Spaces:
Running
Running
Repurpose as ICML-2026 repro logbook: An Odd Estimator for Shapley Values (arXiv:2602.01399, xILwgiWAUk)
19dba42 verified | title: Reproduction - An Odd Estimator for Shapley Values | |
| emoji: π | |
| colorFrom: yellow | |
| colorTo: red | |
| sdk: static | |
| pinned: false | |
| short_description: All 6 claims audited, arXiv 2602.01399 | |
| tags: | |
| - trackio | |
| - open-reproductions | |
| - icml2026-repro | |
| - paper-xILwgiWAUk | |
| # Reproduction bundle | |
| Reproduction of **"An Odd Estimator for Shapley Values"** (arXiv:2602.01399v1, | |
| OpenReview `xILwgiWAUk`). | |
| Claims 1-3 are exact algebraic identities (Observation 3.1, Theorem 3.2, | |
| Theorem 3.5). They are settled by exhaustive enumeration of all 2^d coalitions | |
| and by solving the constrained Fourier regression through its exact KKT system β | |
| no sampling, no estimator β and each is paired with a negative control that must | |
| fail. | |
| Claims 4-6 are assertions about Table 1 / Figure 2. They are audited against the | |
| published table, transcribed from a SHA-256-pinned PDF. | |
| ```bash | |
| python3 exp_odd.py # claims 1-3 | |
| python3 exp_table1.py # claims 4-6 | |
| python3 build_pages.py # regenerate pages/ | |
| ``` | |
| Requires numpy and pypdf. | |