|
1
|
|
|
# Author: Simon Blanke |
|
2
|
|
|
# Email: [email protected] |
|
3
|
|
|
# License: MIT License |
|
4
|
|
|
|
|
5
|
|
|
import time |
|
6
|
|
|
import random |
|
7
|
|
|
|
|
8
|
|
|
import numpy as np |
|
9
|
|
|
from tqdm import tqdm |
|
10
|
|
|
|
|
11
|
|
|
from .init_positions import Initializer |
|
12
|
|
|
from .progress_bar import ProgressBarLVL0, ProgressBarLVL1 |
|
13
|
|
|
from .conv import values2positions, positions2values, position2value |
|
14
|
|
|
from .times_tracker import TimesTracker |
|
15
|
|
|
|
|
16
|
|
|
p_bar_dict = { |
|
17
|
|
|
False: ProgressBarLVL0, |
|
18
|
|
|
True: ProgressBarLVL1, |
|
19
|
|
|
} |
|
20
|
|
|
|
|
21
|
|
|
|
|
22
|
|
|
def time_exceeded(start_time, max_time): |
|
23
|
|
|
run_time = time.time() - start_time |
|
24
|
|
|
return max_time and run_time > max_time |
|
25
|
|
|
|
|
26
|
|
|
|
|
27
|
|
|
def set_random_seed(nth_process, random_state): |
|
28
|
|
|
"""Sets the random seed separately for each thread (to avoid getting the same results in each thread)""" |
|
29
|
|
|
if nth_process is None: |
|
30
|
|
|
nth_process = 0 |
|
31
|
|
|
|
|
32
|
|
|
if random_state is None: |
|
33
|
|
|
random_state = np.random.randint(0, high=2 ** 32 - 2) |
|
34
|
|
|
|
|
35
|
|
|
random.seed(random_state + nth_process) |
|
36
|
|
|
np.random.seed(random_state + nth_process) |
|
37
|
|
|
|
|
38
|
|
|
|
|
39
|
|
|
class Search(TimesTracker): |
|
40
|
|
|
def __init__(self): |
|
41
|
|
|
super().__init__() |
|
42
|
|
|
|
|
43
|
|
|
self.optimizers = [] |
|
44
|
|
|
|
|
45
|
|
|
@TimesTracker.eval_time_dec |
|
46
|
|
|
def _score(self, pos): |
|
47
|
|
|
pos_tuple = tuple(pos) |
|
48
|
|
|
|
|
49
|
|
|
if self.memory is True and pos_tuple in self.memory_dict: |
|
50
|
|
|
return self.memory_dict[pos_tuple] |
|
51
|
|
|
else: |
|
52
|
|
|
score = self.objective_function(pos) |
|
53
|
|
|
self.memory_dict[pos_tuple] = score |
|
54
|
|
|
return score |
|
55
|
|
|
|
|
56
|
|
|
def _init_memory(self, memory): |
|
57
|
|
|
self.memory_dict = {} |
|
58
|
|
|
|
|
59
|
|
|
if isinstance(memory, dict): |
|
60
|
|
|
values_list = memory["values"] |
|
61
|
|
|
scores = memory["scores"] |
|
62
|
|
|
|
|
63
|
|
|
value_tuple_list = list(map(tuple, values_list)) |
|
64
|
|
|
self.memory_dict = dict(zip(value_tuple_list, scores)) |
|
65
|
|
|
|
|
66
|
|
|
@TimesTracker.iter_time_dec |
|
67
|
|
|
def _initialization(self, init_pos): |
|
68
|
|
|
self.init_pos(init_pos) |
|
69
|
|
|
|
|
70
|
|
|
value_new = position2value(self.search_space, init_pos) |
|
71
|
|
|
score_new = self._score(value_new) |
|
72
|
|
|
self.evaluate(score_new) |
|
73
|
|
|
|
|
74
|
|
|
self.p_bar.update(score_new, value_new) |
|
75
|
|
|
|
|
76
|
|
|
@TimesTracker.iter_time_dec |
|
77
|
|
|
def _iteration(self): |
|
78
|
|
|
pos_new = self.iterate() |
|
79
|
|
|
|
|
80
|
|
|
value_new = position2value(self.search_space, pos_new) |
|
81
|
|
|
score_new = self._score(value_new) |
|
82
|
|
|
self.evaluate(score_new) |
|
83
|
|
|
|
|
84
|
|
|
self.p_bar.update(score_new, value_new) |
|
85
|
|
|
|
|
86
|
|
|
def search( |
|
87
|
|
|
self, |
|
88
|
|
|
objective_function, |
|
89
|
|
|
n_iter, |
|
90
|
|
|
initialize={"grid": 4, "random": 2, "vertices": 4}, |
|
91
|
|
|
max_time=None, |
|
92
|
|
|
memory=True, |
|
93
|
|
|
progress_bar=True, |
|
94
|
|
|
print_results=True, |
|
95
|
|
|
random_state=None, |
|
96
|
|
|
nth_process=None, |
|
97
|
|
|
): |
|
98
|
|
|
start_time = time.time() |
|
99
|
|
|
|
|
100
|
|
|
self.objective_function = objective_function |
|
101
|
|
|
self.memory = memory |
|
102
|
|
|
|
|
103
|
|
|
self._init_memory(memory) |
|
104
|
|
|
self.p_bar = p_bar_dict[progress_bar](nth_process, n_iter, objective_function) |
|
105
|
|
|
|
|
106
|
|
|
set_random_seed(nth_process, random_state) |
|
107
|
|
|
|
|
108
|
|
|
# get init positions |
|
109
|
|
|
init = Initializer(self.search_space) |
|
110
|
|
|
init_positions = init.set_pos(initialize) |
|
111
|
|
|
|
|
112
|
|
|
# loop to initialize N positions |
|
113
|
|
|
for init_pos in init_positions: |
|
114
|
|
|
if time_exceeded(start_time, max_time): |
|
115
|
|
|
break |
|
116
|
|
|
self._initialization(init_pos) |
|
117
|
|
|
|
|
118
|
|
|
# loop to do the iterations |
|
119
|
|
|
for nth_iter in range(len(init_positions), n_iter): |
|
120
|
|
|
if time_exceeded(start_time, max_time): |
|
121
|
|
|
break |
|
122
|
|
|
self._iteration() |
|
123
|
|
|
|
|
124
|
|
|
self.values = np.array(list(self.memory_dict.keys())) |
|
125
|
|
|
self.scores = np.array(list(self.memory_dict.values())).reshape(-1, 1) |
|
126
|
|
|
|
|
127
|
|
|
self.p_bar.close(print_results) |
|
128
|
|
|
|
|
129
|
|
|
self.best_score = self.p_bar.score_best |
|
130
|
|
|
self.best_values = self.p_bar.values_best |
|
131
|
|
|
|
|
132
|
|
|
|
SimonBlanke /
Gradient-Free-Optimizers
