def test_predict_1d_fit(self):
x = np.linspace(0, 2 * np.pi, num=8)
u = np.linspace(0, 2 * np.pi, num=50)
y = func_1d(x)
eta = 1e-15
for kernel in kernels_dict.values():
k = kernel()
for tail in tails_dict.values():
t = tail()
if t.degree >= k.dmin:
model = RBF(k, t, eta)
model.fit(x, y)
p, s = model.predict(x)
assert p.shape == (len(x), )
assert s.shape == (len(x), )
assert not np.any(np.isnan(p))
assert not np.any(np.isnan(s))
assert np.all(s >= 0)
assert np.all(s <= 1e-3)
assert np.allclose(p, y)
p, s = model.predict(u)
assert p.shape == (len(u), )
assert s.shape == (len(u), )
assert not np.any(np.isnan(p))
assert not np.any(np.isnan(s))
assert np.all(s >= 0)
def test_predict_2d_fit(self):
x = 2 * np.pi * np.random.random_sample((20, 2))
u = 2 * np.pi * np.random.random_sample((50, 2))
y = func_2d(x)
eta = 1e-15
for kernel in kernels_dict.values():
k = kernel()
for tail in tails_dict.values():
t = tail()
if t.degree >= k.dmin:
model = RBF(k, t, eta)
model.fit(x, y)
p, s = model.predict(x)
assert p.shape == (len(x), )
assert s.shape == (len(x), )
assert not np.any(np.isnan(p))
assert not np.any(np.isnan(s))
assert np.all(s >= 0)
assert np.all(s <= 1e-3)
assert np.allclose(p, y)
p, s = model.predict(u)
assert p.shape == (len(u), )
assert s.shape == (len(u), )
assert not np.any(np.isnan(p))
assert not np.any(np.isnan(s))
assert np.all(s >= 0)
def test_copy_fit_2d(self):
x = 2 * np.pi * np.random.random_sample((20, 2))
y = func_2d(x)
eta = 1e-3
for kernel in kernels_dict.values():
k = kernel(2)
for tail in tails_dict.values():
t = tail()
if t.degree >= k.dmin:
model = RBF(k, t, eta)
model.fit(x, y)
copy_model = model.copy()
assert np.allclose(copy_model._X, model._X)
assert np.allclose(copy_model._y, model._y)
assert copy_model._eta == model._eta
assert isinstance(copy_model._kernel,
model._kernel.__class__)
assert copy_model._kernel.param == model._kernel.param
assert isinstance(copy_model._tail, model._tail.__class__)
assert np.all(
copy_model._tail.params == model._tail.params)
assert np.allclose(copy_model._lambda, model._lambda)
assert np.allclose(copy_model._LU, model._LU)
assert np.allclose(copy_model._piv, model._piv)
assert np.allclose(copy_model.loo_residuals,
model.loo_residuals)
def test_to_dict_fit_2d(self):
x = 2 * np.pi * np.random.random_sample((20, 2))
y = func_2d(x)
eta = 1e-3
for kernel in kernels_dict.values():
k = kernel(2)
for tail in tails_dict.values():
t = tail()
if t.degree >= k.dmin:
model = RBF(k, t, eta)
model.fit(x, y)
output_dict = model.to_dict()
assert "X" in output_dict
assert "y" in output_dict
assert "eta" in output_dict
assert "kernel" in output_dict
assert "tail" in output_dict
assert "lambda" in output_dict
assert "LU" in output_dict
assert "piv" in output_dict
assert "loo" in output_dict
assert np.allclose(np.array(output_dict["X"]), model._X)
assert np.allclose(np.array(output_dict["y"]), model._y)
assert output_dict["eta"] == model._eta
assert output_dict["kernel"] == model._kernel.to_dict()
assert output_dict["tail"] == model._tail.to_dict()
assert np.allclose(np.array(output_dict["lambda"]),
model._lambda)
assert np.allclose(np.array(output_dict["LU"]), model._LU)
assert np.allclose(np.array(output_dict["piv"]),
model._piv)
assert np.allclose(np.array(output_dict["loo"]),
model.loo_residuals)
def test_predict_nofit(self):
u = np.linspace(0, 2 * np.pi, num=50)
for kernel in kernels_dict.values():
k = kernel()
for tail in tails_dict.values():
t = tail()
if t.degree >= k.dmin:
model = RBF(k, t)
with pytest.raises(ValueError):
p, s = model.predict(u)
def test_train_2d(self):
N = 20
x = 2 * np.pi * np.random.random_sample((N, 2))
y = func_2d(x)
for kernel in kernels_dict.values():
k = kernel(30)
for tail in tails_dict.values():
t = tail()
if t.degree >= k.dmin:
model = RBF(k, t)
model.train(x, y, method="Bounded", bounds=[1e-5, 20])
assert model._kernel.param >= 1e-5
assert model._kernel.param <= 20
assert model._is_fitted()
def test_train_1d(self):
N = 8
x = np.linspace(0, 2 * np.pi, num=N)
y = func_1d(x)
for kernel in kernels_dict.values():
k = kernel(30)
for tail in tails_dict.values():
t = tail()
if t.degree >= k.dmin:
model = RBF(k, t)
model.train(x, y, method="Bounded", bounds=[1e-5, 20])
assert model._kernel.param >= 1e-5
assert model._kernel.param <= 20
assert model._is_fitted()
