def test_multi_init(self, gps):
"""Test initialization with multiple GPs"""
gp1, gp2 = gps
opt = GaussianProcessOptimization([gp1, gp2],
fmin=0,
beta=2,
num_contexts=1,
threshold=0,
scaling='auto')
# Check scaling
assert_allclose(opt.scaling, np.array([np.sqrt(2), np.sqrt(4)]))
def test_contexts(self):
"""Test contexts and adding data."""
kernel1 = GPy.kern.RBF(2, variance=2)
kernel2 = GPy.kern.Matern32(2, variance=4)
gp1 = GPy.models.GPRegression(np.array([[0, 0]]),
np.array([[5]]),
kernel=kernel1)
gp2 = GPy.models.GPRegression(np.array([[0, 0]]),
np.array([[6]]),
kernel=kernel2)
opt = GaussianProcessOptimization([gp1, gp2],
fmin=[0, 0],
num_contexts=1)
opt.add_new_data_point(1, [3, 4], context=2)
assert_allclose(opt.x, np.array([[0, 0], [1, 2]]))
assert_allclose(opt.y, np.array([[5, 6], [3, 4]]))
for i, gp in enumerate(opt.gps):
assert_allclose(gp.X, opt.x)
assert_allclose(gp.Y[:, 0], opt.y[:, i])
def test_scaling(self, gps):
"""Test the scaling argument."""
gp1, gp2 = gps
pytest.raises(ValueError,
GaussianProcessOptimization, [gp1, gp2],
2,
scaling=[5])
opt = GaussianProcessOptimization([gp1, gp2],
fmin=[1, 0],
beta=2,
num_contexts=1,
threshold=0,
scaling=[1, 2])
assert_allclose(opt.scaling, np.array([1, 2]))
def test_init(self, gps):
"""Test the initialization and beta functions."""
gp1, gp2 = gps
opt = GaussianProcessOptimization(gp1,
fmin=0,
beta=2,
num_contexts=1,
threshold=0,
scaling='auto')
assert opt.beta(0) == 2
opt = GaussianProcessOptimization(gp1,
fmin=[0],
beta=lambda x: 5,
num_contexts=1,
threshold=0,
scaling='auto')
assert opt.beta(10) == 5
def test_data_adding(self, gps):
"""Test adding data points."""
gp1, gp2 = gps
# Test simple 1D case
gp1.set_XY(np.array([[0.]]), np.array([[1.]]))
opt = GaussianProcessOptimization(gp1, 0)
opt.add_new_data_point(2, 3)
x, y = opt.data
assert_allclose(x, np.array([[0], [2]]))
assert_allclose(y, np.array([[1], [3]]))
# Test 2D case
gp1.set_XY(np.array([[0.]]), np.array([[1.]]))
gp2.set_XY(np.array([[0.]]), np.array([[11.]]))
opt = GaussianProcessOptimization([gp1, gp2], [0, 1])
opt.add_new_data_point(2, [2, 3])
x, y = opt.data
assert_allclose(x, np.array([[0], [2]]))
assert_allclose(y, np.array([[1, 11], [2, 3]]))
# Test adding NAN data
opt.add_new_data_point(3, [2, np.nan])
assert_allclose(opt.x, np.array([[0], [2], [3]]))
assert_allclose(opt.y, np.array([[1, 11], [2, 3], [2, np.nan]]))
for i, gp in enumerate(opt.gps):
not_nan = ~np.isnan(opt.y[:, i])
assert_allclose(gp.X, opt.x[not_nan, :])
assert_allclose(gp.Y[:, 0], opt.y[not_nan, i])
# Test removing data
opt.remove_last_data_point()
assert_allclose(opt.x, np.array([[0], [2]]))
assert_allclose(opt.y, np.array([[1, 11], [2, 3]]))
for i, gp in enumerate(opt.gps):
not_nan = ~np.isnan(opt.y[:, i])
assert_allclose(gp.X, opt.x[not_nan, :])
assert_allclose(gp.Y[:, 0], opt.y[not_nan, i])