graph2mat.bindings.e3nn.modules.node_operations

Classes

E3nnSeparateTSQNodeBlock(irreps_in, irreps_out)	Tensor squares each node features and then sums all outputs.
E3nnSimpleNodeBlock(irreps_in, irreps_out)	Sums all node features and then passes them to a tensor square.

class graph2mat.bindings.e3nn.modules.node_operations.E3nnSeparateTSQNodeBlock(irreps_in: Irreps, irreps_out: Irreps)

Bases: Module

Tensor squares each node features and then sums all outputs.

Example

If we construct a SeparateTSQNodeBlock:

>>> irreps_in = o3.Irreps("3x0e + 2x1o")
>>> irreps_out = o3.Irreps("3x2e")
>>> node_block = SeparateTSQNodeBlock(irreps_in, irreps_out)

and then use it with 2 different nodewise tensors:

>>> node_feats = torch.randn(10, irreps_in.dim)
>>> node_messages = torch.randn(10, irreps_in.dim)
>>> output = node_block(node_feats=node_feats, node_messages=node_messages)

this is equivalent to:

>>> tsq1 = o3.TensorSquare(irreps_in, irreps_out)
>>> tsq2 = o3.TensorSquare(irreps_in, irreps_out)
>>> output = tsq1(node_feats) + tsq2(node_messages)

__init__(irreps_in: Irreps, irreps_out: Irreps)

Initialize internal Module state, shared by both nn.Module and ScriptModule.

forward(**node_kwargs: Tensor) → 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 graph2mat.bindings.e3nn.modules.node_operations.E3nnSimpleNodeBlock(irreps_in: Irreps, irreps_out: Irreps)

Bases: Module

Sums all node features and then passes them to a tensor square.

All node features must have the same irreps.

Example

If we construct a SimpleNodeBlock:

>>> irreps_in = o3.Irreps("2x0e + 2x1o")
>>> irreps_out = o3.Irreps("3x2e")
>>> node_block = SimpleNodeBlock(irreps_in, irreps_out)

and then use it with 2 different nodewise tensors:

>>> node_feats = torch.randn(10, irreps_in.dim)
>>> node_messages = torch.randn(10, irreps_in.dim)
>>> node_block(node_feats=node_feats, node_messages=node_messages)

this is equivalent to:

>>> tsq = o3.TensorSquare(irreps_in, irreps_out)
>>> output = tsq(node_feats + node_messages)

__init__(irreps_in: Irreps, irreps_out: Irreps)

Initialize internal Module state, shared by both nn.Module and ScriptModule.

forward(**node_kwargs: Tensor) → 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.