Torch Geometric Pool: the Pytorch library for pooling in Graph Neural Networks

Filippo Maria Bianchi, Carlo Abate, Ivan Marisca

Abstract

Torch Geometric Pool (tgp) is a pooling library built on top of PyTorch Geometric. Graph pooling methods differ in how they assign nodes to supernodes, how they handle batches, what they return after pooling, and whether they expose auxiliary losses. These differences make it hard to compare methods or reuse the same model code across them. tgp addresses this problem with a common software interface based on the Select-Reduce-Connect-Lift (SRCL) decomposition. The library provides 20 hierarchical poolers, standardized output objects, standalone readout modules, support for dense poolers in batched and unbatched mode, and workflows for caching and pre-coarsening. It is released under the MIT license on GitHub and PyPI, with comprehensive documentation, tutorials, and examples.

Benchmark Results

Tested on 7 datasets

Tables show presented results in global context (best 3 papers + GCN + MLP).

TU Dortmund

NCI1

Global leaderboard →

#	Model	Accuracy ↑	Paper
1	WL-MLP GT global	0.9954	1-WL Expressiveness Is (Almost) All You Need
2	GIN GNN global	0.9934	1-WL Expressiveness Is (Almost) All You Need
3	GCN GNN global	0.9727	1-WL Expressiveness Is (Almost) All You Need
131	GraphSAGE code ↗ GNN global	0.8200 ±0.0100	Weisfeiler and Leman Go Infinite: Spectral and Combinatorial Pre-Colorings code ↗
170	Graclus Trad baseline actual #170 vs paper #391 221 places worse	0.8120 ±0.0340 best reported	actual #170 vs paper #391 221 places worse paper 0.7600 ±0.0300 best at pub: The expressive power of pooling in Graph Neural Networks (NeurIPS 2023) Difference within reported uncertainty; not counted as underperformance. paper 0.7600 ±0.0300 Δ at pub 0.0520
221	MinCut Trad baseline actual #221 vs paper #318 97 places worse	0.8010 ±0.0260 best reported	actual #221 vs paper #318 97 places worse paper 0.7800 ±0.0200 best at pub: The expressive power of pooling in Graph Neural Networks (NeurIPS 2023) Difference within reported uncertainty; not counted as underperformance. paper 0.7800 ±0.0200 Δ at pub 0.0210
231	ecpool GNN baseline actual #231 vs paper #318 87 places worse	0.7980 ±0.0330 best reported	actual #231 vs paper #318 87 places worse paper 0.7800 ±0.0200 best at pub: The expressive power of pooling in Graph Neural Networks (NeurIPS 2023) Difference within reported uncertainty; not counted as underperformance. paper 0.7800 ±0.0200 Δ at pub 0.0180
254	SEP GNN baseline actual #254 vs paper #355 101 places worse	0.7935 ±0.0033 best reported	actual #254 vs paper #355 101 places worse paper 0.7700 ±0.0200 best at pub: Scalable Graph Compressed Convolutions (2024) Difference within reported uncertainty; not counted as underperformance. paper 0.7700 ±0.0200 Δ at pub 0.0235
294	DMoN GNN baseline actual #294 vs paper #355 61 places worse	0.7850 ±0.0140 best reported	actual #294 vs paper #355 61 places worse paper 0.7700 ±0.0200 best at pub: The expressive power of pooling in Graph Neural Networks (NeurIPS 2023) Difference within reported uncertainty; not counted as underperformance. paper 0.7700 ±0.0200 Δ at pub 0.0150
309	ASAP GNN baseline actual #309 vs paper #391 82 places worse	0.7821 ±0.0175 best reported	actual #309 vs paper #391 82 places worse paper 0.7600 ±0.0200 best at pub: Multi-grained Semantics-aware Graph Neural Networks (2020) Difference within reported uncertainty; not counted as underperformance. paper 0.7600 ±0.0200 Δ at pub 0.0221
321	bnpool Hyb baseline	0.7800 ±0.0200	This paper
322	hosc GNN baseline	0.7800 ±0.0200	This paper
323	jbgnn GNN baseline	0.7800 ±0.0200	This paper
360	acc GNN baseline	0.7700 ±0.0200	This paper
361	Diff GNN baseline	0.7700 ±0.0300	This paper
362	kmis Trad baseline	0.7700 ±0.0200	This paper
363	lapool GNN baseline	0.7700 ±0.0200	This paper
364	maxcut Trad baseline	0.7700 ±0.0200	This paper
359	sag GNN baseline actual #359 vs paper #359	0.7700 ±0.0300	actual #359 vs paper #359 paper 0.7700 ±0.0300 prev best: Enhancing Graph Classification Robustness with Singular Pooling (NeurIPS 2025) paper 0.7700 ±0.0300 Δ at pub 0.0660
395	ndp GNN baseline	0.7600 ±0.0200	This paper
414	topk GNN baseline actual #414 vs paper #414	0.7500 ±0.0200	actual #414 vs paper #414 paper 0.7500 ±0.0200 prev best: Towards Efficient Training of Graph Neural Networks: A Multiscale Approach (TMLR 2025) paper 0.7500 ±0.0200 Δ at pub 0.0563
473	pan GNN baseline	0.7100 ±0.0300	This paper
499	eigen Trad baseline	0.6900 ±0.0200	This paper

