def test_rad_sf(self):
with unittest.mock.patch.dict("os.environ", {"CML_SCRATCH": str(self.tmpdir)}):
from cmlkit.representation.sf import SymmetryFunctions
from cmlkit import Dataset, runner_path
print(runner_path)
data = Dataset(
z=np.array([[1, 1]]), r=np.array([[[0.0, 0.0, 0.0], [2.0, 0.0, 0.0]]])
)
for i in range(5):
eta = np.random.random()
mu = np.random.random()
cutoff = 5.0
sf = SymmetryFunctions(
[1], universal=[{"rad": {"eta": eta, "mu": mu, "cutoff": cutoff}}]
)
computed = sf(data).ragged
np.testing.assert_almost_equal(
computed[0][0][0], rad_sf(2.0, eta, mu) * fc(2.0, cutoff)
)
np.testing.assert_almost_equal(
computed[0][1][0], rad_sf(2.0, eta, mu) * fc(2.0, cutoff)
)
def test_parametrized_sf(self):
data = Dataset(z=np.array([[1, 1]]),
r=np.array([[[0.0, 0.0, 0.0], [2.0, 0.0, 0.0]]]))
cutoff = 5.0
sf = SymmetryFunctions([1],
sfs=[{
"rad_centered": {
"n": 3
}
}],
cutoff=cutoff)
delta = (cutoff - 1.5) / 2
computed = sf(data)
print(computed)
np.testing.assert_almost_equal(computed[0][0][0], fc(2.0, cutoff))
np.testing.assert_almost_equal(
computed[0][0][1],
rad_sf(2.0, 0.5 / (0.5 + 0 * delta)**2, 0.0) * fc(2.0, cutoff),
)
np.testing.assert_almost_equal(
computed[0][0][2],
rad_sf(2.0, 0.5 / (0.5 + 1 * delta)**2, 0.0) * fc(2.0, cutoff),
)
np.testing.assert_almost_equal(
computed[0][0][3],
rad_sf(2.0, 0.5 / (0.5 + 2 * delta)**2, 0.0) * fc(2.0, cutoff),
)
def test_rad_sf(self):
data = Dataset(z=np.array([[1, 1]]),
r=np.array([[[0.0, 0.0, 0.0], [2.0, 0.0, 0.0]]]))
for i in range(5):
eta = np.random.random()
mu = np.random.random()
cutoff = 6.0
sf = SymmetryFunctions(elems=[1],
cutoff=cutoff,
sfs=[{
"rad": {
"eta": eta,
"mu": mu
}
}])
computed = sf(data)
print(computed.shape)
np.testing.assert_almost_equal(computed[0][0][0], fc(2.0, cutoff))
np.testing.assert_almost_equal(
computed[0][0][1],
rad_sf(2.0, eta, mu) * fc(2.0, cutoff))
np.testing.assert_almost_equal(computed[0][1][0], fc(2.0, cutoff))
np.testing.assert_almost_equal(
computed[0][1][1],
rad_sf(2.0, eta, mu) * fc(2.0, cutoff))
def test_mbtr2_chunked(self):
# naughty -- this is testing functionality that is common
# to ALL representations!
data = Dataset(
z=np.array([[0, 0], [0, 0], [0, 0]]), r=np.array([[[0.0, 0.0, 0.0], [2.0, 0.0, 0.0]], [[0.0, 0.0, 0.0], [2.0, 2.0, 3.0]], [[0.0, 3.0, 0.0], [2.0, 1.0, 0.0]]])
)
mbtr_chunked = MBTR2(
start=0,
stop=1,
num=3,
geomf="1/distance",
weightf="unity",
broadening=0.001,
eindexf="noreversals",
aindexf="noreversals",
elems=[0],
flatten=True,
context={"chunk_size": 2}
)
mbtr = MBTR2(
start=0,
stop=1,
num=3,
geomf="1/distance",
weightf="unity",
broadening=0.001,
eindexf="noreversals",
aindexf="noreversals",
elems=[0],
flatten=True,
context={"chunk_size": None}
)
np.testing.assert_array_equal(mbtr_chunked(data).array, mbtr(data).array)
def test_ang_sf(self):
with unittest.mock.patch.dict("os.environ", {"CML_SCRATCH": str(self.tmpdir)}):
from cmlkit.representation.sf import SymmetryFunctions
from cmlkit import Dataset, runner_path
print(runner_path)
data = Dataset(
z=np.array([[1, 1, 1]]),
r=np.array([[[0.0, 0.0, 0.0], [1.0, 0.0, 0.0], [0.0, 1.0, 0.0]]]),
)
for i in range(5):
eta = np.random.random()
zeta = np.random.random()
cutoff = 5.0
lambd = 1.0 + np.random.random() * 0.3
sf = SymmetryFunctions(
[1],
universal=[
{
"ang": {
"eta": eta,
"zeta": zeta,
"cutoff": cutoff,
"lambd": lambd,
}
}
],
)
computed = sf(data).ragged
print(computed)
np.testing.assert_almost_equal(
computed[0][0][0],
ang_sf(np.pi / 2, 1.0, 1.0, np.sqrt(2.0), lambd, zeta, eta)
* fc(1.0, cutoff)
* fc(1.0, cutoff)
* fc(np.sqrt(2.0), cutoff),
)
np.testing.assert_almost_equal(
computed[0][1][0],
