|
1
|
|
|
# gradient_free_optimizers/hilbert_grid_search.py |
|
2
|
|
|
# Author: Simon Blanke |
|
3
|
|
|
# Email: [email protected] |
|
4
|
|
|
# License: MIT License |
|
5
|
|
|
|
|
6
|
|
|
import numpy as np |
|
7
|
|
|
from numpy_hilbert_curve import decode |
|
8
|
|
|
from ..base_optimizer import BaseOptimizer |
|
9
|
|
|
|
|
10
|
|
|
|
|
11
|
|
|
class HilbertGridSearchOptimizer(BaseOptimizer): |
|
12
|
|
|
def __init__( |
|
13
|
|
|
self, |
|
14
|
|
|
search_space, |
|
15
|
|
|
initialize={"grid": 4, "random": 2, "vertices": 4}, |
|
16
|
|
|
constraints=[], |
|
17
|
|
|
random_state=None, |
|
18
|
|
|
rand_rest_p=0, |
|
19
|
|
|
nth_process=None, |
|
20
|
|
|
step_size=1, |
|
21
|
|
|
): |
|
22
|
|
|
super().__init__( |
|
23
|
|
|
search_space=search_space, |
|
24
|
|
|
initialize=initialize, |
|
25
|
|
|
constraints=constraints, |
|
26
|
|
|
random_state=random_state, |
|
27
|
|
|
rand_rest_p=rand_rest_p, |
|
28
|
|
|
nth_process=nth_process, |
|
29
|
|
|
) |
|
30
|
|
|
self.step_size = step_size |
|
31
|
|
|
self.Z = 0 # Current Hilbert integer |
|
32
|
|
|
self.valid_count = 0 # Counter for valid points |
|
33
|
|
|
|
|
34
|
|
|
def hilbert_move(self): |
|
35
|
|
|
while True: |
|
36
|
|
|
# Decode the current Hilbert integer to get an nD point |
|
37
|
|
|
point = decode(np.array([self.Z]), self.conv.n_dim, self.conv.n_dim)[0] |
|
38
|
|
|
self.Z += 1 |
|
39
|
|
|
# Check if the point is within the grid bounds |
|
40
|
|
|
if all(point[i] < self.conv.dim_sizes[i] for i in range(self.conv.n_dim)): |
|
41
|
|
|
self.valid_count += 1 |
|
42
|
|
|
# Take every step_size-th valid point |
|
43
|
|
|
if self.valid_count % self.step_size == 1: |
|
44
|
|
|
return np.array(point) |
|
45
|
|
|
# Continue if point is out of bounds |
|
46
|
|
|
|
|
47
|
|
|
@BaseOptimizer.track_new_pos |
|
48
|
|
|
def iterate(self): |
|
49
|
|
|
pos_new = self.hilbert_move() |
|
50
|
|
|
pos_new = self.conv2pos(pos_new) |
|
51
|
|
|
return pos_new |
|
52
|
|
|
|
|
53
|
|
|
@BaseOptimizer.track_new_score |
|
54
|
|
|
def evaluate(self, score_new): |
|
55
|
|
|
BaseOptimizer.evaluate(self, score_new) |
SimonBlanke /
Gradient-Free-Optimizers
