{"id":51169406,"url":"https://github.com/borgwardtlab/gelato","last_synced_at":"2026-06-26T23:03:20.315Z","repository":{"id":363349245,"uuid":"1151444668","full_name":"BorgwardtLab/Gelato","owner":"BorgwardtLab","description":"Code and Data for the paper: \"Gelato: Graph Edit Distance via Autoregressive Neural Combinatorial Optimization\" (ICLR 2026)","archived":false,"fork":false,"pushed_at":"2026-06-08T13:28:43.000Z","size":6535,"stargazers_count":4,"open_issues_count":0,"forks_count":0,"subscribers_count":0,"default_branch":"main","last_synced_at":"2026-06-08T15:25:12.377Z","etag":null,"topics":["gnns","graph-edit-distance","iclr2026"],"latest_commit_sha":null,"homepage":"","language":"Python","has_issues":true,"has_wiki":null,"has_pages":null,"mirror_url":null,"source_name":null,"license":"mit","status":null,"scm":"git","pull_requests_enabled":true,"icon_url":"https://github.com/BorgwardtLab.png","metadata":{"files":{"readme":"readme.md","changelog":null,"contributing":null,"funding":null,"license":"LICENSE","code_of_conduct":null,"threat_model":null,"audit":null,"citation":null,"codeowners":null,"security":null,"support":null,"governance":null,"roadmap":null,"authors":null,"dei":null,"publiccode":null,"codemeta":null,"zenodo":null,"notice":null,"maintainers":null,"copyright":null,"agents":null,"dco":null,"cla":null}},"created_at":"2026-02-06T13:28:16.000Z","updated_at":"2026-06-08T13:29:31.000Z","dependencies_parsed_at":null,"dependency_job_id":null,"html_url":"https://github.com/BorgwardtLab/Gelato","commit_stats":null,"previous_names":["borgwardtlab/gelato"],"tags_count":null,"template":false,"template_full_name":null,"purl":"pkg:github/BorgwardtLab/Gelato","repository_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/BorgwardtLab%2FGelato","tags_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/BorgwardtLab%2FGelato/tags","releases_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/BorgwardtLab%2FGelato/releases","manifests_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/BorgwardtLab%2FGelato/manifests","owner_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners/BorgwardtLab","download_url":"https://codeload.github.com/BorgwardtLab/Gelato/tar.gz/refs/heads/main","sbom_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/BorgwardtLab%2FGelato/sbom","scorecard":null,"host":{"name":"GitHub","url":"https://github.com","kind":"github","repositories_count":286080680,"owners_count":34835782,"icon_url":"https://github.com/github.png","version":null,"created_at":"2022-05-30T11:31:42.601Z","updated_at":"2026-05-26T15:22:16.424Z","status":"online","status_checked_at":"2026-06-26T02:00:06.560Z","response_time":106,"last_error":null,"robots_txt_status":"success","robots_txt_updated_at":"2025-07-24T06:49:26.215Z","robots_txt_url":"https://github.com/robots.txt","online":true,"can_crawl_api":true,"host_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub","repositories_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories","repository_names_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repository_names","owners_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners"}},"keywords":["gnns","graph-edit-distance","iclr2026"],"created_at":"2026-06-26T23:03:14.057Z","updated_at":"2026-06-26T23:03:20.305Z","avatar_url":"https://github.com/BorgwardtLab.png","language":"Python","funding_links":[],"categories":[],"sub_categories":[],"readme":"# Gelato\n\nDatasets and code for the paper: \"Gelato: Graph Edit Distance via Autoregressive Neural Combinatorial Optimization\", published at ICLR 2026. The paper is available [here](https://openreview.net/forum?id=6ZTcLNmguc).\n\nThe repository provides code for training and testing the Gelato model. \nMoreover, ```src/dataset.py``` contains a dataset class with the GED datasets used in the paper. \n\n### Training\n\nThe following commands can be used to train Gelato on the main datasets used in the paper.\n```\npython train.py --data aids --save_ckp checkpoints/model_aids.pt --train_pairs 88000\npython train.py --data linux --save_ckp checkpoints/model_linux.pt --train_pairs 25000\npython train.py --data imdb-16 --save_ckp checkpoints/model_imdb.pt --train_pairs 25000\npython train.py --data zinc-16 --save_ckp checkpoints/model_zinc.pt --train_pairs 125000\npython train.py --data molhiv-16 --save_ckp ckp/model_molhiv.pt --train_pairs 200000\npython train.py --data code2-22 --save_ckp ckp/model_code.pt --train_pairs 100000\n```\n\n### Testing\n\nCheckpoints for Gelato are available in the ```checkpoints/``` folder. \n\nExample usage for in-distribution testing: \n```\npython test.py --data zinc-16 --load_ckp checkpoints/model_zinc.pt\n```\nExample usage for out-of-distribution testing:\n```\npython test.py --data zinc-16 --load_ckp checkpoints/model_zinc.pt --split larger --size_bounds 17 18 --num_samples 500\n```\n\n## Datasets\n\nIn ```src/dataset.py```, we provide easy-to-use datasets with several improvements over existing ones:\n- We provide pre-computed train-val-test splits with **no data leakage** (due to graph isomorphism) across splits.\n- The datasets have both edge-labeled and edge-unlabeled variants of graphs.\n- We provide optimal solutions for graphs **up to 30 nodes** to test for out-of-distribution generalization, in the ```larger``` data split.\n\nExample usage:\n\n```python\nfrom src.dataset import GraphMatchingDataset\n\n# Get 1000 graph pairs from the test split of the AIDS dataset\ndataset = GraphMatchingDataset(name='aids', root='data/', num_pairs=1000, split='test')\n\n# Get 1000 graph pairs from the 'larger' split of the ZINC-16 dataset\ndataset = GraphMatchingDataset(name='zinc-16', root='data/', num_pairs=1000, split='larger')\n\n# Get 1000 graph pairs with graphs between 23 and 26 nodes from the 'larger' split of the code2-22 dataset\ndataset = GraphMatchingDataset(name='code2-22', root='data/', num_pairs=1000, split='larger', bounds=(23, 26))\n\nfor data in dataset:\n  graph_1 = Data(x=data.x_s, edge_index=data.edge_index_s, edge_attr=data.edge_attr_s)\n  graph_2 = Data(x=data.x_t, edge_index=data.edge_index_t, edge_attr=data.edge_attr_t)\n  optimal_matching = data.matching.long()\n```\n\n\n## Citing our work\n\nPlease cite our ICLR 2026 paper in case you find Gelato useful for your applications.\n\n\u003e Paolo Pellizzoni, Till Hendrik Schulz, and Karsten Borgwardt. _Gelato: Graph Edit Distance via Autoregressive Neural Combinatorial Optimization_, in ICLR, 2026.\n\n```\n@inproceedings{\n  pellizzoni2026gelato,\n  title={Gelato: Graph Edit Distance via Autoregressive Neural Combinatorial Optimization},\n  author={Paolo Pellizzoni and Till Hendrik Schulz and Karsten Borgwardt},\n  booktitle={International Conference on Learning Representations},\n  year={2026},\n}\n```\n","project_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fborgwardtlab%2Fgelato","html_url":"https://awesome.ecosyste.ms/projects/github.com%2Fborgwardtlab%2Fgelato","lists_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fborgwardtlab%2Fgelato/lists"}