|
1
|
|
|
# Author: Simon Blanke |
|
2
|
|
|
# Email: [email protected] |
|
3
|
|
|
# License: MIT License |
|
4
|
|
|
|
|
5
|
|
|
import os |
|
6
|
|
|
import time |
|
7
|
|
|
import numpy as np |
|
8
|
|
|
|
|
9
|
|
|
from .candidate import Candidate |
|
10
|
|
|
|
|
11
|
|
|
from hypermemory import Hypermemory |
|
12
|
|
|
from importlib import import_module |
|
13
|
|
|
|
|
14
|
|
|
|
|
15
|
|
|
def meta_data_path(): |
|
16
|
|
|
current_path = os.path.realpath(__file__) |
|
17
|
|
|
return current_path.rsplit("/", 1)[0] + "/meta_data/" |
|
18
|
|
|
|
|
19
|
|
|
|
|
20
|
|
|
optimizer_dict = { |
|
21
|
|
|
"HillClimbing": "HillClimbingOptimizer", |
|
22
|
|
|
"StochasticHillClimbing": "StochasticHillClimbingOptimizer", |
|
23
|
|
|
"TabuSearch": "TabuOptimizer", |
|
24
|
|
|
"RandomSearch": "RandomSearchOptimizer", |
|
25
|
|
|
"RandomRestartHillClimbing": "RandomRestartHillClimbingOptimizer", |
|
26
|
|
|
"RandomAnnealing": "RandomAnnealingOptimizer", |
|
27
|
|
|
"SimulatedAnnealing": "SimulatedAnnealingOptimizer", |
|
28
|
|
|
"StochasticTunneling": "StochasticTunnelingOptimizer", |
|
29
|
|
|
"ParallelTempering": "ParallelTemperingOptimizer", |
|
30
|
|
|
"ParticleSwarm": "ParticleSwarmOptimizer", |
|
31
|
|
|
"EvolutionStrategy": "EvolutionStrategyOptimizer", |
|
32
|
|
|
"Bayesian": "BayesianOptimizer", |
|
33
|
|
|
"TPE": "TreeStructuredParzenEstimators", |
|
34
|
|
|
"DecisionTree": "DecisionTreeOptimizer", |
|
35
|
|
|
} |
|
36
|
|
|
|
|
37
|
|
|
|
|
38
|
|
|
class ShortTermMemory: |
|
39
|
|
|
def __init__(self, _space_, _main_args_, _cand_): |
|
40
|
|
|
self._space_ = _space_ |
|
41
|
|
|
self._main_args_ = _main_args_ |
|
42
|
|
|
|
|
43
|
|
|
self.pos_best = None |
|
44
|
|
|
self.score_best = -np.inf |
|
45
|
|
|
|
|
46
|
|
|
self.memory_type = _main_args_.memory |
|
47
|
|
|
self.memory_dict = {} |
|
48
|
|
|
self.memory_dict_new = {} |
|
49
|
|
|
|
|
50
|
|
|
self.meta_data_found = False |
|
51
|
|
|
|
|
52
|
|
|
self.n_dims = None |
|
53
|
|
|
|
|
54
|
|
|
|
|
55
|
|
|
class HypermemoryWrapper: |
|
56
|
|
|
def __init__(self): |
|
57
|
|
|
pass |
|
58
|
|
|
|
|
59
|
|
|
def load_memory(self, X, y): |
|
60
|
|
|
self.mem = Hypermemory(X, y, self.obj_func, self.search_space,) |
|
61
|
|
|
self.eval_pos = self.eval_pos_Mem |
|
62
|
|
|
self.memory_dict = self.mem.load() |
|
63
|
|
|
|
|
64
|
|
|
|
|
65
|
|
|
class SearchProcess: |
|
66
|
|
|
def __init__(self, nth_process, pro_arg, verb): |
|
67
|
|
|
self.nth_process = nth_process |
|
68
|
|
|
self.pro_arg = pro_arg |
|
69
|
|
|
self.verb = verb |
|
70
|
|
|
|
|
71
|
|
|
kwargs = self.pro_arg.kwargs |
|
72
|
|
|
module = import_module("gradient_free_optimizers") |
|
73
|
|
|
|
|
74
|
|
|
self.opt_class = getattr(module, optimizer_dict[pro_arg.optimizer]) |
|
75
|
|
|
self.obj_func = kwargs["objective_function"] |
|
76
|
|
|
self.func_para = kwargs["function_parameter"] |
|
77
|
|
|
self.search_space = kwargs["search_space"] |
|
78
|
|
|
self.n_iter = kwargs["n_iter"] |
|
79
|
|
|
self.n_jobs = kwargs["n_jobs"] |
|
80
|
|
|
self.memory = kwargs["memory"] |
|
81
|
|
|
self.init_para = kwargs["init_para"] |
|
82
|
|
|
self.distribution = kwargs["distribution"] |
|
83
|
|
|
|
|
84
|
|
|
self.cand = Candidate( |
|
85
|
|
|
self.obj_func, |
|
86
|
|
|
self.func_para, |
|
87
|
|
|
self.search_space, |
|
88
|
|
|
self.init_para, |
|
89
|
|
|
self.memory, |
|
90
|
|
|
verb, |
|
91
|
|
|
) |
|
92
|
|
|
|
|
93
|
|
|
# self.space = SearchSpace(kwargs["search_space"], verb) |
|
