SchNet

class nfflr.models.SchNet(config: SchNetConfig = SchNetConfig(transform=PeriodicRadiusGraph(), cutoff=Cosine(), layers=4, norm=None, atom_features='cgcnn', edge_input_features=128, d_model=128, output_features=1, compute_forces=False, energy_units='eV', reference_energies=None))

Methods

forward(g)
reset_output_scale(avg_num_nodes)	Initialize output layer so average total energy at init has std 1.

reset_atomic_reference_energies
reset_parameters

reset_output_scale(avg_num_nodes: int)

Initialize output layer so average total energy at init has std 1.

class nfflr.models.SchNetConfig(transform: Callable = PeriodicRadiusGraph(), cutoff: Module = Cosine(), layers: int = 4, norm: Literal['batchnorm', 'layernorm'] | None = None, atom_features: str | Module = 'cgcnn', edge_input_features: int = 128, d_model: int = 128, output_features: int = 1, compute_forces: bool = False, energy_units: Literal['eV', 'eV/atom'] = 'eV', reference_energies: Literal['fixed', 'trainable'] | None = None)

Hyperparameter schema for nfflr.models.gnn.alignn.

Attributes:

norm

reference_energies

Methods

cutoff
transform