ang_sf(np.pi / 4, 1.0, 1.0, np.sqrt(2.0), lambd, zeta, eta)
* fc(1.0, cutoff)
* fc(1.0, cutoff)
* fc(np.sqrt(2.0), cutoff),
)
def setUp(self):
self.n = 100
self.n_atoms = np.random.randint(1, high=10, size=self.n)
r = [5 * np.random.random((na, 3)) for na in self.n_atoms]
self.r = np.array(r, dtype=object)
self.z = np.array(
[np.random.randint(1, high=3, size=na) for na in self.n_atoms], dtype=object
)
self.data = Dataset(z=self.z, r=self.r)
self.tmpdir = (pathlib.Path(__file__) / "..").resolve() / "tmp_test_composed"
self.tmpdir.mkdir(exist_ok=True)
def test_ang_sf(self):
data = Dataset(
z=np.array([[1, 1, 1]]),
r=np.array([[[0.0, 0.0, 0.0], [1.0, 0.0, 0.0], [0.0, 1.0, 0.0]]]),
)
for i in range(5):
eta = np.random.random()
zeta = np.random.random()
cutoff = 5.0
lambd = 1.0 + np.random.random() * 0.3
sf = SymmetryFunctions(
[1],
sfs=[{
"ang": {
"eta": eta,
"zeta": zeta,
"lambd": lambd
}
}],
cutoff=cutoff,
)
computed = sf(data)
print(computed)
np.testing.assert_almost_equal(computed[0][0][0],
fc(1.0, cutoff) + fc(1.0, cutoff))
np.testing.assert_almost_equal(
computed[0][0][1],
ang_sf(np.pi / 2, 1.0, 1.0, np.sqrt(2.0), lambd, zeta, eta) *
fc(1.0, cutoff) * fc(1.0, cutoff) * fc(np.sqrt(2.0), cutoff),
)
np.testing.assert_almost_equal(
computed[0][1][0],
fc(1.0, cutoff) + fc(np.sqrt(2.0), cutoff))
np.testing.assert_almost_equal(
computed[0][1][1],
ang_sf(np.pi / 4, 1.0, 1.0, np.sqrt(2.0), lambd, zeta, eta) *
fc(1.0, cutoff) * fc(1.0, cutoff) * fc(np.sqrt(2.0), cutoff),
)
def test_smoke(self):
# since quippy is impossible to install,
# we can't test it here.
# but we can at least check that things are not TOO broken
with unittest.mock.patch.dict(
"os.environ",
{
"CML_QUIPPY_PYTHONPATH": str(self.tmpdir),
"CML_QUIPPY_PYTHON_EXE": "",
"CML_SCRATCH": str(self.tmpdir),
},
):
res = np.array([[np.array([1.0]), np.array([2.0])]], dtype=object)
def fake_output(*args, **kwargs):
outfolder = list(self.tmpdir.glob("soap_*"))[0]
np.save(outfolder / "out", res)
return "", ""
fake_task = unittest.mock.MagicMock(side_effect=fake_output)
with unittest.mock.patch(
"cmlkit.representation.soap.quippy_interface.run_task",
fake_task):
from cmlkit.representation.soap import SOAP
from cmlkit import Dataset
data = Dataset(
z=np.array([[1, 1]]),
r=np.array([[[0.0, 0.0, 0.0], [2.0, 0.0, 0.0]]]),
)
soap = SOAP(elems=[1], sigma=0.1, n_max=2, l_max=3, cutoff=5.0)
computed = soap(data).ragged
np.testing.assert_array_equal(computed, res)
def test_rad_sf_periodic(self):
data = Dataset(
z=np.array([[1, 1]]),
r=np.array([[[0.0, 0.0, 0.0], [2.0, 0.0, 0.0]]]),
b=np.array([[[8.0, 0, 0], [0, 8.0, 0], [0, 0, 8.0]]]),
)
# just testing that nothing is broken; cutoff is smaller than cell
for i in range(5):
eta = np.random.random()
mu = np.random.random()
cutoff = 6.0
sf = SymmetryFunctions(elems=[1],
cutoff=cutoff,
sfs=[{
"rad": {
"eta": eta,
"mu": mu
}
}])
computed = sf(data)
print(computed.shape)
np.testing.assert_almost_equal(computed[0][0][0], fc(2.0, cutoff))
np.testing.assert_almost_equal(
computed[0][0][1],
rad_sf(2.0, eta, mu) * fc(2.0, cutoff))
np.testing.assert_almost_equal(computed[0][1][0], fc(2.0, cutoff))
np.testing.assert_almost_equal(
computed[0][1][1],
rad_sf(2.0, eta, mu) * fc(2.0, cutoff))
def test_parametrized_sf(self):
with unittest.mock.patch.dict("os.environ", {"CML_SCRATCH": str(self.tmpdir)}):
from cmlkit.representation.sf import SymmetryFunctions
from cmlkit import Dataset, runner_path
print(runner_path)
data = Dataset(
z=np.array([[1, 1]]), r=np.array([[[0.0, 0.0, 0.0], [2.0, 0.0, 0.0]]])
)
cutoff = 5.0
sf = SymmetryFunctions(
[1], universal=[{"rad_centered": {"n": 3, "cutoff": cutoff}}]
)