REDDIT-BINARY

Global leaderboard →

#	Model	Accuracy ↑	Paper
1	GIN-0 GNN global	0.9240 ±0.0250	Contrastive Multi-View Representation Learning on Graphs
2	GIN-ε GNN global	0.9220 ±0.0230	Contrastive Multi-View Representation Learning on Graphs
4	bnpool Hyb baseline	0.9100 ±0.0200	This paper
3	DMoN GNN baseline actual #3 vs paper #3	0.9100 ±0.0100	actual #3 vs paper #3 paper 0.9100 ±0.0100 prev best: Graph Pooling via Ricci Flow (TMLR 2024) paper 0.9100 ±0.0100 Δ at pub 0.0600
5	hosc GNN baseline	0.9100 ±0.0100	This paper
6	jbgnn GNN baseline	0.9100 ±0.0200	This paper
7	kmis Trad baseline	0.9100 ±0.0300	This paper
10	acc GNN baseline	0.9000 ±0.0300	This paper
8	Diff GNN baseline actual #8 vs paper #8	0.9000 ±0.0200	actual #8 vs paper #8 paper 0.9000 ±0.0200 prev best: Graph Pooling via Ricci Flow (TMLR 2024) paper 0.9000 ±0.0200 Δ at pub 0.0400
11	ecpool GNN baseline	0.9000 ±0.0200	This paper
12	eigen Trad baseline	0.9000 ±0.0300	This paper
13	Graclus Trad baseline	0.9000 ±0.0200	This paper
14	lapool GNN baseline	0.9000 ±0.0100	This paper
9	MinCut Trad baseline actual #9 vs paper #9	0.9000 ±0.0200	actual #9 vs paper #9 paper 0.9000 ±0.0200 prev best: Graph Pooling via Ricci Flow (TMLR 2024) paper 0.9000 ±0.0200 Δ at pub 0.0500
16	SEP GNN baseline	0.8900 ±0.0300	This paper
21	ASAP GNN baseline	0.8800 ±0.0300	This paper
23	maxcut Trad baseline	0.8700 ±0.0300	This paper
24	ndp GNN baseline	0.8700 ±0.0200	This paper
25	pan GNN baseline	0.8700 ±0.0500	This paper
26	sag GNN baseline	0.8700 ±0.0300	This paper
29	topk GNN baseline	0.8600 ±0.0300	This paper
49	GCN code ↗ GNN global	0.5653 ±0.0260	Local Virtual Nodes for Alleviating Over-Squashing in Graph Neural Networks code ↗

Heterophily Benchmark

Roman-empire

Global leaderboard →

#	Model	Accuracy ↑	Paper
1	Dir-Poly code ↗ GNN global	0.9451 ±0.0022	Graph Transductive Sharpening: Leveraging Unlabeled Predictions in Node Classification code ↗
2	D-CDE code ↗ GNN global	0.9387 ±0.0041	Distributed-Order Fractional Graph Operating Network code ↗
3	LargeScaleNet code ↗ Hyb global	0.9358 ±0.0024	Scale-aware Message Passing For Graph Node Classification code ↗
26	GraphSAGE code ↗ GNN global	0.9106 ±0.0027	Directed Semi-Simplicial Learning with Applications to Brain Activity Decoding code ↗
135	MLP code ↗ Trad global	0.6665 ±0.0047	HyperAggregation: Aggregating over Graph Edges with Hypernetworks code ↗
206	maxcut Trad baseline	0.3600 ±0.0200	This paper
212	kmis Trad baseline	0.3200 ±0.0100	This paper
216	ASAP GNN baseline	0.2700 ±0.0100	This paper
218	ndp GNN baseline	0.2300 ±0.0100	This paper
224	topk GNN baseline	0.2100 ±0.0200	This paper

