|
1
|
|
|
# Author: Simon Blanke |
|
2
|
|
|
# Email: [email protected] |
|
3
|
|
|
# License: MIT License |
|
4
|
|
|
|
|
5
|
|
|
|
|
6
|
|
|
import random |
|
7
|
|
|
|
|
8
|
|
|
import numpy as np |
|
9
|
|
|
|
|
10
|
|
|
from .simulated_annealing import SimulatedAnnealingOptimizer |
|
11
|
|
|
from ..local import HillClimbingPositioner |
|
12
|
|
|
|
|
13
|
|
|
|
|
14
|
|
|
class ParallelTemperingOptimizer(SimulatedAnnealingOptimizer): |
|
15
|
|
|
def __init__(self, _opt_args_): |
|
16
|
|
|
super().__init__(_opt_args_) |
|
17
|
|
|
self.n_iter_swap = _opt_args_.n_iter_swap |
|
18
|
|
|
self.n_positioners = len(_opt_args_.system_temperatures) |
|
19
|
|
|
|
|
20
|
|
|
def _init_annealer(self, _cand_): |
|
21
|
|
|
temp = self._opt_args_.system_temperatures[self.i] |
|
22
|
|
|
_p_ = System(**self._opt_args_.kwargs_opt, temp=temp) |
|
23
|
|
|
|
|
24
|
|
|
_p_.pos_new = _cand_._space_.get_random_pos() |
|
25
|
|
|
|
|
26
|
|
|
return _p_ |
|
27
|
|
|
|
|
28
|
|
|
def _swap_pos(self, _cand_): |
|
29
|
|
|
_p_list_temp = self.p_list[:] |
|
30
|
|
|
|
|
31
|
|
|
for _p1_ in self.p_list: |
|
32
|
|
|
rand = random.uniform(0, 1) |
|
33
|
|
|
_p2_ = np.random.choice(_p_list_temp) |
|
34
|
|
|
|
|
35
|
|
|
p_accept = self._accept_swap(_p1_, _p2_) |
|
36
|
|
|
if p_accept > rand: |
|
37
|
|
|
temp_temp = _p1_.temp # haha! |
|
38
|
|
|
_p1_.temp = _p2_.temp |
|
39
|
|
|
_p2_.temp = temp_temp |
|
40
|
|
|
|
|
41
|
|
|
def _accept_swap(self, _p1_, _p2_): |
|
42
|
|
|
denom = _p1_.score_current + _p2_.score_current |
|
43
|
|
|
|
|
44
|
|
|
if denom == 0: |
|
45
|
|
|
return 100 |
|
46
|
|
|
elif abs(denom) == np.inf: |
|
47
|
|
|
return 0 |
|
48
|
|
|
else: |
|
49
|
|
|
score_diff_norm = (_p1_.score_current - _p2_.score_current) / denom |
|
50
|
|
|
|
|
51
|
|
|
temp = (1 / _p1_.temp) - (1 / _p2_.temp) |
|
52
|
|
|
return np.exp(score_diff_norm * temp) |
|
53
|
|
|
|
|
54
|
|
|
def _anneal_system(self, _cand_, _p_): |
|
55
|
|
|
self._p_ = _p_ |
|
56
|
|
|
super()._iterate(0, _cand_) |
|
57
|
|
|
|
|
58
|
|
|
def _iterate(self, i, _cand_): |
|
59
|
|
|
_p_current = self.p_list[i % len(self.p_list)] |
|
60
|
|
|
|
|
61
|
|
|
self._anneal_system(_cand_, _p_current) |
|
62
|
|
|
|
|
63
|
|
|
if self.n_iter_swap != 0 and i % self.n_iter_swap == 0: |
|
64
|
|
|
self._swap_pos(_cand_) |
|
65
|
|
|
|
|
66
|
|
|
return _cand_ |
|
67
|
|
|
|
|
68
|
|
|
def _init_iteration(self, _cand_): |
|
69
|
|
|
p = self._init_annealer(_cand_) |
|
70
|
|
|
|
|
71
|
|
|
self._optimizer_eval(_cand_, p) |
|
72
|
|
|
self._update_pos(_cand_, p) |
|
73
|
|
|
|
|
74
|
|
|
return p |
|
75
|
|
|
|
|
76
|
|
|
def _finish_search(self): |
|
77
|
|
|
self._pbar_.close_p_bar() |
|
78
|
|
|
|
|
79
|
|
|
|
|
80
|
|
|
class System(HillClimbingPositioner): |
|
81
|
|
|
def __init__(self, *args, **kwargs): |
|
82
|
|
|
super().__init__(*args, **kwargs) |
|
83
|
|
|
self.temp = kwargs["temp"] |
|
84
|
|
|
|
SimonBlanke /
Hyperactive