delta = (cutoff - 1.5) / 2
computed = sf(data).ragged
# note that the ordering of the results is not the same as
# the one you'd expect -- runner internally reorders.
# I can't be bothered to find out how for now.
np.testing.assert_almost_equal(
computed[0][0][2],
rad_sf(2.0, 0.5 / (0.5 + 0 * delta) ** 2, 0.0) * fc(2.0, cutoff),
)
np.testing.assert_almost_equal(
computed[0][0][1],
rad_sf(2.0, 0.5 / (0.5 + 1 * delta) ** 2, 0.0) * fc(2.0, cutoff),
)
np.testing.assert_almost_equal(
computed[0][0][0],
rad_sf(2.0, 0.5 / (0.5 + 2 * delta) ** 2, 0.0) * fc(2.0, cutoff),
)
def test_rad_sf_multiple_elements(self):
data = Dataset(
z=np.array([[1, 2, 3]]),
r=np.array([[[0.0, 0.0, 0.0], [1.0, 0.0, 0.0], [0.0, 1.0, 0.0]]]),
)
for i in range(5):
eta = np.random.random()
mu = np.random.random()
cutoff = 5.0
sf = SymmetryFunctions([1, 2, 3],
sfs=[{
"rad": {
"eta": eta,
"mu": mu
}
}],
cutoff=cutoff)
computed = sf(data)
print(computed)
print(computed.shape)
# Atom 0
# Other Z = 1
np.testing.assert_almost_equal(computed[0][0][0], 0.0)
np.testing.assert_almost_equal(computed[0][0][1], 0.0)
# Other Z = 2
np.testing.assert_almost_equal(computed[0][0][2], fc(1.0, cutoff))
np.testing.assert_almost_equal(
computed[0][0][3],
rad_sf(1.0, eta, mu) * fc(1.0, cutoff))
# Other Z = 3
np.testing.assert_almost_equal(computed[0][0][4], fc(1.0, cutoff))
np.testing.assert_almost_equal(
computed[0][0][5],
rad_sf(1.0, eta, mu) * fc(1.0, cutoff))
# Atom 1
# Other Z = 1
np.testing.assert_almost_equal(computed[0][1][6 + 0],
fc(1.0, cutoff))
np.testing.assert_almost_equal(
computed[0][1][6 + 1],
rad_sf(1.0, eta, mu) * fc(1.0, cutoff))
# Other Z = 2
np.testing.assert_almost_equal(computed[0][1][6 + 2], 0.0)
np.testing.assert_almost_equal(computed[0][1][6 + 3], 0.0)
# Other Z = 3
np.testing.assert_almost_equal(computed[0][1][6 + 4],
fc(np.sqrt(2.0), cutoff))
np.testing.assert_almost_equal(
computed[0][1][6 + 5],
rad_sf(np.sqrt(2.0), eta, mu) * fc(np.sqrt(2.0), cutoff),
)
# Atom 2
# Other Z = 1
np.testing.assert_almost_equal(computed[0][2][12 + 0],
fc(1.0, cutoff))
np.testing.assert_almost_equal(
computed[0][2][12 + 1],
rad_sf(1.0, eta, mu) * fc(1.0, cutoff))
# Other Z = 2
np.testing.assert_almost_equal(computed[0][2][12 + 2],
fc(np.sqrt(2.0), cutoff))
np.testing.assert_almost_equal(
computed[0][2][12 + 3],
rad_sf(np.sqrt(2.0), eta, mu) * fc(np.sqrt(2.0), cutoff),
)
# Other Z = 3
np.testing.assert_almost_equal(computed[0][2][12 + 4], 0.0)
np.testing.assert_almost_equal(computed[0][2][12 + 5], 0.0)
def setUp(self):
self.data = Dataset(
z=np.array([[1, 2, 2]]),
r=np.array([[[0.0, 0.0, 0.0], [1.0, 0.0, 0.0], [2.0, 0.0, 0.0]]]),
)
def setUp(self):
self.data = Dataset(z=np.array([[0, 1]]), r=np.array([[[1, 2, 3], [1, 2, 3]]]))
def setUp(self):
self.data = Dataset(
z=np.array([[0, 0, 0]]),
r=np.array([[[0.0, 0.0, 0.0], [1.0, 0.0, 0.0], [0.0, 1.0, 0.0]]]),
)
def setUp(self):
self.data = Dataset(z=np.array([[2, 3]]),
r=np.array([[[0, 0, 0], [2.0, 0, 0]]]))
