...search import Search |
|
12
|
|
|
from ..local_opt import SimulatedAnnealingOptimizer |
|
13
|
|
|
|
|
14
|
|
|
|
|
15
|
|
|
class ParallelTemperingOptimizer(BasePopulationOptimizer, Search): |
|
16
|
|
|
name = "Parallel Tempering" |
|
17
|
|
|
|
|
18
|
|
|
def __init__( |
|
19
|
|
|
self, |
|
20
|
|
|
*args, |
|
21
|
|
|
population=5, |
|
22
|
|
|
n_iter_swap=5, |
|
23
|
|
|
**kwargs, |
|
24
|
|
|
): |
|
25
|
|
|
super().__init__(*args, **kwargs) |
|
26
|
|
|
|
|
27
|
|
|
self.population = population |
|
28
|
|
|
self.n_iter_swap = n_iter_swap |
|
29
|
|
|
|
|
30
|
|
|
self.systems = self._create_population(SimulatedAnnealingOptimizer) |
|
31
|
|
|
for system in self.systems: |
|
32
|
|
|
system.temp = 1.1 ** random.uniform(0, 25) |
|
33
|
|
|
self.optimizers = self.systems |
|
34
|
|
|
|
|
35
|
|
View Code Duplication |
def _swap_pos(self): |
|
|
|
|
|
|
36
|
|
|
for _p1_ in self.systems: |
|
37
|
|
|
_systems_temp = copy.copy(self.systems) |
|
38
|
|
|
if len(_systems_temp) > 1: |
|
39
|
|
|
_systems_temp.remove(_p1_) |
|
40
|
|
|
|
|
41
|
|
|
rand = random.uniform(0, 1) * 100 |
|
42
|
|
|
_p2_ = np.random.choice(_systems_temp) |
|
43
|
|
|
|
|
44
|
|
|
p_accept = self._accept_swap(_p1_, _p2_) |
|
45
|
|
|
if p_accept > rand: |
|
46
|
|
|
_p1_.temp, _p2_.temp = (_p2_.temp, _p1_.temp) |
|
47
|
|
|
|
|
48
|
|
|
def _accept_swap(self, _p1_, _p2_): |
|
49
|
|
|
denom = _p1_.score_current + _p2_.score_current |
|
50
|
|
|
|
|
51
|
|
|
if denom == 0: |
|
52
|
|
|
return 100 |
|
53
|
|
|
elif abs(denom) == np.inf: |
|
54
|
|
|
return 0 |
|
55
|
|
|
else: |
|
56
|
|
|
score_diff_norm = (_p1_.score_current - _p2_.score_current) / denom |
|
57
|
|
|
|
|
58
|
|
|
temp = (1 / _p1_.temp) - (1 / _p2_.temp) |
|
59
|
|
|
return np.exp(score_diff_norm * temp) * 100 |
|
60
|
|
|
|
|
61
|
|
|
def init_pos(self, pos): |
|
62
|
|
|
nth_pop = self.nth_iter % len(self.systems) |
|
63
|
|
|
|
|
64
|
|
|
self.p_current = self.systems[nth_pop] |
|
65
|
|
|
self.p_current.init_pos(pos) |
|
66
|
|
|
|
|
67
|
|
|
def iterate(self): |
|
68
|
|
|
nth_iter = self._iterations(self.systems) |
|
69
|
|
|
self.p_current = self.systems[nth_iter % len(self.systems)] |
|
70
|
|
|
|
|
71
|
|
|
return self.p_current.iterate() |
|
72
|
|
|
|
|
73
|
|
|
def evaluate(self, score_new): |
|
74
|
|
|
nth_iter = self._iterations(self.systems) |
|
75
|
|
|
|
|
76
|
|
|
notZero = self.n_iter_swap != 0 |
|
77
|
|
|
modZero = nth_iter % self.n_iter_swap == 0 |
|
78
|
|
|
|
|
79
|
|
|
if notZero and modZero: |
|
80
|
|
|
self._swap_pos() |
|
81
|
|
|
|
|
82
|
|
|
self.p_current.evaluate(score_new) |
|
83
|
|
|
|
SimonBlanke /
Gradient-Free-Optimizers
Loading history...