def test___init__tails(self):
for name, cls in tails_dict.items():
_tail = cls()
for kernel in kernels_dict.values():
k = kernel()
if _tail.degree >= k.dmin:
model = RBF(k, tail=name)
assert isinstance(model._tail, cls)
model = RBF(k, tail=cls())
assert isinstance(model._tail, cls)
else:
with pytest.raises(ValueError):
model = RBF(k, tail=name)
with pytest.raises(ValueError):
model = RBF(k, tail=cls())
def test_to_dict_nofit(self):
eta = 1
for kernel in kernels_dict.values():
k = kernel(2)
for tail in tails_dict.values():
t = tail()
if t.degree >= k.dmin:
model = RBF(k, t, eta)
output_dict = model.to_dict()
assert "eta" in output_dict
assert "kernel" in output_dict
assert "tail" in output_dict
assert output_dict["eta"] == model._eta
assert output_dict["kernel"] == model._kernel.to_dict()
assert output_dict["tail"] == model._tail.to_dict()
def test_fit_2d_default(self):
N = 20
x = 2 * np.pi * np.random.random_sample((N, 2))
y = func_2d(x)
eta = 1e-15
for kernel in kernels_dict.values():
k = kernel()
for tail in tails_dict.values():
t = tail()
if t.degree >= k.dmin:
model = RBF(k, t, eta)
model.fit(x, y)
assert hasattr(model, "_lambda")
assert model._lambda.shape == (N, )
assert hasattr(model, "_LU")
assert model._LU.shape == (N, N) if t.degree == -1 else (
N + 1, N + 1) if t.degree == 0 else (N + 2, N + 2)
assert hasattr(model, "_piv")
assert model._piv.shape == (N, ) if t.degree == -1 else (
N + 1, ) if t.degree == 0 else (N + 2, )
assert hasattr(model, "_loo_residuals")
assert model._loo_residuals.shape == (N, )
assert model._kernel.param == k.param
if t.degree == -1:
assert np.allclose(
np.dot(model._Phi(model._X),
model._lambda.reshape((-1, 1))),
y.reshape((-1, 1)))
elif t.degree == 0:
assert np.allclose(
np.dot(model._Phi(model._X),
model._lambda.reshape(
(-1, 1))) + model._tail._p * np.ones(
(N, 1)), y.reshape((-1, 1)))
assert np.sum(model._lambda) < 1e-5
assert np.sum(model._lambda) > -1e-5
elif t.degree == 1:
assert np.allclose(
np.dot(model._Phi(model._X),
model._lambda.reshape((-1, 1))) + np.dot(
model._tail.compute_P(model._X),
np.array(model._tail.params).reshape(
(-1, 1))), y.reshape((-1, 1)))
assert np.allclose(
np.dot(
np.transpose(model._tail.compute_P(model._X)),
model._lambda.reshape((-1, 1))),
np.zeros((3, 1)))
def test_copy_nofit(self):
eta = 1
for kernel in kernels_dict.values():
k = kernel(2)
for tail in tails_dict.values():
t = tail()
if t.degree >= k.dmin:
model = RBF(k, t, eta)
copy_model = model.copy()
assert copy_model._eta == model._eta
assert isinstance(copy_model._kernel,
model._kernel.__class__)
assert copy_model._kernel.param == model._kernel.param
assert isinstance(copy_model._tail, model._tail.__class__)
assert np.all(
copy_model._tail.params == model._tail.params)
def test_from_dict_nofit(self):
eta = 1
for kernel in kernels_dict.values():
k = kernel(2)
for tail in tails_dict.values():
t = tail()
input_dict = {
"eta": eta,
"kernel": k.to_dict(),
"tail": t.to_dict()
}
if t.degree >= k.dmin:
model = RBF.from_dict(input_dict)
assert model._eta == eta
assert isinstance(model._kernel, kernel)
assert model._kernel.param == 2
assert isinstance(model._tail, tail)
else:
with pytest.raises(ValueError):
model = RBF.from_dict(input_dict)
def test_from_dict_fit_1d(self):
x = np.linspace(0, 2 * np.pi, num=8)
y = func_1d(x)
eta = 1e-3
p = 3
for kernel in kernels_dict.values():
k = kernel(2)
for tail in tails_dict.values():
t = tail()
if t.degree >= k.dmin:
_model = RBF(k, t, eta)
_model.fit(x, y, p)
input_dict = {
"X": x.reshape((-1, 1)).tolist(),
"y": y.tolist(),
"eta": eta,
"kernel": k.to_dict(),
"lambda": _model._lambda.tolist(),
"LU": _model._LU.tolist(),
"piv": _model._piv.tolist(),
"loo": _model._loo_residuals.tolist(),
"tail": _model._tail.to_dict()
}
model = RBF.from_dict(input_dict)
assert model._X.shape == (8, 1)
assert np.allclose(model._X.ravel(), x)
assert model._y.shape == (8, )
assert np.allclose(model._y, y)
assert model._eta == eta
assert isinstance(model._kernel, kernel)
assert model._kernel.param == p
assert np.allclose(model._lambda, _model._lambda)
assert np.allclose(model._LU, _model._LU)
assert np.allclose(model._piv, _model._piv)
assert np.allclose(model._loo_residuals,
_model._loo_residuals)
assert isinstance(model._tail, _model._tail.__class__)
assert np.all(model._tail.params == _model._tail.params)
