1
|
|
|
"""NSGA-II multi-objective optimizer.""" |
2
|
|
|
# copyright: hyperactive developers, MIT License (see LICENSE file) |
3
|
|
|
|
4
|
|
|
from .._adapters._base_optuna_adapter import _BaseOptunaAdapter |
5
|
|
|
|
6
|
|
|
|
7
|
|
|
class NSGAIIOptimizer(_BaseOptunaAdapter): |
8
|
|
|
"""NSGA-II multi-objective optimizer. |
9
|
|
|
|
10
|
|
|
Parameters |
11
|
|
|
---------- |
12
|
|
|
param_space : dict[str, tuple or list or optuna distributions] |
13
|
|
|
The search space to explore. Dictionary with parameter names |
14
|
|
|
as keys and either tuples/lists of (low, high) or |
15
|
|
|
optuna distribution objects as values. |
16
|
|
|
n_trials : int, default=100 |
17
|
|
|
Number of optimization trials. |
18
|
|
|
initialize : dict[str, int], default=None |
19
|
|
|
The method to generate initial positions. A dictionary with |
20
|
|
|
the following key literals and the corresponding value type: |
21
|
|
|
{"grid": int, "vertices": int, "random": int, "warm_start": list[dict]} |
22
|
|
|
random_state : None, int, default=None |
23
|
|
|
If None, create a new random state. If int, create a new random state |
24
|
|
|
seeded with the value. |
25
|
|
|
early_stopping : int, default=None |
26
|
|
|
Number of trials after which to stop if no improvement. |
27
|
|
|
max_score : float, default=None |
28
|
|
|
Maximum score threshold. Stop optimization when reached. |
29
|
|
|
population_size : int, default=50 |
30
|
|
|
Population size for NSGA-II. |
31
|
|
|
mutation_prob : float, default=0.1 |
32
|
|
|
Mutation probability for NSGA-II. |
33
|
|
|
crossover_prob : float, default=0.9 |
34
|
|
|
Crossover probability for NSGA-II. |
35
|
|
|
experiment : BaseExperiment, optional |
36
|
|
|
The experiment to optimize parameters for. |
37
|
|
|
Optional, can be passed later via ``set_params``. |
38
|
|
|
|
39
|
|
|
Examples |
40
|
|
|
-------- |
41
|
|
|
Basic usage of NSGAIIOptimizer with a scikit-learn experiment: |
42
|
|
|
|
43
|
|
|
>>> from hyperactive.experiment.integrations import SklearnCvExperiment |
44
|
|
|
>>> from hyperactive.opt.optuna import NSGAIIOptimizer |
45
|
|
|
>>> from sklearn.datasets import load_iris |
46
|
|
|
>>> from sklearn.svm import SVC |
47
|
|
|
>>> X, y = load_iris(return_X_y=True) |
48
|
|
|
>>> sklearn_exp = SklearnCvExperiment(estimator=SVC(), X=X, y=y) |
49
|
|
|
>>> param_space = { |
50
|
|
|
... "C": (0.01, 10), |
51
|
|
|
... "gamma": (0.0001, 10), |
52
|
|
|
... } |
53
|
|
|
>>> optimizer = NSGAIIOptimizer( |
54
|
|
|
... param_space=param_space, n_trials=50, experiment=sklearn_exp |
55
|
|
|
... ) |
56
|
|
|
>>> best_params = optimizer.run() |
57
|
|
|
""" |
58
|
|
|
|
59
|
|
|
_tags = { |
60
|
|
|
"info:name": "NSGA-II Optimizer", |
61
|
|
|
"info:local_vs_global": "global", |
62
|
|
|
"info:explore_vs_exploit": "mixed", |
63
|
|
|
"info:compute": "high", |
64
|
|
|
"python_dependencies": ["optuna"], |
65
|
|
|
} |
66
|
|
|
|
67
|
|
|
def __init__( |
68
|
|
|
self, |
69
|
|
|
param_space=None, |
70
|
|
|
n_trials=100, |
71
|
|
|
initialize=None, |
72
|
|
|
