structlearning/isonetpp-benchmark

IsoNet++ Benchmark Dataset

The IsoNet++ Benchmark is a subgraph retrieval benchmark derived from TUDataset graph datasets including:

• AIDS
• MUTAG
• PTC (FM, FR, MM, MR)

The benchmark is used to evaluate models that learn graph representations for:

• Graph similarity search
• Subgraph matching
• Retrieval at scale

This benchmark was introduced to evaluate the IsoNet++ model.

---

Dataset Structure

isonetpp-benchmark/
├─ corpus/                      # Searchable  graph collections
│   ├─ aids240k_corpus_subgraphs.pkl
│   ├─ mutag240k_corpus_subgraphs.pkl
│   ├─ ptc_fm240k_corpus_subgraphs.pkl
│   ├─ ptc_fr240k_corpus_subgraphs.pkl
│   ├─ ptc_mm240k_corpus_subgraphs.pkl
│   └─ ptc_mr240k_corpus_subgraphs.pkl
└─ splits/                      # Query → relevance evaluation sets
    ├─ train/
    │   ├─ train_<dataset>_query_subgraphs.pkl
    │   └─ train_<dataset>_rel_nx_is_subgraph_iso.pkl
    ├─ val/
    │   ├─ val_<dataset>_query_subgraphs.pkl
    │   └─ val_<dataset>_rel_nx_is_subgraph_iso.pkl
    └─ test/
        ├─ test_<dataset>_query_subgraphs.pkl
        └─ test_<dataset>_rel_nx_is_subgraph_iso.pkl

Where <dataset> ∈ {aids240k, mutag240k, ptc_fm240k, ptc_fr240k, ptc_mm240k, ptc_mr240k}.

---

Data Format

All .pkl files use Python pickle serialization:

File Pattern	Description
*_corpus_subgraphs.pkl	List of NetworkX graphs representing the retrieval corpus
*_query_subgraphs.pkl	List of NetworkX graphs serving as query graphs
*_rel_nx_is_subgraph_iso.pkl	Binary labels from exact subgraph isomorphism (NetworkX VF2)

---

Load Examples

Load Corpus

from huggingface_hub import hf_hub_download
import pickle

path = hf_hub_download(
    "structlearning/isonetpp-benchmark",
    filename="large_dataset/corpus/aids240k_corpus_subgraphs.pkl",
    repo_type="dataset"
)
with open(path, "rb") as f:
    corpus_graphs = pickle.load(f)

Load Query Split

from huggingface_hub import hf_hub_download
import pickle

queries = pickle.load(open(
    hf_hub_download("structlearning/isonetpp-benchmark",
                    filename="large_dataset/splits/train/train_aids240k_query_subgraphs.pkl",
                    repo_type="dataset"),
    "rb"
))

labels = pickle.load(open(
    hf_hub_download("structlearning/isonetpp-benchmark",
                    filename="large_dataset/splits/train/train_aids240k_rel_nx_is_subgraph_iso.pkl",
                    repo_type="dataset"),
    "rb"
))

---

Intended Use

This dataset is suitable for:

• Graph retrieval model evaluation
• Learning subgraph-aware representations
• Benchmarking hashing, GNN-based retrieval systems
• Reproducing IsoNet++ results

---

Citation

If you use this dataset in research, please cite:

@inproceedings{ramachandraniteratively,
  title={Iteratively Refined Early Interaction Alignment for Subgraph Matching based Graph Retrieval},
  author={Ramachandran, Ashwin and Raj, Vaibhav and Roy, Indradyumna and Chakrabarti, Soumen and De, Abir},
  booktitle={The Thirty-eighth Annual Conference on Neural Information Processing Systems}
}

---

License

This dataset is released under CC-BY-4.0.