GOOD.networks.models.CIGAGNN

Implementation of the CIGA algorithm from “Learning Causally Invariant Representations for Out-of-Distribution Generalization on Graphs” paper

Copied from https://github.com/LFhase/GOOD.

Functions

clear_masks(model)
relabel(x, edge_index, batch[, pos])
set_masks(mask, model)
sparse_sort(src, index[, dim, descending, eps])	Adopt from <https://github.com/rusty1s/pytorch_scatter/issues/48>_.
sparse_topk(src, index, ratio[, dim, ...])
split_batch(g)
split_graph(data, edge_score, ratio)

GOOD.networks.models.CIGAGNN.clear_masks(model: Module)

GOOD.networks.models.CIGAGNN.relabel(x, edge_index, batch, pos=None)

GOOD.networks.models.CIGAGNN.set_masks(mask: Tensor, model: Module)

GOOD.networks.models.CIGAGNN.sparse_sort(src: Tensor, index: Tensor, dim=0, descending=False, eps=1e-12)

Adopt from <https://github.com/rusty1s/pytorch_scatter/issues/48>_.

GOOD.networks.models.CIGAGNN.sparse_topk(src: Tensor, index: Tensor, ratio: float, dim=0, descending=False, eps=1e-12)

GOOD.networks.models.CIGAGNN.split_batch(g)

GOOD.networks.models.CIGAGNN.split_graph(data, edge_score, ratio)

Classes

CIGAGIN(config)
CIGAvGIN(config)
CIGAvGINNB(config)
CIGAvGINNC(config)	using a simple GNN to encode spurious subgraph
GAEAttNet(causal_ratio, config, **kwargs)

class GOOD.networks.models.CIGAGNN.CIGAGIN(config: Union[CommonArgs, Munch])

Bases: GNNBasic

forward(*args, **kwargs)

The CIGA model implementation.

Parameters

• *args (list) – argument list for the use of arguments_read. Refer to arguments_read

• **kwargs (dict) – key word arguments for the use of arguments_read. Refer to arguments_read

Returns (Tensor):

label predictions

class GOOD.networks.models.CIGAGNN.CIGAvGIN(config: Union[CommonArgs, Munch])

Bases: CIGAGIN

class GOOD.networks.models.CIGAGNN.CIGAvGINNB(config: Union[CommonArgs, Munch])

Bases: CIGAGIN

class GOOD.networks.models.CIGAGNN.CIGAvGINNC(config: Union[CommonArgs, Munch])

Bases: CIGAGIN

using a simple GNN to encode spurious subgraph

class GOOD.networks.models.CIGAGNN.GAEAttNet(causal_ratio, config, **kwargs)

Bases: Module

forward(*args, **kwargs)

Defines 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.