|
1
|
|
|
"""Grid search optimizer.""" |
|
2
|
|
|
|
|
3
|
|
|
from collections.abc import Sequence |
|
4
|
|
|
|
|
5
|
|
|
import numpy as np |
|
6
|
|
|
|
|
7
|
|
|
from sklearn.model_selection import ParameterGrid, ParameterSampler, check_cv |
|
8
|
|
|
|
|
9
|
|
|
from hyperactive.base import BaseOptimizer |
|
10
|
|
|
|
|
11
|
|
|
|
|
12
|
|
|
class GridSearch(BaseOptimizer): |
|
13
|
|
|
"""Grid search optimizer. |
|
14
|
|
|
|
|
15
|
|
|
Parameters |
|
16
|
|
|
---------- |
|
17
|
|
|
experiment : BaseExperiment, optional |
|
18
|
|
|
The experiment to optimize parameters for. |
|
19
|
|
|
Optional, can be passed later in ``add_search``. |
|
20
|
|
|
error_score : float, default=np.nan |
|
21
|
|
|
The score to assign if an error occurs during the evaluation of a parameter set. |
|
22
|
|
|
param_grid : dict[str, list] |
|
23
|
|
|
The search space to explore. A dictionary with parameter |
|
24
|
|
|
names as keys and a numpy array as values. |
|
25
|
|
|
|
|
26
|
|
|
Example |
|
27
|
|
|
------- |
|
28
|
|
|
Grid search applied to scikit-learn parameter tuning: |
|
29
|
|
|
|
|
30
|
|
|
1. defining the experiment to optimize: |
|
31
|
|
|
>>> from hyperactive.experiment import Experiment |
|
32
|
|
|
>>> from hyperactive.experiment.integrations import SklearnCvExperiment |
|
33
|
|
|
>>> from sklearn.datasets import load_iris |
|
34
|
|
|
>>> from sklearn.svm import SVC |
|
35
|
|
|
>>> |
|
36
|
|
|
>>> X, y = load_iris(return_X_y=True) |
|
37
|
|
|
>>> |
|
38
|
|
|
>>> sklearn_exp = SklearnCvExperiment( |
|
39
|
|
|
... estimator=SVC(), |
|
40
|
|
|
... X=X, |
|
41
|
|
|
... y=y, |
|
42
|
|
|
... ) |
|
43
|
|
|
|
|
44
|
|
|
2. setting up the grid search optimizer: |
|
45
|
|
|
>>> from hyperactive import GridSearch |
|
46
|
|
|
>>> param_grid = { |
|
47
|
|
|
... "C": [0.01, 0.1, 1, 10], |
|
48
|
|
|
... "gamma": [0.0001, 0.01, 0.1, 1, 10], |
|
49
|
|
|
... } |
|
50
|
|
|
... grid_search = GridSearch(sklearn_exp, param_grid=param_grid) |
|
51
|
|
|
|
|
52
|
|
|
3. running the grid search: |
|
53
|
|
|
>>> best_params = grid_search.run() |
|
54
|
|
|
|
|
55
|
|
|
Best parameters can also be accessed via the attributes: |
|
56
|
|
|
>>> best_params = grid_search.best_params_ |
|
57
|
|
|
""" |
|
58
|
|
|
|
|
59
|
|
|
def __init__( |
|
60
|
|
|
self, |
|
61
|
|
|
experiment=None, |
|
62
|
|
|
error_score=np.nan, |
|
63
|
|
|
param_grid=None, |
|
64
|
|
|
): |
|
65
|
|
|
self.experiment = experiment |
|
66
|
|
|
self.param_grid = param_grid |
|
67
|
|
|
self.error_score = error_score |
|
68
|
|
|
|
|
69
|
|
|
super().__init__() |
|
70
|
|
|
|
|
71
|
|
|
def _check_param_grid(self, param_grid): |
|
72
|
|
|
"""_check_param_grid from sklearn 1.0.2, before it was removed.""" |
|
73
|
|
|
if hasattr(param_grid, "items"): |
|
74
|
|
|
param_grid = [param_grid] |
|
75
|
|
|
|
|
76
|
|
|
for p in param_grid: |
|
77
|
|
|
for name, v in p.items(): |
|
78
|
|
|
if isinstance(v, np.ndarray) and v.ndim > 1: |
|
79
|
|
|
raise ValueError("Parameter array should be one-dimensional.") |
|
80
|
|
|
|
|
81
|
|
|
if isinstance(v, str) or not isinstance(v, (np.ndarray, Sequence)): |
|
82
|
|
|
raise ValueError( |
|
83
|
|
|
f"Parameter grid for parameter ({name}) needs to" |
|
84
|
|
|
f" be a list or numpy array, but got ({type(v)})." |
|
85
|
|
|
" Single values need to be wrapped in a list" |
|
86
|
|
|
" with one element." |
|
87
|
|
|
) |
|
88
|
|
|
|
|
89
|
|
|
if len(v) == 0: |
|
90
|
|
|
raise ValueError( |
|
91
|
|
|
f"Parameter values for parameter ({name}) need " |
|
92
|
|
|
"to be a non-empty sequence." |
|
93
|
|
|
) |
|
94
|
|
|
|
|
95
|
|
|
def _run(self, experiment, param_grid, error_score): |
|
96
|
|
|
"""Run the optimization search process.""" |
|
97
|
|
|
self._check_param_grid(param_grid) |
|
98
|
|
|
candidate_params = list(ParameterGrid(param_grid)) |
|
99
|
|
|
|
|
100
|
|
|
scores = [] |
|
101
|
|
|
for candidate_param in candidate_params: |
|
102
|
|
|
try: |
|
103
|
|
|
score = experiment(**candidate_param) |
|
104
|
|
|
except Exception: # noqa: B904 |
|
105
|
|
|
# Catch all exceptions and assign error_score |
|
106
|
|
|
score = error_score |
|
107
|
|
|
scores.append(score) |
|
108
|
|
|
|
|
109
|
|
|
best_index = np.argmin(scores) |
|
110
|
|
|
best_params = candidate_params[best_index] |
|
111
|
|
|
|
|
112
|
|
|
self.best_index_ = best_index |
|
113
|
|
|
self.best_params_ = best_params |
|
114
|
|
|
self.best_score_ = scores[best_index] |
|
115
|
|
|
|
|
116
|
|
|
return best_params |
|
117
|
|
|
|
SimonBlanke /
Hyperactive
