MolecularDiffusion.modules.models.ligandiff.gvp_model

GVP network. Verbatim port of LigandDiff’s src/gvp_model.py.

Classes

GVPNetwork

Stack of GVP conv layers over a geometric graph.

Module Contents

class MolecularDiffusion.modules.models.ligandiff.gvp_model.GVPNetwork(in_dims: Tuple[int, int], out_dims: Tuple[int, int], hidden_dims: Tuple[int, int], num_layers: int, drop_rate: float = 0.0, attention: bool = False, normalization_factor: float = 100.0, aggr: str = 'add', activations=(F.silu, None), vector_gate: bool = True, eps: float = 0.0001)

Bases: torch.nn.Module

Stack of GVP conv layers over a geometric graph.

forward(h, pos, edge_index) → MolecularDiffusion.modules.models.ligandiff.gvp.s_V

get_edge_attr(edge_index, pos) → MolecularDiffusion.modules.models.ligandiff.gvp.s_V

edge_embedding

embedding_in

embedding_out

eps = 0.0001

layers