diffalign.models.encoder package

Submodules

diffalign.models.encoder.edge module

class diffalign.models.encoder.edge.MLPEdgeEncoder(hidden_dim=100, activation='relu')

Bases: PatchedModule

forward(edge_length, edge_type)

Input:

edge_length: The length of edges, shape=(E, 1). edge_type: The type pf edges, shape=(E,)

Returns:

The representation of edges. (E, self.hidden_dim)

Return type:

edge_attr

property out_channels

diffalign.models.encoder.egnn module

class diffalign.models.encoder.egnn.EGNN(in_node_nf, in_edge_nf, hidden_nf, device='cpu', act_fn=SiLU(), n_layers=4, recurrent=True, attention=False, norm_diff=True, out_node_nf=None, tanh=False, coords_range=15, agg='sum', norm_constant=0, inv_sublayers=1, sin_embedding=False)

Bases: PatchedModule

forward(h, x, edges, edge_attr=None, node_mask=None, edge_mask=None, coord_mask=None)

Define the computation performed at every call.

Should be overridden by all subclasses.

Note

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the registered hooks while the latter silently ignores them.

class diffalign.models.encoder.egnn.EGNN_old(in_node_nf, in_edge_nf, hidden_nf, device='cpu', act_fn=SiLU(), n_layers=4, recurrent=True, attention=False, norm_diff=True, out_node_nf=None, tanh=False, coords_range=15, agg='sum')

Bases: PatchedModule

forward(h, x, edges, edge_attr=None, node_mask=None, edge_mask=None)

Define the computation performed at every call.

Should be overridden by all subclasses.

Note

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the registered hooks while the latter silently ignores them.

class diffalign.models.encoder.egnn.E_GCL(input_nf, output_nf, hidden_nf, edges_in_d=0, nodes_att_dim=0, act_fn=SiLU(), attention=False, norm_diff=True, tanh=False, coords_range=1, norm_constant=0)

Bases: PatchedModule

Graph Neural Net with global state and fixed number of nodes per graph. :param hidden_dim: Number of hidden units. :param num_nodes: Maximum number of nodes (for self-attentive pooling). :param global_agg: Global aggregation function (‘attn’ or ‘sum’). :param temp: Softmax temperature.

coord2radial(edge_index, coord)

coord_model(coord, edge_index, coord_diff, radial, edge_feat, node_mask, edge_mask, coord_mask)

edge_model(source, target, radial, edge_attr, edge_mask)

forward(h, edge_index, coord, edge_attr=None, node_attr=None, node_mask=None, edge_mask=None, coord_mask=None)

Define the computation performed at every call.

Should be overridden by all subclasses.

Note

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the registered hooks while the latter silently ignores them.

node_model(x, edge_index, edge_attr, node_attr)

class diffalign.models.encoder.egnn.GNN(in_node_nf, in_edge_nf, hidden_nf, device='cpu', act_fn=SiLU(), n_layers=4, attention=False, out_node_nf=None)

Bases: PatchedModule

forward(h, edges, edge_attr=None, node_mask=None, edge_mask=None)

Define the computation performed at every call.

Should be overridden by all subclasses.

Note

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the registered hooks while the latter silently ignores them.

class diffalign.models.encoder.egnn.MLP(dim_in: int, dim_out: int, width: int, nb_layers: int, skip=1, bias=True)

Bases: PatchedModule

forward(x: Tensor)

Define the computation performed at every call.

Should be overridden by all subclasses.

Note

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the registered hooks while the latter silently ignores them.

class diffalign.models.encoder.egnn.TransformerNN(in_node_nf, in_edge_nf, hidden_nf, device='cpu', act_fn=SiLU(), n_layers=4, recurrent=True, attention=False, norm_diff=True, out_node_nf=None, tanh=False, coords_range=15, agg='sum', norm_constant=0)

Bases: PatchedModule

class SetDecoder(cfg)

Bases: PatchedModule

forward(x, latent)

x: batch_size, n, channels latent: batch_size, channels2.

forward(h, x, edges, edge_attr=None, node_mask=None, edge_mask=None)

x: batch_size, n, channels latent: batch_size, channels2.

diffalign.models.encoder.egnn.unsorted_segment_sum(data, segment_ids, num_segments)

Custom PyTorch op to replicate TensorFlow’s unsorted_segment_sum.

Module contents