pyagc.transforms.GSSLTransform

class GSSLTransform(p_feat_mask: float = 0.5, p_edge_drop: float = 0.5, node_attrs: Optional[List[str]] = ['x'], edge_attrs: Optional[List[str]] = ['edge_attr'])[source]

Applies random feature masking and random edge dropping for Graph Self-Supervised Learning (functional name: gssl_transform).

This transform is commonly used in graph self-supervised learning methods such as GRACE, CCA-SSG, and BGRL.

For each node attribute in node_attrs, randomly masks features. For each edge attribute in edge_attrs, randomly drops edges.

Works for both homogeneous and heterogeneous graphs.

Only keeps specified node attributes and edge attributes in the returned data.

Parameters:

p_feat_mask (float, optional) – Probability of masking node features. (default: 0.5)
p_edge_drop (float, optional) – Probability of dropping edges. (default: 0.5)
node_attrs (List[str], optional) – Node attributes to transform and keep. (default: ["x"])
edge_attrs (List[str], optional) – Edge attributes to transform and keep. (default: ["edge_attr"])

__init__(p_feat_mask: float = 0.5, p_edge_drop: float = 0.5, node_attrs: Optional[List[str]] = ['x'], edge_attrs: Optional[List[str]] = ['edge_attr'])[source]

Methods

__init__([p_feat_mask, p_edge_drop, ...])