Poolers Cheatsheet¶

This cheatsheet provides a quick reference for the pooling operators available in tgp. It shows each pooler’s alias (for tgp.poolers.get_pooler()), whether it supports sparse operations, has trainable parameters, and provides auxiliary losses.

sparse: If checked (✓), the pooler operates on sparse adjacency matrices and returns sparse connectivity and \(\mathbf{S}\) matrices, e.g., suitable for sparse GNN layers like GCNConv.
trainable: If checked (✓), the pooler contains trainable parameters that are updated during training.
aux_loss: If checked (✓), the pooler computes auxiliary loss terms that can be used for regularization during training.

Graph Pooling Operators¶

Alias	Class	sparse	trainable	aux_loss
`asap`	`ASAPooling` [1]	✓	✓
`acc`	`AsymCheegerCutPooling` [1]		✓	✓
`bnpool`	`BNPool` [1]		✓	✓
`dmon`	`DMoNPooling` [1]		✓	✓
`diff`	`DiffPool` [1]		✓	✓
`ec`	`EdgeContractionPooling` [1, 2, 3]	✓	✓
`eigen`	`EigenPooling` [1]
`graclus`	`GraclusPooling` [1]	✓
`hosc`	`HOSCPooling` [1]		✓	✓
`jb`	`JustBalancePooling` [1]		✓	✓
`kmis`	`KMISPooling` [1]	✓
`lap`	`LaPooling` [1]
`maxcut`	`MaxCutPooling` [1]	✓	✓	✓
`mincut`	`MinCutPooling` [1]		✓	✓
`ndp`	`NDPPooling` [1]	✓
`nmf`	`NMFPooling` [1]
`nopool`	`NoPool`	✓
`pan`	`PANPooling` [1]	✓	✓
`sag`	`SAGPooling` [1]	✓	✓
`sep`	`SEPPooling` [1]	✓
`topk`	`TopkPooling` [1, 2, 3]	✓	✓