|
1
|
|
|
import numpy as np |
|
2
|
|
|
import random |
|
3
|
|
|
from sklearn.model_selection import cross_val_score |
|
4
|
|
|
from sklearn.tree import DecisionTreeClassifier |
|
5
|
|
|
from sklearn.datasets import load_breast_cancer |
|
6
|
|
|
from hyperactive import Hyperactive |
|
7
|
|
|
|
|
8
|
|
|
data = load_breast_cancer() |
|
9
|
|
|
X, y = data.data, data.target |
|
10
|
|
|
|
|
11
|
|
|
X_list, y_list = [X], [y] |
|
12
|
|
|
|
|
13
|
|
|
def data_aug(X, y, sample_multi=5, feature_multi=5): |
|
14
|
|
|
X_list, y_list = [], [] |
|
15
|
|
|
|
|
16
|
|
|
n_samples = X.shape[0] |
|
17
|
|
|
n_features = X.shape[1] |
|
18
|
|
|
|
|
19
|
|
|
for sample in range(1, sample_multi+1): |
|
20
|
|
|
idx_sample = np.random.randint(n_samples, size=int(n_samples / sample)) |
|
21
|
|
|
|
|
22
|
|
|
for feature in range(1, feature_multi+1): |
|
23
|
|
|
idx_feature = np.random.randint(n_features, size=int(n_features / feature)) |
|
24
|
|
|
|
|
25
|
|
|
X_temp_ = X[idx_sample, :] |
|
26
|
|
|
X_temp = X_temp_[:, idx_feature] |
|
27
|
|
|
y_temp = y[idx_sample] |
|
28
|
|
|
|
|
29
|
|
|
X_list.append(X_temp) |
|
30
|
|
|
y_list.append(y_temp) |
|
31
|
|
|
|
|
32
|
|
|
return X_list, y_list |
|
33
|
|
|
|
|
34
|
|
|
def model(para, X, y): |
|
35
|
|
|
model = DecisionTreeClassifier( |
|
36
|
|
|
max_depth=para["max_depth"], |
|
37
|
|
|
min_samples_split=para["min_samples_split"], |
|
38
|
|
|
min_samples_leaf=para["min_samples_leaf"], |
|
39
|
|
|
) |
|
40
|
|
|
scores = cross_val_score(model, X, y, cv=3) |
|
41
|
|
|
|
|
42
|
|
|
return scores.mean() |
|
43
|
|
|
|
|
44
|
|
|
search_config_model = { |
|
45
|
|
|
model: { |
|
46
|
|
|
"max_depth": range(2, 50), |
|
47
|
|
|
"min_samples_split": range(2, 50), |
|
48
|
|
|
"min_samples_leaf": range(1, 50), |
|
49
|
|
|
} |
|
50
|
|
|
} |
|
51
|
|
|
|
|
52
|
|
|
|
|
53
|
|
|
def meta_opt(para, X_list, y_list): |
|
54
|
|
|
scores = [] |
|
55
|
|
|
|
|
56
|
|
|
for X, y in zip(X_list, y_list): |
|
57
|
|
|
X_list, y_list = data_aug(X, y, sample_multi=3, feature_multi=3) |
|
58
|
|
|
|
|
59
|
|
|
for X, y in zip(X_list, y_list): |
|
60
|
|
|
|
|
61
|
|
|
for n_iter in [10, 25, 50, 100]: |
|
62
|
|
|
opt = Hyperactive( |
|
63
|
|
|
search_config_model, |
|
64
|
|
|
optimizer={ |
|
65
|
|
|
"ParticleSwarm": {"inertia": para["inertia"], "cognitive_weight": para["cognitive_weight"], "social_weight": para["social_weight"]} |
|
66
|
|
|
}, |
|
67
|
|
|
n_iter=n_iter, |
|
68
|
|
|
verbosity=None, |
|
69
|
|
|
) |
|
70
|
|
|
opt.search(X, y) |
|
71
|
|
|
score = opt.score_best |
|
72
|
|
|
scores.append(score) |
|
73
|
|
|
|
|
74
|
|
|
return np.array(scores).mean() |
|
75
|
|
|
|
|
76
|
|
|
|
|
77
|
|
|
search_config_meta = { |
|
78
|
|
|
meta_opt: { |
|
79
|
|
|
"inertia": np.arange(0, 1, 0.01), |
|
80
|
|
|
"cognitive_weight": np.arange(0, 1, 0.01), |
|
81
|
|
|
"social_weight": np.arange(0, 1, 0.01), |
|
82
|
|
|
} |
|
83
|
|
|
} |
|
84
|
|
|
|
|
85
|
|
|
opt = Hyperactive(search_config_meta, optimizer="Bayesian", n_iter=30) |
|
86
|
|
|
opt.search(X_list, y_list) |
|
87
|
|
|
|
SimonBlanke /
Hyperactive
