Skip to content

Latest commit

 

History

History
42 lines (35 loc) · 2.31 KB

README.md

File metadata and controls

42 lines (35 loc) · 2.31 KB
Raw

MSGC

Code for KBS paper "Multiple Sparse Graphs Condensation".

Abstract—The growing graph scale presents challenges for applying graph neural networks (GNNs) practically. Recently, Graph condensation (GCond) was proposed to condense the original large-scale graph into a small-scale graph. The goal is to make GNNs trained on the condensed graph achieve the performance close to those trained on the original graph. GCond achieves satisfactory performance on some datasets. However, GCond uses a single fully connected graph to model the condensed graph, which limits the diversity of embeddings obtained, especially when there are few synthetic nodes. We propose Multiple Sparse Graphs Condensation (MSGC), which condenses the original large-scale graph into multiple small-scale sparse graphs. MSGC takes standard neighborhood patterns as the basic substructures and can construct various connection schemes. Correspondingly, GNNs can obtain multiple sets of embeddings, which significantly enriches the diversity of embeddings. Experiments show that compared with GCond and other baselines, MSGC has significant advantages under the same condensed graph scale. MSGC can retain nearly 100% performance on Flickr and Citeseer datasets while reducing their graph scale by over 99.0%.

Environment configuration

  • python 3.8
  • torch 1.12.1+cu113
  • torch_geometric 2.1.0
  • torch_sparse 0.6.15
  • torch_scatter 2.0.9
  • prettytable 3.4.1
  • sklearn 1.1.2
  • tqdm 4.64.0
  • numpy 1.21.5

Dataset

In utils_data.py, change argument root in function load_data to your path.

The dataset will be downloaded automatically. If the download fails, you can view the source code of torch_geometric.datasets and update the url.

Run

python -u main.py --dataset=cora --basic-model=SGC --val-model=GCN --n-syn=35 --batch-size-syn=16--num-run=5 --val-run=5

  • basic-model: The relay model in gradient matching loss calculation.
  • val-model: The model to evaluate the condensed graph.
  • n-syn: The number of synthetic nodes in the condensed graph.
  • batch-size-syn: The number of synthetic adjacency matrices in the condensed graph.
  • num-run: To run the condensation process multiple times.
  • val-run: To evalue the condensed graph multiple times.

Modify parameters.py for more detailed setup.