94
|
|
|
# self.model = Model(self.obj_func, self.func_para, verb) |
|
95
|
|
|
# self.init = InitSearchPosition(self.init_para, self.space, verb) |
|
96
|
|
|
|
|
97
|
|
|
self.start_time = time.time() |
|
98
|
|
|
self.i = 0 |
|
99
|
|
|
|
|
100
|
|
|
self.memory_dict = {} |
|
101
|
|
|
self.memory_dict_new = {} |
|
102
|
|
|
|
|
103
|
|
|
self._score = -np.inf |
|
104
|
|
|
self._pos = None |
|
105
|
|
|
|
|
106
|
|
|
self.score_best = -np.inf |
|
107
|
|
|
self.pos_best = None |
|
108
|
|
|
|
|
109
|
|
|
self.score_list = [] |
|
110
|
|
|
self.pos_list = [] |
|
111
|
|
|
|
|
112
|
|
|
self.eval_time = [] |
|
113
|
|
|
self.iter_times = [] |
|
114
|
|
|
|
|
115
|
|
|
# self._memory_processor() |
|
116
|
|
|
|
|
117
|
|
|
def _memory_processor(self): |
|
118
|
|
|
if self.memory == "long": |
|
119
|
|
|
self.mem = Hypermemory( |
|
120
|
|
|
self.func_para["features"], |
|
121
|
|
|
self.func_para["target"], |
|
122
|
|
|
self.obj_func, |
|
123
|
|
|
self.search_space, |
|
124
|
|
|
) |
|
125
|
|
|
self.eval_pos = self.eval_pos_Mem |
|
126
|
|
|
|
|
127
|
|
|
self.memory_dict = self.mem.load() |
|
128
|
|
|
|
|
129
|
|
|
else: |
|
130
|
|
|
print("Warning: Memory not defined") |
|
131
|
|
|
self.mem = None |
|
132
|
|
|
self.eval_pos = self.eval_pos_noMem |
|
133
|
|
|
|
|
134
|
|
|
if self.mem: |
|
135
|
|
|
if self.mem.meta_data_found: |
|
136
|
|
|
self.pos_best = self.mem.pos_best |
|
137
|
|
|
self.score_best = self.mem.score_best |
|
138
|
|
|
|
|
139
|
|
|
def _get_warm_start(self): |
|
140
|
|
|
return self.space.pos2para(self.pos_best) |
|
141
|
|
|
|
|
142
|
|
|
def _process_results(self): |
|
143
|
|
|
self.total_time = time.time() - self.start_time |
|
144
|
|
|
start_point = self.verb.info.print_start_point(self) |
|
145
|
|
|
|
|
146
|
|
|
if self.memory == "long": |
|
147
|
|
|
self.mem.dump(self.memory_dict_new) |
|
148
|
|
|
|
|
149
|
|
|
return start_point |
|
150
|
|
|
|
|
151
|
|
|
def search(self, start_time, max_time, nth_process): |
|
152
|
|
|
self._initialize_search(nth_process) |
|
153
|
|
|
|
|
154
|
|
|
# loop to initialize N positions |
|
155
|
|
|
for nth_init in range(len(self.opt.init_positions)): |
|
156
|
|
|
pos_new = self.opt.init_pos(nth_init) |
|
157
|
|
|
score_new = self.cand.get_score(pos_new, 0) |
|
158
|
|
|
self.opt.evaluate(score_new) |
|
159
|
|
|
|
|
160
|
|
|
# loop to do the iterations |
|
161
|
|
|
for nth_iter in range(len(self.opt.init_positions), self.n_iter): |
|
162
|
|
|
pos_new = self.opt.iterate(nth_iter) |
|
163
|
|
|
score_new = self.cand.get_score(pos_new, nth_iter) |
|
164
|
|
|
self.opt.evaluate(score_new) |
|
165
|
|
|
|
|
166
|
|
|
if self._time_exceeded(start_time, max_time): |
|
167
|
|
|
break |
|
168
|
|
|
|
|
169
|
|
|
self.verb.p_bar.close_p_bar() |
|
170
|
|
|
|
|
171
|
|
|
return self.opt.p_list |
|
172
|
|
|
|
|
173
|
|
|
def _time_exceeded(self, start_time, max_time): |
|
174
|
|
|
run_time = time.time() - start_time |
|
175
|
|
|
return max_time and run_time > max_time |
|
176
|
|
|
|
|
177
|
|
|
def _initialize_search(self, nth_process): |
|
178
|
|
|
n_positions = self.pro_arg.n_positions |
|
179
|
|
|
init_positions = self.cand.init.set_start_pos(n_positions) |
|
180
|
|
|
self.opt = self.opt_class(init_positions, self.cand.space.dim, opt_para={}) |
|
181
|
|
|
|
|
182
|
|
|
self.pro_arg.set_random_seed(nth_process) |
|
183
|
|
|
self.verb.p_bar.init_p_bar(nth_process, self.n_iter, self.obj_func) |
|
184
|
|
|
|
SimonBlanke /
Hyperactive
