def test_fit_features(self):
trans = Shell(input_type="filename")
feats = trans.fit_transform(ALL)
a = AtomKernel()
values = list(zip(feats, ALL_NUMS))
a.fit(values)
self.assertEqual(ALL_NUMS, list(a._numbers))
self.assertEqual(list(ALL_FEATURES), list(a._features))
def test_fit_transform_transformer(self):
trans = Shell(input_type="filename")
a = AtomKernel(transformer=trans)
res = a.fit_transform(ALL)
try:
numpy.testing.assert_array_almost_equal(RBF_KERNEL, res)
except AssertionError as e:
self.fail(e)
def test_fit_transform_features(self):
trans = Shell(input_type="filename")
feats = trans.fit_transform(ALL)
a = AtomKernel()
values = list(zip(feats, ALL_NUMS))
res = a.fit_transform(values)
try:
numpy.testing.assert_array_almost_equal(RBF_KERNEL, res)
except AssertionError as e:
self.fail(e)
def test_laplace_kernel(self):
# Set depth=2 so the comparison is not trivial
trans = Shell(input_type="filename", depth=2)
a = AtomKernel(transformer=trans, kernel="laplace", gamma=1.)
a.fit(ALL)
res = a.transform(ALL)
try:
numpy.testing.assert_array_almost_equal(LAPLACE_KERNEL, res)
except AssertionError as e:
self.fail(e)
def test_same_element(self):
# Set depth=2 so the comparison is not trivial
trans = Shell(input_type="filename", depth=2)
# Set gamma=1 to make the differences more noticeable
a = AtomKernel(transformer=trans, same_element=False, gamma=1.)
res = a.fit_transform(ALL)
expected = numpy.array([[17.00000033, 14.58016505],
[14.58016505, 32.76067832]])
try:
numpy.testing.assert_array_almost_equal(expected, res)
except AssertionError as e:
self.fail(e)
def test_custom_kernel(self):
# Set depth=2 so the comparison is not trivial
trans = Shell(input_type="filename", depth=2)
# Simple linear kernel
a = AtomKernel(transformer=trans,
kernel=lambda x, y: numpy.dot(x, numpy.transpose(y)))
a.fit(ALL)
res = a.transform(ALL)
try:
numpy.testing.assert_array_almost_equal(LINEAR_KERNEL, res)
except AssertionError as e:
self.fail(e)
feat = CoulombMatrix()
H2_conn = (H2_ELES, H2_COORDS, H2_CONNS)
HCN_conn = (HCN_ELES, HCN_COORDS, HCN_CONNS)
print(feat.fit_transform([H2_conn, HCN_conn]))
print()
# Example of generating the Coulomb matrix using a specified input_type
print("User specified input_type")
feat = CoulombMatrix(input_type=("coords", "numbers"))
H2_spec = (H2_COORDS, H2_NUMS)
HCN_spec = (HCN_COORDS, HCN_NUMS)
print(feat.fit_transform([H2_spec, HCN_spec]))
print()
# Example of generating the Local Coulomb matrix (atom-wise
# representation)
print("Atom feature")
feat = LocalCoulombMatrix()
print(feat.fit_transform([H2, HCN]))
# Example of generating AtomKernel
print("Atom Kernel")
feat = AtomKernel(transformer=LocalCoulombMatrix())
print(feat.fit_transform([H2, HCN]))
# Example of using arbitrary function to load data
# This example is useless, but it shows the possibility
feat = CoulombMatrix(input_type=lambda x: LazyValues(elements=HCN_ELES,
coords=HCN_COORDS))
feat.fit_transform(list(range(10)))
def test_transform_before_fit(self):
a = AtomKernel()
with self.assertRaises(ValueError):
a.transform(ALL)
def test_fit_transformer(self):
trans = Shell(input_type="filename")
a = AtomKernel(transformer=trans)
a.fit(ALL)
self.assertEqual(ALL_NUMS, [x.tolist() for x in a._numbers])
self.assertEqual(list(ALL_FEATURES), list(a._features))
def test_invalid_kernel(self):
with self.assertRaises(ValueError):
trans = Shell(input_type="filename")
a = AtomKernel(kernel=1, transformer=trans)
a.fit_transform(ALL)
from molml.kernel import AtomKernel
from utils import load_qm7
if __name__ == "__main__":
# This is just boiler plate code to load the data
Xin_train, Xin_test, y_train, y_test = load_qm7()
# Look at just a few examples to be quick
n_train = 200
n_test = 200
Xin_train = Xin_train[:n_train]
y_train = y_train[:n_train]
Xin_test = Xin_test[:n_test]
y_test = y_test[:n_test]
gamma = 1e-7
alpha = 1e-7
kern = AtomKernel(gamma=gamma,
transformer=LocalEncodedBond(n_jobs=-1),
n_jobs=-1)
K_train = kern.fit_transform(Xin_train)
K_test = kern.transform(Xin_test)
clf = KernelRidge(alpha=alpha, kernel="precomputed")
clf.fit(K_train, y_train)
train_error = MAE(clf.predict(K_train), y_train)
test_error = MAE(clf.predict(K_test), y_test)
print("Train MAE: %.4f Test MAE: %.4f" % (train_error, test_error))
print()