|
1
|
|
|
# Author: Simon Blanke |
|
2
|
|
|
# Email: [email protected] |
|
3
|
|
|
# License: MIT License |
|
4
|
|
|
|
|
5
|
|
|
|
|
6
|
|
|
import multiprocessing |
|
7
|
|
|
from tqdm import tqdm |
|
8
|
|
|
|
|
9
|
|
|
from .optimizers import RandomSearchOptimizer |
|
10
|
|
|
from .run_search import run_search |
|
11
|
|
|
|
|
12
|
|
|
|
|
13
|
|
|
def set_n_jobs(n_jobs): |
|
14
|
|
|
"""Sets the number of jobs to run in parallel""" |
|
15
|
|
|
num_cores = multiprocessing.cpu_count() |
|
16
|
|
|
if n_jobs == -1 or n_jobs > num_cores: |
|
17
|
|
|
return num_cores |
|
18
|
|
|
else: |
|
19
|
|
|
return n_jobs |
|
20
|
|
|
|
|
21
|
|
|
|
|
22
|
|
|
class HyperactiveResults: |
|
23
|
|
|
def __init__(*args, **kwargs): |
|
24
|
|
|
pass |
|
25
|
|
|
|
|
26
|
|
|
def _sort_results_objFunc(self, objective_function): |
|
27
|
|
|
import numpy as np |
|
28
|
|
|
import pandas as pd |
|
29
|
|
|
|
|
30
|
|
|
best_score = -np.inf |
|
31
|
|
|
best_para = None |
|
32
|
|
|
results_list = [] |
|
33
|
|
|
|
|
34
|
|
|
for results in self.results_list: |
|
35
|
|
|
nth_process = results["nth_process"] |
|
36
|
|
|
|
|
37
|
|
|
process_infos = self.process_infos[nth_process] |
|
38
|
|
|
objective_function_ = process_infos["objective_function"] |
|
39
|
|
|
|
|
40
|
|
|
if objective_function_ != objective_function: |
|
41
|
|
|
continue |
|
42
|
|
|
|
|
43
|
|
|
if results["best_score"] > best_score: |
|
44
|
|
|
best_score = results["best_score"] |
|
45
|
|
|
best_para = results["best_para"] |
|
46
|
|
|
|
|
47
|
|
|
results_list.append(results["results"]) |
|
48
|
|
|
|
|
49
|
|
|
results = pd.concat(results_list) |
|
50
|
|
|
|
|
51
|
|
|
self.objFunc2results[objective_function] = { |
|
52
|
|
|
"best_para": best_para, |
|
53
|
|
|
"best_score": best_score, |
|
54
|
|
|
"results": results, |
|
55
|
|
|
} |
|
56
|
|
|
|
|
57
|
|
|
def _sort_results_search_id(self, search_id): |
|
58
|
|
|
for results in self.results_list: |
|
59
|
|
|
nth_process = results["nth_process"] |
|
60
|
|
|
search_id_ = self.process_infos[nth_process]["search_id"] |
|
61
|
|
|
|
|
62
|
|
|
if search_id_ != search_id: |
|
63
|
|
|
continue |
|
64
|
|
|
|
|
65
|
|
|
best_score = results["best_score"] |
|
66
|
|
|
best_para = results["best_para"] |
|
67
|
|
|
results = results["results"] |
|
68
|
|
|
|
|
69
|
|
|
self.search_id2results[search_id] = { |
|
70
|
|
|
"best_para": best_para, |
|
71
|
|
|
"best_score": best_score, |
|
72
|
|
|
"results": results, |
|
73
|
|
|
} |
|
74
|
|
|
|
|
75
|
|
|
def _get_one_result(self, id_, result_name): |
|
76
|
|
|
if isinstance(id_, str): |
|
77
|
|
|
if id_ not in self.search_id2results: |
|
78
|
|
|
self._sort_results_search_id(id_) |
|
79
|
|
|
|
|
80
|
|
|
return self.search_id2results[id_][result_name] |
|
81
|
|
|
|
|
82
|
|
|
else: |
|
83
|
|
|
if id_ not in self.objFunc2results: |
|
84
|
|
|
self._sort_results_objFunc(id_) |
|
85
|
|
|
|
|
86
|
|
|
return self.objFunc2results[id_][result_name] |
|
87
|
|
|
|
|
88
|
|
|
def best_para(self, id_): |
|
89
|
|
|
return self._get_one_result(id_, "best_para") |
|
90
|
|
|
|
|
91
|
|
|
def best_score(self, id_): |
|
92
|
|
|
return self._get_one_result(id_, "best_score") |
|
93
|
|
|
|
|
94
|
|
|
def results(self, id_): |
|
95
|
|
|
return self._get_one_result(id_, "results") |
|
96
|
|
|
|
|
97
|
|
|
|
|
98
|
|
|
class HyperactiveLongTermMemory: |
|
99
|
|
|
def __init__(*args, **kwargs): |
|
100
|
|
|
pass |
|
101
|
|
|
|
|
102
|
|
|
def _load_ltm(self, memory_warm_start): |
|
103
|
|
|
if self.long_term_memory is not None: |
|
104
|
|
|
return self.long_term_memory.load() |
|
105
|
|
|
else: |
|
106
|
|
|
return memory_warm_start |
|
107
|
|
|
|
|
108
|
|
|
def _save_ltm(self): |
|
109
|
|
|
for nth_process in self.process_infos.keys(): |
|
110
|
|
|
long_term_memory = self.process_infos[nth_process]["long_term_memory"] |
|
111
|
|
|
objective_function = self.process_infos[nth_process]["objective_function"] |
|
112
|
|
|
