NCI1
Description
NCI1
Dataset Overview
Lifting Methods
Structural-based Liftings
- Cellular: Cycle-based lifting
- Simplicial: Clique complex lifting
- Hypergraph: k-hop lifting
Feature Lifting
- Projected sum
Model Performance
| Model | Accuracy (%) | Std Dev (±) |
|---|---|---|
| GCN | 72.86 | 0.69 |
| GIN | 74.26 | 0.96 |
| GAT | 75.00 | 0.99 |
| AST | 75.18 | 1.24 |
| EDGNN | 73.97 | 0.82 |
| UniGNN2 | 73.02 | 0.92 |
| CWN | 73.93 | 1.87 |
| CCCN | 76.67 | 1.48 |
| SCCNN | 76.60 | 1.75 |
| SCN | 74.49 | 1.03 |
Key Insights
- CCCN achieves the best performance with 76.67% accuracy
- SCCNN is a close second with 76.60% accuracy
- Most models perform in the 73–76% range
- CWN shows the highest variability (±1.87)
- GCN shows the most stable results with lowest std dev (±0.69)
Reproducibility
python -m topobench model=cell/cccn dataset=graph/NCI1 optimizer.parameters.lr=0.01 model.feature_encoder.out_channels=64 model.backbone.n_layers=4 model.readout.readout_name=NoReadOut model.feature_encoder.proj_dropout=0.5 dataset.dataloader_params.batch_size=128 transforms.graph2cell_lifting.max_cell_length=10 dataset.split_params.data_seed=0,3,5,7,9 trainer.max_epochs=500 trainer.min_epochs=50 trainer.check_val_every_n_epoch=5 callbacks.early_stopping.patience=10 logger.wandb.project=TopoBench --multirun