random_state=None, |
73
|
|
|
early_stopping=None, |
74
|
|
|
max_score=None, |
75
|
|
|
population_size=50, |
76
|
|
|
mutation_prob=0.1, |
77
|
|
|
crossover_prob=0.9, |
78
|
|
|
experiment=None, |
79
|
|
|
): |
80
|
|
|
self.population_size = population_size |
81
|
|
|
self.mutation_prob = mutation_prob |
82
|
|
|
self.crossover_prob = crossover_prob |
83
|
|
|
|
84
|
|
|
super().__init__( |
85
|
|
|
param_space=param_space, |
86
|
|
|
n_trials=n_trials, |
87
|
|
|
initialize=initialize, |
88
|
|
|
random_state=random_state, |
89
|
|
|
early_stopping=early_stopping, |
90
|
|
|
max_score=max_score, |
91
|
|
|
experiment=experiment, |
92
|
|
|
) |
93
|
|
|
|
94
|
|
|
def _get_optimizer(self): |
95
|
|
|
"""Get the NSGA-II optimizer. |
96
|
|
|
|
97
|
|
|
Returns |
98
|
|
|
------- |
99
|
|
|
optimizer |
100
|
|
|
The Optuna NSGAIIOptimizer instance |
101
|
|
|
""" |
102
|
|
|
import optuna |
103
|
|
|
|
104
|
|
|
optimizer_kwargs = { |
105
|
|
|
"population_size": self.population_size, |
106
|
|
|
"mutation_prob": self.mutation_prob, |
107
|
|
|
"crossover_prob": self.crossover_prob, |
108
|
|
|
} |
109
|
|
|
|
110
|
|
|
if self.random_state is not None: |
111
|
|
|
optimizer_kwargs["seed"] = self.random_state |
112
|
|
|
|
113
|
|
|
return optuna.samplers.NSGAIISampler(**optimizer_kwargs) |
114
|
|
|
|
115
|
|
View Code Duplication |
@classmethod |
|
|
|
|
116
|
|
|
def get_test_params(cls, parameter_set="default"): |
117
|
|
|
"""Return testing parameter settings for the optimizer.""" |
118
|
|
|
from sklearn.datasets import load_iris |
119
|
|
|
from sklearn.ensemble import RandomForestClassifier |
120
|
|
|
|
121
|
|
|
from hyperactive.experiment.integrations import SklearnCvExperiment |
122
|
|
|
|
123
|
|
|
# Test case 1: Basic single-objective (inherits from base) |
124
|
|
|
params = super().get_test_params(parameter_set) |
125
|
|
|
params[0].update( |
126
|
|
|
{ |
127
|
|
|
"population_size": 20, |
128
|
|
|
"mutation_prob": 0.2, |
129
|
|
|
"crossover_prob": 0.8, |
130
|
|
|
} |
131
|
|
|
) |
132
|
|
|
|
133
|
|
|
# Test case 2: Multi-objective with mixed parameter types |
134
|
|
|
X, y = load_iris(return_X_y=True) |
135
|
|
|
rf_exp = SklearnCvExperiment( |
136
|
|
|
estimator=RandomForestClassifier(random_state=42), X=X, y=y |
137
|
|
|
) |
138
|
|
|
|
139
|
|
|
mixed_param_space = { |
140
|
|
|
"n_estimators": (10, 50), # Continuous integer |
141
|
|
|
"max_depth": [3, 5, 7, None], # Mixed discrete/None |
142
|
|
|
"criterion": ["gini", "entropy"], # Categorical |
143
|
|
|
"min_samples_split": (2, 10), # Continuous integer |
144
|
|
|
"bootstrap": [True, False], # Boolean categorical |
145
|
|
|
} |
146
|
|
|
|
147
|
|
|
params.append( |
148
|
|
|
{ |
149
|
|
|
"param_space": mixed_param_space, |
150
|
|
|
"n_trials": 15, # Smaller for faster testing |
151
|
|
|
"experiment": rf_exp, |
152
|
|
|
"population_size": 8, # Smaller population for testing |
153
|
|
|
"mutation_prob": 0.1, |
154
|
|
|
"crossover_prob": 0.9, |
155
|
|
|
} |
156
|
|
|
) |
157
|
|
|
|
158
|
|
|
return params |
159
|
|
|
|