megnet.data.crystal module

Crystal graph related

class CrystalGraph(nn_strategy: Union[str, pymatgen.analysis.local_env.NearNeighbors] = 'MinimumDistanceNNAll', atom_converter: Optional[megnet.data.graph.Converter] = None, bond_converter: Optional[megnet.data.graph.Converter] = None, cutoff: float = 5.0)

Bases: megnet.data.graph.StructureGraphFixedRadius

Convert a crystal into a graph with z as atomic feature and distance as bond feature one can optionally include state features

Convert the structure into crystal graph :param nn_strategy: NearNeighbor strategy :type nn_strategy: str :param atom_converter: atom features converter :type atom_converter: Converter :param bond_converter: bond features converter :type bond_converter: Converter :param cutoff: cutoff radius :type cutoff: float

class CrystalGraphDisordered(nn_strategy: Union[str, pymatgen.analysis.local_env.NearNeighbors] = 'MinimumDistanceNNAll', atom_converter: megnet.data.graph.Converter = <megnet.data.crystal._AtomEmbeddingMap object>, bond_converter: Optional[megnet.data.graph.Converter] = None, cutoff: float = 5.0)

Bases: megnet.data.graph.StructureGraphFixedRadius

Enable disordered site predictions

Convert the structure into crystal graph :param nn_strategy: NearNeighbor strategy :type nn_strategy: str :param atom_converter: atom features converter :type atom_converter: Converter :param bond_converter: bond features converter :type bond_converter: Converter :param cutoff: cutoff radius :type cutoff: float

static get_atom_features(structure) → List[dict]

For a structure return the list of dictionary for the site occupancy for example, Fe0.5Ni0.5 site will be returned as {“Fe”: 0.5, “Ni”: 0.5}

Parameters

structure (Structure) – pymatgen Structure with potential site disorder

Returns

a list of site fraction description

class CrystalGraphWithBondTypes(nn_strategy: Union[str, pymatgen.analysis.local_env.NearNeighbors] = 'VoronoiNN', atom_converter: Optional[megnet.data.graph.Converter] = None, bond_converter: Optional[megnet.data.graph.Converter] = None)

Bases: megnet.data.graph.StructureGraph

Overwrite the bond attributes with bond types, defined simply by the metallicity of the atoms forming the bond. Three types of scenario is considered, nonmetal-nonmetal (type 0), metal-nonmetal (type 1), and metal-metal (type 2)

Parameters

• nn_strategy (str) – NearNeighbor strategy

• atom_converter (Converter) – atom features converter

• bond_converter (Converter) – bond features converter

convert(structure: pymatgen.core.structure.Structure, state_attributes: Optional[List] = None) → Dict

Convert structure into graph :param structure: pymatgen Structure :type structure: Structure :param state_attributes: state attributes :type state_attributes: list

Returns: graph dictionary

get_elemental_embeddings() → Dict

Provides the pre-trained elemental embeddings using formation energies, which can be used to speed up the training of other models. The embeddings are also extremely useful elemental descriptors that encode chemical similarity that may be used in other ways. See

“Graph Networks as a Universal Machine Learning Framework for Molecules and Crystals”, https://arxiv.org/abs/1812.05055

Returns

Dict of elemental embeddings as {symbol: length 16 string}