|
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
|
|
|
import pandas as pd |
|
10
|
|
|
|
|
11
|
|
|
from multiprocessing.managers import DictProxy |
|
12
|
|
|
|
|
13
|
|
|
from .progress_bar import ProgressBarLVL0, ProgressBarLVL1 |
|
14
|
|
|
from .times_tracker import TimesTracker |
|
15
|
|
|
from .memory import Memory |
|
16
|
|
|
from .print_info import print_info |
|
17
|
|
|
from .stop_run import StopRun |
|
18
|
|
|
|
|
19
|
|
|
|
|
20
|
|
View Code Duplication |
def set_random_seed(nth_process, random_state): |
|
|
|
|
|
|
21
|
|
|
""" |
|
22
|
|
|
Sets the random seed separately for each thread |
|
23
|
|
|
(to avoid getting the same results in each thread) |
|
24
|
|
|
""" |
|
25
|
|
|
if nth_process is None: |
|
26
|
|
|
nth_process = 0 |
|
27
|
|
|
|
|
28
|
|
|
if random_state is None: |
|
29
|
|
|
random_state = np.random.randint(0, high=2 ** 31 - 2, dtype=np.int64) |
|
30
|
|
|
|
|
31
|
|
|
random.seed(random_state + nth_process) |
|
32
|
|
|
np.random.seed(random_state + nth_process) |
|
33
|
|
|
|
|
34
|
|
|
|
|
35
|
|
|
class Search(TimesTracker): |
|
36
|
|
|
def __init__(self): |
|
37
|
|
|
super().__init__() |
|
38
|
|
|
|
|
39
|
|
|
self.optimizers = [] |
|
40
|
|
|
self.new_results_list = [] |
|
41
|
|
|
self.all_results_list = [] |
|
42
|
|
|
|
|
43
|
|
|
self.score_l = [] |
|
44
|
|
|
self.pos_l = [] |
|
45
|
|
|
|
|
46
|
|
|
@TimesTracker.eval_time |
|
47
|
|
|
def _score(self, pos): |
|
48
|
|
|
return self.score(pos) |
|
49
|
|
|
|
|
50
|
|
|
@TimesTracker.iter_time |
|
51
|
|
|
def _initialization(self, init_pos, nth_iter): |
|
52
|
|
|
self.nth_iter = nth_iter |
|
53
|
|
|
self.best_score = self.p_bar.score_best |
|
54
|
|
|
|
|
55
|
|
|
self.init_pos(init_pos) |
|
56
|
|
|
|
|
57
|
|
|
score_new = self._score(init_pos) |
|
58
|
|
|
self.evaluate(score_new) |
|
59
|
|
|
|
|
60
|
|
|
self.pos_l.append(init_pos) |
|
61
|
|
|
self.score_l.append(score_new) |
|
62
|
|
|
|
|
63
|
|
|
self.p_bar.update(score_new, init_pos, nth_iter) |
|
64
|
|
|
|
|
65
|
|
|
@TimesTracker.iter_time |
|
66
|
|
|
def _iteration(self, nth_iter): |
|
67
|
|
|
self.nth_iter = nth_iter |
|
68
|
|
|
self.best_score = self.p_bar.score_best |
|
69
|
|
|
|
|
70
|
|
|
pos_new = self.iterate() |
|
71
|
|
|
|
|
72
|
|
|
score_new = self._score(pos_new) |
|
73
|
|
|
self.evaluate(score_new) |
|
74
|
|
|
|
|
75
|
|
|
self.pos_l.append(pos_new) |
|
76
|
|
|
self.score_l.append(score_new) |
|
77
|
|
|
|
|
78
|
|
|
self.p_bar.update(score_new, pos_new, nth_iter) |
|
79
|
|
|
|
|
80
|
|
|
def _init_search(self): |
|
81
|
|
|
if "progress_bar" in self.verbosity: |
|
82
|
|
|
self.p_bar = ProgressBarLVL1( |
|
83
|
|
|
self.nth_process, self.n_iter, self.objective_function |
|
84
|
|
|
) |
|
85
|
|
|
else: |
|
86
|
|
|
self.p_bar = ProgressBarLVL0( |
|
87
|
|
|
self.nth_process, self.n_iter, self.objective_function |
|
88
|
|
|
) |
|
89
|
|
|
|
|
90
|
|
|
def print_info(self, *args): |
|
91
|
|
|
print_info(*args) |
|
92
|
|
|
