The dataset viewer is not available because its heuristics could not detect any supported data files. You can try uploading some data files, or configuring the data files location manually.
YAML Metadata Warning:empty or missing yaml metadata in repo card
Check out the documentation for more information.
Data directory — SA²GFM
Structure-Aware Semantic Augmentation for Robust Graph Foundation Models
This folder is the optional local data root for the official SA²GFM implementation in the parent repository (../README.md). SA²GFM is a graph foundation model that uses structure-aware semantic augmentation (communities → text → language-model features), a self-supervised information bottleneck for pretraining, and mixture-of-experts routing (with a null expert) for robust downstream adaptation.
The code (community detection, feature enhancement, pretraining, downstream MoE, attacks) lives under ../community_detection/, ../node_feature_enhance/, ../pretrain/, ../downstream/, and ../attack_data_gen/. This data/ tree only holds datasets and derived artifacts consumed by those modules.
Pointing the code at this folder
From the repository root:
export SA2GFM_DATA_ROOT="$(pwd)/data"
If SA2GFM_DATA_ROOT is unset, the project may default to this repo’s data/ when no sibling test_pretrain/data layout is detected—see path resolution in the codebase and the root README.
Expected layout under the data root
| Path | Role |
|---|---|
ori/{dataset}.pt |
PyTorch Geometric Data: at minimum x, edge_index, y. Pretrain and downstream expect enhanced_x_64 (or enhanced_x) after running node_feature_enhance. |
communities/{dataset}_communities.pt |
Produced by community_detection/ |
few_shot/{dataset}/{k}shot/split_*.pt |
Produced by few_shot_gen/ |
save_model/{name}.pt |
Expert checkpoints from pretrain/ |
Attack outputs typically live under ../attack_data_gen/outputs/ when you run the attack pipeline.
What is in this checkout right now
data/
└── ori/
├── arxiv.pt
├── citeseer.pt
├── cora.pt
├── pubmed.pt
├── P-home.pt
├── P-tech.pt
└── wikics.pt
These are canonical graph tensors for the listed datasets. Before running the full SA²GFM pipeline, generate communities, enhanced features, few-shot splits, and per-source experts as described in the root README.
Naming note
Graph files should be named {dataset}.pt (e.g. cora.pt) so scripts resolve ori/{dataset}.pt consistently. Duplicate-download names such as dataset (2).pt should be renamed to avoid path mismatches.
More documentation
- Full install, pipeline, expert–target table, and commands:
../README.md
- Downloads last month
- 3