memory_results = self.results_list[nth_process]["memory_values_df"] |
|
113
|
|
|
|
|
114
|
|
|
if long_term_memory is not None: |
|
115
|
|
|
long_term_memory.save(memory_results, objective_function) |
|
116
|
|
|
|
|
117
|
|
|
|
|
118
|
|
|
class Hyperactive(HyperactiveResults, HyperactiveLongTermMemory): |
|
119
|
|
|
def __init__( |
|
120
|
|
|
self, |
|
121
|
|
|
verbosity=["progress_bar", "print_results", "print_times"], |
|
122
|
|
|
distribution={ |
|
123
|
|
|
"multiprocessing": { |
|
124
|
|
|
"initializer": tqdm.set_lock, |
|
125
|
|
|
"initargs": (tqdm.get_lock(),), |
|
126
|
|
|
} |
|
127
|
|
|
}, |
|
128
|
|
|
): |
|
129
|
|
|
super().__init__() |
|
130
|
|
|
if verbosity is False: |
|
131
|
|
|
verbosity = [] |
|
132
|
|
|
|
|
133
|
|
|
self.verbosity = verbosity |
|
134
|
|
|
self.distribution = distribution |
|
135
|
|
|
self.search_ids = [] |
|
136
|
|
|
|
|
137
|
|
|
self.process_infos = {} |
|
138
|
|
|
|
|
139
|
|
|
self.objFunc2results = {} |
|
140
|
|
|
self.search_id2results = {} |
|
141
|
|
|
|
|
142
|
|
|
def _add_search_processes( |
|
143
|
|
|
self, |
|
144
|
|
|
random_state, |
|
145
|
|
|
objective_function, |
|
146
|
|
|
search_space, |
|
147
|
|
|
optimizer, |
|
148
|
|
|
n_iter, |
|
149
|
|
|
n_jobs, |
|
150
|
|
|
max_score, |
|
151
|
|
|
memory, |
|
152
|
|
|
memory_warm_start, |
|
153
|
|
|
long_term_memory, |
|
154
|
|
|
search_id, |
|
155
|
|
|
): |
|
156
|
|
|
for nth_job in range(set_n_jobs(n_jobs)): |
|
157
|
|
|
nth_process = len(self.process_infos) |
|
158
|
|
|
|
|
159
|
|
|
self.process_infos[nth_process] = { |
|
160
|
|
|
"random_state": random_state, |
|
161
|
|
|
"verbosity": self.verbosity, |
|
162
|
|
|
"nth_process": nth_process, |
|
163
|
|
|
"objective_function": objective_function, |
|
164
|
|
|
"search_space": search_space, |
|
165
|
|
|
"optimizer": optimizer, |
|
166
|
|
|
"n_iter": n_iter, |
|
167
|
|
|
"max_score": max_score, |
|
168
|
|
|
"memory": memory, |
|
169
|
|
|
"memory_warm_start": memory_warm_start, |
|
170
|
|
|
"long_term_memory": long_term_memory, |
|
171
|
|
|
"search_id": search_id, |
|
172
|
|
|
} |
|
173
|
|
|
|
|
174
|
|
|
def add_search( |
|
175
|
|
|
self, |
|
176
|
|
|
objective_function, |
|
177
|
|
|
search_space, |
|
178
|
|
|
n_iter, |
|
179
|
|
|
search_id=None, |
|
180
|
|
|
optimizer="default", |
|
181
|
|
|
n_jobs=1, |
|
182
|
|
|
initialize={"grid": 4, "random": 2, "vertices": 4}, |
|
183
|
|
|
max_score=None, |
|
184
|
|
|
random_state=None, |
|
185
|
|
|
memory=True, |
|
186
|
|
|
memory_warm_start=None, |
|
187
|
|
|
long_term_memory=None, |
|
188
|
|
|
): |
|
189
|
|
|
self.long_term_memory = long_term_memory |
|
190
|
|
|
|
|
191
|
|
|
if isinstance(optimizer, str): |
|
192
|
|
|
if optimizer == "default": |
|
193
|
|
|
optimizer = RandomSearchOptimizer() |
|
194
|
|
|
optimizer.init(search_space, initialize) |
|
195
|
|
|
|
|
196
|
|
|
if search_id is not None: |
|
197
|
|
|
search_id = search_id |
|
198
|
|
|
self.search_ids.append(search_id) |
|
199
|
|
|
else: |
|
200
|
|
|
search_id = str(len(self.search_ids)) |
|
201
|
|
|
self.search_ids.append(search_id) |
|
202
|
|
|
|
|
203
|
|
|
memory_warm_start = self._load_ltm(memory_warm_start) |
|
204
|
|
|
|
|
205
|
|
|
self._add_search_processes( |
|
206
|
|
|
random_state, |
|
207
|
|
|
objective_function, |
|
208
|
|
|
search_space, |
|
209
|
|
|
optimizer, |
|
210
|
|
|
n_iter, |
|
211
|
|
|
n_jobs, |
|
212
|
|
|
max_score, |
|
213
|
|
|
memory, |
|
214
|
|
|
memory_warm_start, |
|
215
|
|
|
long_term_memory, |
|
216
|
|
|
search_id, |
|
217
|
|
|
) |
|
218
|
|
|
|
|
219
|
|
|
def run( |
|
220
|
|
|
self, |
|
221
|
|
|
max_time=None, |
|
222
|
|
|
): |
|
223
|
|
|
for nth_process in self.process_infos.keys(): |
|
224
|
|
|
self.process_infos[nth_process]["max_time"] = max_time |
|
225
|
|
|
|
|
226
|
|
|
self.results_list = run_search(self.process_infos, self.distribution) |
|
227
|
|
|
|
|
228
|
|
|
self._save_ltm() |
|
229
|
|
|
|
SimonBlanke /
Hyperactive