|
|
93
|
|
|
def search( |
|
94
|
|
|
self, |
|
95
|
|
|
objective_function, |
|
96
|
|
|
n_iter, |
|
97
|
|
|
max_time=None, |
|
98
|
|
|
max_score=None, |
|
99
|
|
|
early_stopping=None, |
|
100
|
|
|
memory=True, |
|
101
|
|
|
memory_warm_start=None, |
|
102
|
|
|
verbosity=["progress_bar", "print_results", "print_times"], |
|
103
|
|
|
): |
|
104
|
|
|
self.start_time = time.time() |
|
105
|
|
|
|
|
106
|
|
|
if verbosity is False: |
|
107
|
|
|
verbosity = [] |
|
108
|
|
|
|
|
109
|
|
|
self.objective_function = objective_function |
|
110
|
|
|
self.n_iter = n_iter |
|
111
|
|
|
self.max_time = max_time |
|
112
|
|
|
self.max_score = max_score |
|
113
|
|
|
self.early_stopping = early_stopping |
|
114
|
|
|
self.memory = memory |
|
115
|
|
|
self.memory_warm_start = memory_warm_start |
|
116
|
|
|
self.verbosity = verbosity |
|
117
|
|
|
|
|
118
|
|
|
self.stop = StopRun(max_time, max_score, early_stopping) |
|
119
|
|
|
|
|
120
|
|
|
self._init_search() |
|
121
|
|
|
|
|
122
|
|
|
if isinstance(memory, DictProxy): |
|
123
|
|
|
mem = Memory(memory_warm_start, self.conv, dict_proxy=memory) |
|
124
|
|
|
self.score = self.results_mang.score(mem.memory(objective_function)) |
|
125
|
|
|
elif memory is True: |
|
126
|
|
|
mem = Memory(memory_warm_start, self.conv) |
|
127
|
|
|
self.score = self.results_mang.score(mem.memory(objective_function)) |
|
128
|
|
|
else: |
|
129
|
|
|
self.score = self.results_mang.score(objective_function) |
|
130
|
|
|
|
|
131
|
|
|
# loop to initialize N positions |
|
132
|
|
|
for init_pos, nth_iter in zip(self.init_positions, range(n_iter)): |
|
133
|
|
|
if self.stop.check(self.start_time, self.p_bar.score_best, self.score_l): |
|
134
|
|
|
break |
|
135
|
|
|
self._initialization(init_pos, nth_iter) |
|
136
|
|
|
|
|
137
|
|
|
self.finish_initialization() |
|
138
|
|
|
|
|
139
|
|
|
# loop to do the iterations |
|
140
|
|
|
for nth_iter in range(len(self.init_positions), n_iter): |
|
141
|
|
|
if self.stop.check(self.start_time, self.p_bar.score_best, self.score_l): |
|
142
|
|
|
break |
|
143
|
|
|
self._iteration(nth_iter) |
|
144
|
|
|
|
|
145
|
|
|
self.search_data = pd.DataFrame(self.results_mang.results_list) |
|
146
|
|
|
|
|
147
|
|
|
self.best_score = self.p_bar.score_best |
|
148
|
|
|
self.best_value = self.conv.position2value(self.p_bar.pos_best) |
|
149
|
|
|
self.best_para = self.conv.value2para(self.best_value) |
|
150
|
|
|
|
|
151
|
|
|
if memory not in [False, None]: |
|
152
|
|
|
self.memory_dict = mem.memory_dict |
|
|
|
|
|
|
153
|
|
|
else: |
|
154
|
|
|
self.memory_dict = {} |
|
155
|
|
|
|
|
156
|
|
|
self.p_bar.close() |
|
157
|
|
|
|
|
158
|
|
|
self.print_info( |
|
159
|
|
|
verbosity, |
|
160
|
|
|
self.objective_function, |
|
161
|
|
|
self.best_score, |
|
162
|
|
|
self.best_para, |
|
163
|
|
|
self.eval_times, |
|
164
|
|
|
self.iter_times, |
|
165
|
|
|
self.n_iter, |
|
166
|
|
|
) |
|
167
|
|
|
|
SimonBlanke /
Gradient-Free-Optimizers
Loading history...