Datasetpublic

rakshi-the-neural-nexus/repro-grace-artifacts

Artifacts — reproduction of GRACE (ICML 2026, OpenReview tSZaHvpxCd) Raw outputs for the logbook at https://huggingface.co/spaces/rakshi-the-neural-nexus/repro-gradient-based-causal-tree-ensembles-hte File What it is run_grace.py the driver that produced every cell: GRACE + four scikit-learn baselines (claim 1) and the nn.Linear -> GRACE_layer swap (claim 2) results/grace.jsonl one line per (method, dataset, seed) cell, 60 cells, as written during the run… See the full description on the dataset page: https://huggingface.co/datasets/rakshi-the-neural-nexus/repro-grace-artifacts.

sourceHugging Facemitupdated 2mo agoView on Hugging Face

0likes8downloads

Dataset Card

Artifacts — reproduction of GRACE (ICML 2026, OpenReview `tSZaHvpxCd`)

Raw outputs for the logbook at https://huggingface.co/spaces/rakshi-the-neural-nexus/repro-gradient-based-causal-tree-ensembles-hte

File	What it is
`run_grace.py`	the driver that produced every cell: GRACE + four scikit-learn baselines (claim 1) and the `nn.Linear` -> `GRACE_layer` swap (claim 2)
`results/grace.jsonl`	one line per (method, dataset, seed) cell, 60 cells, as written during the run
`results/summary_pehe.csv`	mean and standard error of sqrt-PEHE per (dataset, method)

Upstream code: https://github.com/ysk-kano/GRACE (unmodified; max_steps reduced from the config default 50,000 to 3,000, which handicaps GRACE rather than flattering it). Data is generated locally by the repository's own data/data_generator.py. CPU only, 60 cells, about 12 minutes wall on one 24-thread laptop CPU.