1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
|
import numpy as np
import pytest
from numpy.testing import assert_array_almost_equal, assert_array_equal, assert_raises
from scipy import optimize
from sklearn.multioutput import MultiOutputRegressor
from skopt.acquisition import (
_gaussian_acquisition,
gaussian_acquisition_1D,
gaussian_ei,
gaussian_lcb,
gaussian_pi,
)
from skopt.learning import GaussianProcessRegressor
from skopt.learning.gaussian_process.kernels import Matern, WhiteKernel
from skopt.space import Space
from skopt.utils import cook_estimator
class ConstSurrogate:
def predict(self, X, return_std=True):
X = np.array(X)
return np.zeros(X.shape[0]), np.ones(X.shape[0])
# This is used to test that given constant acquisition values at
# different points, acquisition functions "EIps" and "PIps"
# prefer candidate points that take lesser time.
# The second estimator mimics the GP regressor that is fit on
# the log of the input.
class ConstantGPRSurrogate:
def __init__(self, space):
self.space = space
def fit(self, X, y):
"""The first estimator returns a constant value.
The second estimator is a gaussian process regressor that models
the logarithm of the time.
"""
X = np.array(X)
gpr = cook_estimator("GP", self.space, normalize_y=False)
gpr.fit(X, np.log(np.ravel(X)))
self.estimators_ = []
self.estimators_.append(ConstSurrogate())
self.estimators_.append(gpr)
return self
@pytest.mark.fast_test
def test_acquisition_ei_correctness():
# check that it works with a vector as well
X = 10 * np.ones((4, 2))
ei = gaussian_ei(X, ConstSurrogate(), -0.5, xi=0.0)
assert_array_almost_equal(ei, [0.1977966] * 4)
@pytest.mark.fast_test
def test_acquisition_pi_correctness():
# check that it works with a vector as well
X = 10 * np.ones((4, 2))
pi = gaussian_pi(X, ConstSurrogate(), -0.5, xi=0.0)
assert_array_almost_equal(pi, [0.308538] * 4)
@pytest.mark.fast_test
def test_acquisition_variance_correctness():
# check that it works with a vector as well
X = 10 * np.ones((4, 2))
var = gaussian_lcb(X, ConstSurrogate(), kappa='inf')
assert_array_almost_equal(var, [-1.0] * 4)
@pytest.mark.fast_test
def test_acquisition_lcb_correctness():
# check that it works with a vector as well
X = 10 * np.ones((4, 2))
lcb = gaussian_lcb(X, ConstSurrogate(), kappa=0.3)
assert_array_almost_equal(lcb, [-0.3] * 4)
@pytest.mark.fast_test
def test_acquisition_api():
rng = np.random.RandomState(0)
X = rng.randn(10, 2)
y = rng.randn(10)
gpr = GaussianProcessRegressor()
gpr.fit(X, y)
for method in [gaussian_ei, gaussian_lcb, gaussian_pi]:
assert_array_equal(method(X, gpr).shape, 10)
assert_raises(ValueError, method, rng.rand(10), gpr)
def check_gradient_correctness(X_new, model, acq_func, y_opt):
def num_grad_func(x):
return gaussian_acquisition_1D(x, model, y_opt, acq_func=acq_func)[0]
analytic_grad = gaussian_acquisition_1D(X_new, model, y_opt, acq_func)[1]
num_grad = optimize.approx_fprime(X_new, num_grad_func, 1e-5)
assert_array_almost_equal(analytic_grad, num_grad, 3)
@pytest.mark.fast_test
def test_acquisition_gradient():
rng = np.random.RandomState(0)
X = rng.randn(20, 5)
y = rng.randn(20)
X_new = rng.randn(5)
mat = Matern()
wk = WhiteKernel()
gpr = GaussianProcessRegressor(kernel=mat + wk)
gpr.fit(X, y)
for acq_func in ["LCB", "PI", "EI"]:
check_gradient_correctness(X_new, gpr, acq_func, np.max(y))
@pytest.mark.fast_test
def test_acquisition_gradient_cookbook():
rng = np.random.RandomState(0)
X = rng.randn(20, 5)
y = rng.randn(20)
X_new = rng.randn(5)
gpr = cook_estimator("GP", Space(((-5.0, 5.0),)), random_state=0)
gpr.fit(X, y)
for acq_func in ["LCB", "PI", "EI"]:
check_gradient_correctness(X_new, gpr, acq_func, np.max(y))
@pytest.mark.fast_test
@pytest.mark.parametrize("acq_func", ["EIps", "PIps"])
def test_acquisition_per_second(acq_func):
X = np.reshape(np.linspace(4.0, 8.0, 10), (-1, 1))
y = np.vstack((np.ones(10), np.ravel(np.log(X)))).T
cgpr = ConstantGPRSurrogate(Space(((1.0, 9.0),)))
cgpr.fit(X, y)
X_pred = np.reshape(np.linspace(1.0, 11.0, 20), (-1, 1))
indices = np.arange(6)
vals = _gaussian_acquisition(X_pred, cgpr, y_opt=1.0, acq_func=acq_func)
for fast, slow in zip(indices[:-1], indices[1:]):
assert vals[slow] > vals[fast]
acq_wo_time = _gaussian_acquisition(
X, cgpr.estimators_[0], y_opt=1.2, acq_func=acq_func[:2]
)
acq_with_time = _gaussian_acquisition(X, cgpr, y_opt=1.2, acq_func=acq_func)
assert_array_almost_equal(acq_wo_time / acq_with_time, np.ravel(X), 2)
def test_gaussian_acquisition_check_inputs():
model = ConstantGPRSurrogate(Space(((1.0, 9.0),)))
with pytest.raises(ValueError) as err:
_gaussian_acquisition(np.arange(1, 5), model)
assert "it must be 2-dimensional" in err.value.args[0]
@pytest.mark.fast_test
@pytest.mark.parametrize("acq_func", ["EIps", "PIps"])
def test_acquisition_per_second_gradient(acq_func):
rng = np.random.RandomState(0)
X = rng.randn(20, 10)
# Make the second component large, so that mean_grad and std_grad
# do not become zero.
y = np.vstack((X[:, 0], np.abs(X[:, 0]) ** 3)).T
for X_new in [rng.randn(10), rng.randn(10)]:
gpr = cook_estimator("GP", Space(((-5.0, 5.0),)), random_state=0)
mor = MultiOutputRegressor(gpr)
mor.fit(X, y)
check_gradient_correctness(X_new, mor, acq_func, 1.5)
|
