|
1
|
|
|
# Author: Simon Blanke |
|
2
|
|
|
# Email: [email protected] |
|
3
|
|
|
# License: MIT License |
|
4
|
|
|
|
|
5
|
|
|
import random |
|
6
|
|
|
import numpy as np |
|
7
|
|
|
|
|
8
|
|
|
|
|
9
|
|
|
from . import HillClimbingOptimizer |
|
10
|
|
|
from ...base_positioner import BasePositioner |
|
11
|
|
|
from scipy.spatial.distance import euclidean |
|
12
|
|
|
|
|
13
|
|
|
|
|
14
|
|
|
def gaussian(distance, sig): |
|
15
|
|
|
return sig * np.exp(-np.power(distance, 2.0) / (2 * np.power(sig, 2.0))) |
|
16
|
|
|
|
|
17
|
|
|
|
|
18
|
|
|
class TabuOptimizer(HillClimbingOptimizer): |
|
19
|
|
|
def __init__(self, *args, **kwargs): |
|
20
|
|
|
super().__init__(*args, **kwargs) |
|
21
|
|
|
|
|
22
|
|
|
def _tabu_pos(self, pos, _p_): |
|
23
|
|
|
_p_.add_tabu(pos) |
|
24
|
|
|
|
|
25
|
|
|
return _p_ |
|
26
|
|
|
|
|
27
|
|
|
def _iterate(self, i, _cand_, _p_, X, y): |
|
28
|
|
|
_p_.pos_new = _p_.move_climb(_cand_, _p_.pos_current) |
|
29
|
|
|
_p_.score_new = _cand_.eval_pos(_p_.pos_new, X, y) |
|
30
|
|
|
|
|
31
|
|
|
if _p_.score_new > _cand_.score_best: |
|
32
|
|
|
_cand_, _p_ = self._update_pos(_cand_, _p_) |
|
33
|
|
|
else: |
|
34
|
|
|
_p_ = self._tabu_pos(_p_.pos_new, _p_) |
|
35
|
|
|
|
|
36
|
|
|
return _cand_ |
|
37
|
|
|
|
|
38
|
|
|
def _init_opt_positioner(self, _cand_, X, y): |
|
39
|
|
|
return super()._init_base_positioner( |
|
40
|
|
|
_cand_, positioner=TabuPositioner, pos_para=self.pos_para |
|
41
|
|
|
) |
|
42
|
|
|
|
|
43
|
|
|
|
|
44
|
|
|
class TabuPositioner(BasePositioner): |
|
45
|
|
|
def __init__(self, epsilon): |
|
46
|
|
|
super().__init__(epsilon) |
|
47
|
|
|
self.tabus = [] |
|
48
|
|
|
|
|
49
|
|
|
def add_tabu(self, tabu): |
|
50
|
|
|
self.tabus.append(tabu) |
|
51
|
|
|
|
|
52
|
|
|
def move_climb(self, _cand_, pos, epsilon_mod=1): |
|
53
|
|
|
sigma = 3 + _cand_._space_.dim * self.epsilon * epsilon_mod |
|
54
|
|
|
pos_normal = np.random.normal(pos, sigma, pos.shape) |
|
55
|
|
|
pos_new_int = np.rint(pos_normal) |
|
56
|
|
|
|
|
57
|
|
|
for tabu in self.tabus: |
|
58
|
|
|
distance = euclidean(pos_new_int, tabu) |
|
59
|
|
|
sigma_mean = sigma.mean() |
|
60
|
|
|
p_discard = gaussian(distance, sigma_mean) |
|
61
|
|
|
rand = random.uniform(0, 1) |
|
62
|
|
|
|
|
63
|
|
|
if p_discard > rand: |
|
64
|
|
|
pos_normal = np.random.normal(pos, sigma, pos.shape) |
|
65
|
|
|
pos_new_int = np.rint(pos_normal) |
|
66
|
|
|
|
|
67
|
|
|
n_zeros = [0] * len(_cand_._space_.dim) |
|
68
|
|
|
pos = np.clip(pos_new_int, n_zeros, _cand_._space_.dim) |
|
69
|
|
|
|
|
70
|
|
|
return pos |
|
71
|
|
|
|
SimonBlanke /
Hyperactive