Amazon-ratings

Global leaderboard →

#	Model	Accuracy ↑	Paper
1	UniGAP global	0.6375 ±0.0110	UniGAP: A Universal and Adaptive Graph Upsampling Approach to Mitigate Over-Smoothing in Node …
2	H2GCN + HalfHop GNN global	0.5961 ±0.0100	UniGAP: A Universal and Adaptive Graph Upsampling Approach to Mitigate Over-Smoothing in Node …
3	GraphSAGE + HalfHop GNN global	0.5897 ±0.0150	UniGAP: A Universal and Adaptive Graph Upsampling Approach to Mitigate Over-Smoothing in Node …
12	SAGE code ↗ GNN global	0.5569 ±0.0023	Graph Transductive Sharpening: Leveraging Unlabeled Predictions in Node Classification code ↗
84	MLP Trad global	0.5058	Understanding Heterophily for Graph Neural Networks
159	kmis Trad baseline	0.4400 ±0.0100	This paper
166	ndp GNN baseline	0.4300 ±0.0100	This paper
181	maxcut Trad baseline	0.4200 ±0.0100	This paper
188	ASAP GNN baseline	0.4000 ±0.0200	This paper
192	topk GNN baseline	0.3800 ±0.0100	This paper

Minesweeper

Global leaderboard →

#	Model	ROC-AUC ↑	Paper
1	BuNN code ↗ Hyb global	0.9899 ±0.0016	Bundle Neural Networks for message diffusion on graphs code ↗
2	DC-GNN GNN global	0.9850 ±0.0021	Differentiable Cluster Graph Neural Network
3	Tango -GPS global	0.9839 ±0.0054	TANGO: Graph Neural Dynamics via Learned Energy and Tangential Flows
42	GraphSAGE code ↗ GNN global	0.9358 ±0.0005	Gumbel-MPNN: Graph Rewiring with Gumbel-Softmax code ↗
122	ASAP GNN baseline	0.7800 ±0.0500	This paper
139	kmis Trad baseline	0.7100 ±0.0100	This paper
145	topk GNN baseline	0.6600 ±0.0200	This paper
147	maxcut Trad baseline	0.6300 ±0.0400	This paper
149	ndp GNN baseline	0.6200 ±0.0100	This paper
156	MLP Trad global	0.5099 ±0.0147	Graph Homophily Booster: Rethinking the Role of Discrete Features on Heterophilic Graphs

Tolokers

Global leaderboard →

#	Model	ROC-AUC ↑	Paper
1	Polynormer-r code ↗ GT global	0.8591 ±0.0074	Polynormer: Polynomial-Expressive Graph Transformer in Linear Time code ↗
2	DC-GNN GNN global	0.8588 ±0.0081	Differentiable Cluster Graph Neural Network
3	TANGO-GPS Hyb global	0.8566 ±0.0101	TANGO: Graph Neural Dynamics via Learned Energy and Tangential Flows
50	GCN GNN global	0.8364 ±0.0067	Beyond Oversquashing: Understanding Signal Propagation in GNNs Via Observables
95	maxcut Trad baseline	0.8000 ±0.0100	This paper
101	ASAP GNN baseline	0.7900 ±0.0100	This paper
128	kmis Trad baseline	0.7700 ±0.0500	This paper
139	MLP code ↗ Trad global	0.7362 ±0.0020	Gumbel-MPNN: Graph Rewiring with Gumbel-Softmax code ↗
148	ndp GNN baseline	0.7100 ±0.0600	This paper
151	topk GNN baseline	0.7000 ±0.0300	This paper

Questions

Global leaderboard →

#	Model	ROC-AUC ↑	Paper
1	NID_GAT code ↗ global	0.9703 ±0.0002	Node Identifiers: Compact, Discrete Representations for Efficient Graph Learning code ↗
2	NID_GCN code ↗ global	0.9685 ±0.0010	Node Identifiers: Compact, Discrete Representations for Efficient Graph Learning code ↗
3	UniGAP global	0.8421 ±0.0080	UniGAP: A Universal and Adaptive Graph Upsampling Approach to Mitigate Over-Smoothing in Node …
16	GCN GNN global	0.7902 ±0.0060	UniGAP: A Universal and Adaptive Graph Upsampling Approach to Mitigate Over-Smoothing in Node …
96	MLP code ↗ Trad global	0.7082 ±0.0010	Gumbel-MPNN: Graph Rewiring with Gumbel-Softmax code ↗
123	kmis Trad baseline	0.6600 ±0.0100	This paper
127	maxcut Trad baseline	0.6500 ±0.0200	This paper
133	ASAP GNN baseline	0.6200 ±0.0600	This paper
134	topk GNN baseline	0.6200 ±0.0300	This paper
139	ndp GNN baseline	0.5900 ±0.0600	This paper

SOTA performance (not significantly different from the best) Baseline / comparison (this paper) Global context (other papers) Best at pub — reported value improved over the best known when published