|
1
|
|
|
from tqdm import tqdm |
|
2
|
|
|
import numpy as np |
|
3
|
|
|
import pandas as pd |
|
4
|
|
|
|
|
5
|
|
|
from gradient_free_optimizers import ( |
|
6
|
|
|
HillClimbingOptimizer, |
|
7
|
|
|
StochasticHillClimbingOptimizer, |
|
8
|
|
|
TabuOptimizer, |
|
9
|
|
|
RandomSearchOptimizer, |
|
10
|
|
|
RandomRestartHillClimbingOptimizer, |
|
11
|
|
|
RandomAnnealingOptimizer, |
|
12
|
|
|
SimulatedAnnealingOptimizer, |
|
13
|
|
|
ParallelTemperingOptimizer, |
|
14
|
|
|
ParticleSwarmOptimizer, |
|
15
|
|
|
EvolutionStrategyOptimizer, |
|
16
|
|
|
BayesianOptimizer, |
|
17
|
|
|
TreeStructuredParzenEstimators, |
|
18
|
|
|
DecisionTreeOptimizer, |
|
19
|
|
|
) |
|
20
|
|
|
|
|
21
|
|
|
|
|
22
|
|
|
optimizer_dict = { |
|
23
|
|
|
"HillClimbing": HillClimbingOptimizer, |
|
24
|
|
|
"StochasticHillClimbingOptimizer": StochasticHillClimbingOptimizer, |
|
25
|
|
|
} |
|
26
|
|
|
|
|
27
|
|
|
|
|
28
|
|
|
def create_convergence_data(optimizer_key): |
|
29
|
|
|
def objective_function(para): |
|
30
|
|
|
score = -para["x1"] * para["x1"] |
|
31
|
|
|
return score |
|
32
|
|
|
|
|
33
|
|
|
search_space = {"x1": np.arange(-100, 101, 1)} |
|
34
|
|
|
initialize = {"vertices": 2} |
|
35
|
|
|
|
|
36
|
|
|
n_opts = 30 |
|
37
|
|
|
n_iter = 100 |
|
38
|
|
|
|
|
39
|
|
|
scores_list = [] |
|
40
|
|
|
for rnd_st in tqdm(range(n_opts)): |
|
41
|
|
|
opt = optimizer_dict[optimizer_key](search_space) |
|
42
|
|
|
opt.search( |
|
43
|
|
|
objective_function, |
|
44
|
|
|
n_iter=n_iter, |
|
45
|
|
|
random_state=rnd_st, |
|
46
|
|
|
memory=False, |
|
47
|
|
|
verbosity=False, |
|
48
|
|
|
initialize=initialize, |
|
49
|
|
|
) |
|
50
|
|
|
|
|
51
|
|
|
scores_list.append(opt.results["score"]) |
|
52
|
|
|
|
|
53
|
|
|
convergence_data = pd.concat(scores_list, axis=1) |
|
54
|
|
|
convergence_data.to_csv( |
|
55
|
|
|
"./data/" + optimizer_key + "_convergence_data", index=False |
|
56
|
|
|
) |
|
57
|
|
|
|
|
58
|
|
|
|
|
59
|
|
|
for opt_key in optimizer_dict.keys(): |
|
60
|
|
|
create_convergence_data(opt_key) |
|
61
|
|
|
|
SimonBlanke /
Gradient-Free-Optimizers
