|
1
|
|
|
import time |
|
2
|
|
|
import numpy as np |
|
3
|
|
|
import pandas as pd |
|
4
|
|
|
from sklearn.datasets import load_breast_cancer |
|
5
|
|
|
from sklearn.model_selection import cross_val_score |
|
6
|
|
|
from sklearn.tree import DecisionTreeClassifier |
|
7
|
|
|
from gradient_free_optimizers import RandomSearchOptimizer |
|
8
|
|
|
from sklearn.ensemble import GradientBoostingClassifier |
|
9
|
|
|
|
|
10
|
|
|
|
|
11
|
|
|
def objective_function(para): |
|
12
|
|
|
score = -para["x1"] * para["x1"] |
|
13
|
|
|
return score |
|
14
|
|
|
|
|
15
|
|
|
|
|
16
|
|
|
search_space = { |
|
17
|
|
|
"x1": np.arange(0, 10, 1), |
|
18
|
|
|
} |
|
19
|
|
|
|
|
20
|
|
|
|
|
21
|
|
|
def test_memory_timeSave_0(): |
|
22
|
|
|
data = load_breast_cancer() |
|
23
|
|
|
X, y = data.data, data.target |
|
24
|
|
|
|
|
25
|
|
|
def objective_function(para): |
|
26
|
|
|
dtc = DecisionTreeClassifier( |
|
27
|
|
|
min_samples_split=para["min_samples_split"] |
|
28
|
|
|
) |
|
29
|
|
|
scores = cross_val_score(dtc, X, y, cv=5) |
|
30
|
|
|
|
|
31
|
|
|
return scores.mean() |
|
32
|
|
|
|
|
33
|
|
|
search_space = { |
|
34
|
|
|
"min_samples_split": np.arange(2, 20), |
|
35
|
|
|
} |
|
36
|
|
|
|
|
37
|
|
|
c_time1 = time.time() |
|
38
|
|
|
opt = RandomSearchOptimizer(search_space) |
|
39
|
|
|
opt.search(objective_function, n_iter=100) |
|
40
|
|
|
diff_time1 = time.time() - c_time1 |
|
41
|
|
|
|
|
42
|
|
|
c_time2 = time.time() |
|
43
|
|
|
opt = RandomSearchOptimizer(search_space) |
|
44
|
|
|
opt.search(objective_function, n_iter=100, memory=False) |
|
45
|
|
|
diff_time2 = time.time() - c_time2 |
|
46
|
|
|
|
|
47
|
|
|
assert diff_time1 < diff_time2 * 0.8 |
|
48
|
|
|
|
|
49
|
|
|
|
|
50
|
|
|
def test_memory_timeSave_1(): |
|
51
|
|
|
data = load_breast_cancer() |
|
52
|
|
|
X, y = data.data, data.target |
|
53
|
|
|
|
|
54
|
|
|
def objective_function(para): |
|
55
|
|
|
dtc = DecisionTreeClassifier(max_depth=para["max_depth"]) |
|
56
|
|
|
scores = cross_val_score(dtc, X, y, cv=5) |
|
57
|
|
|
|
|
58
|
|
|
return scores.mean() |
|
59
|
|
|
|
|
60
|
|
|
search_space = { |
|
61
|
|
|
"max_depth": np.arange(1, 101), |
|
62
|
|
|
} |
|
63
|
|
|
|
|
64
|
|
|
results = pd.DataFrame(np.arange(1, 101), columns=["max_depth"]) |
|
65
|
|
|
results["score"] = 0 |
|
66
|
|
|
|
|
67
|
|
|
c_time1 = time.time() |
|
68
|
|
|
opt = RandomSearchOptimizer(search_space) |
|
69
|
|
|
opt.search(objective_function, n_iter=300, memory_warm_start=results) |
|
70
|
|
|
diff_time1 = time.time() - c_time1 |
|
71
|
|
|
|
|
72
|
|
|
assert diff_time1 < 1 |
|
73
|
|
|
|
|
74
|
|
|
|
|
75
|
|
|
def test_memory_warm_start(): |
|
76
|
|
|
data = load_breast_cancer() |
|
77
|
|
|
X, y = data.data, data.target |
|
78
|
|
|
|
|
79
|
|
|
def objective_function(para): |
|
80
|
|
|
dtc = DecisionTreeClassifier( |
|
81
|
|
|
max_depth=para["max_depth"], |
|
82
|
|
|
min_samples_split=para["min_samples_split"], |
|
83
|
|
|
) |
|
84
|
|
|
scores = cross_val_score(dtc, X, y, cv=5) |
|
85
|
|
|
|
|
86
|
|
|
return scores.mean() |
|
87
|
|
|
|
|
88
|
|
|
search_space = { |
|
89
|
|
|
"max_depth": np.arange(1, 10), |
|
90
|
|
|
"min_samples_split": np.arange(2, 20), |
|
91
|
|
|
} |
|
92
|
|
|
|
|
93
|
|
|
c_time1 = time.time() |
|
94
|
|
|
opt0 = RandomSearchOptimizer(search_space) |
|
95
|
|
|
opt0.search(objective_function, n_iter=300) |
|
96
|
|
|
diff_time1 = time.time() - c_time1 |
|
97
|
|
|
|
|
98
|
|
|
c_time2 = time.time() |
|
99
|
|
|
opt1 = RandomSearchOptimizer(search_space) |
|
100
|
|
|
opt1.search(objective_function, n_iter=300, memory_warm_start=opt0.results) |
|
101
|
|
|
diff_time2 = time.time() - c_time2 |
|
102
|
|
|
|
|
103
|
|
|
print("\n opt0.results \n", opt0.results) |
|
104
|
|
|
|
|
105
|
|
|
assert diff_time2 < diff_time1 * 0.5 |
|
106
|
|
|
|
|
107
|
|
|
|
|
108
|
|
|
def test_memory_warm_start_manual(): |
|
109
|
|
|
data = load_breast_cancer() |
|
110
|
|
|
X, y = data.data, data.target |
|
111
|
|
|
|
|
112
|
|
|
def objective_function(para): |
|
113
|
|
|
dtc = GradientBoostingClassifier(n_estimators=para["n_estimators"],) |
|
114
|
|
|
scores = cross_val_score(dtc, X, y, cv=5) |
|
115
|
|
|
|
|
116
|
|
|
return scores.mean() |
|
117
|
|
|
|
|
118
|
|
|
search_space = { |
|
119
|
|
|
"n_estimators": np.arange(500, 502), |
|
120
|
|
|
} |
|
121
|
|
|
|
|
122
|
|
|
memory_warm_start = pd.DataFrame( |
|
123
|
|
|
[[500, 0.9], [501, 0.91]], columns=["n_estimators", "score"] |
|
124
|
|
|
) |
|
125
|
|
|
|
|
126
|
|
|
c_time = time.time() |
|
127
|
|
|
opt = RandomSearchOptimizer(search_space) |
|
128
|
|
|
opt.search( |
|
129
|
|
|
objective_function, n_iter=10, memory_warm_start=memory_warm_start |
|
130
|
|
|
) |
|
131
|
|
|
diff_time = time.time() - c_time |
|
132
|
|
|
|
|
133
|
|
|
assert diff_time < 1 |
|
134
|
|
|
|
|
135
|
|
|
|
SimonBlanke /
Gradient-Free-Optimizers
