def test_comp_descriptor_1():
desc = Compositions(n_jobs=1)
desc.fit_transform(pd.Series([{'H': 2}], name='composition'))
desc.fit_transform(pd.Series([{'H': 2}], name='other'))
tmp1 = desc.fit_transform(pd.Series([{
'H': 2
}], name='other'),
composition='other')
tmp2 = desc.fit_transform([{'H': 2}])
assert tmp1.shape == (1, 290)
assert isinstance(tmp1, pd.DataFrame)
assert isinstance(tmp2, pd.DataFrame)
assert np.all(tmp1.values == tmp2.values)
tmp = desc.transform([{'H': 2}], featurizers=['WeightedAverage'])
assert tmp.shape == (1, 58)
tmp = desc.transform([{'H': 2}], featurizers='all')
assert tmp.shape == (1, 500)
tmp = desc.transform([{'H': 2}], featurizers=Compositions.classic)
assert tmp.shape == (1, 290)
class OrganicCompDescriptor(BaseFeaturizer):
def __init__(self,
n_jobs=-1,
*,
featurizers='all',
on_errors='raise',
return_type='any'):
"""
A featurizer for extracting XenonPy compositional descriptors from SMILES or MOL
"""
# fix n_jobs to be 0 to skip automatic wrapper in XenonPy BaseFeaturizer class
super().__init__(n_jobs=0,
on_errors=on_errors,
return_type=return_type)
self._cal = Compositions(n_jobs=n_jobs,
featurizers=featurizers,
on_errors=on_errors)
def featurize(self, x):
# check if type(x) = list
if isinstance(x, pd.Series):
x = x.tolist()
if not isinstance(x, list):
x = [x]
# check input format, assume SMILES if not RDKit-MOL
if not isinstance(x[0], Chem.rdchem.Mol):
x_mol = []
for z in x:
x_mol.append(Chem.MolFromSmiles(z))
if x_mol[-1] is None:
raise ValueError('can not convert Mol from SMILES %s' % z)
else:
x_mol = x
# convert to counting dictionary
mol = [Chem.AddHs(z) for z in x_mol]
d_list = [
dict(Counter([atom.GetSymbol() for atom in z.GetAtoms()]))
for z in mol
]
self.output = self._cal.transform(d_list)
return self.output
@property
def feature_labels(self):
return self.output.columns