Top pick: DeepLINK-T + e-BH (knockpy + e-BH (MIT + open)). Fills the gap: FDR aggregation (meta) — which the production baseline (Pearson + Spearman + h_norm + Tier-2 MI-vs-target) cannot detect.
Lane = how usable the metric is today. Top lane = production-baseline duplicates (Pearson, Spearman). Below = permissive FOSS / GPL / unmaintained / code-released-but-not-packaged / no-FOSS. Per-circle text = first 3 letters of language (Pyt / R / Jav).
| Metric | Library | License | Lang | Maint | Complexity | Param-free? | URL |
|---|---|---|---|---|---|---|---|
| DeepLINK-T | Reference code with paper | TBD | Python (PyTorch/TF) | ⚠️ | O(T N hidden^2) per epoch | yes | ↗ |
| Derandomized Knockoffs (Ren-Barber JRSSB 2024) | Author code | varies | Python | ⚠️ | ? | ↗ | |
| MRMR | mrmr-selection | GPL | Python | ✅ | O(K N T) | yes | ↗ |
| Model-X knockoffs — R | knockoff | GPL | R | ✅ | ? | ↗ | |
| Stability Selection (Meinshausen-Buhlmann) | stabs / stability-selection | GPL / BSD | R / Python | ✅ | B x cost(selection) | ? | ↗ |
| ARFS (Leshy + BoostAGroota + GrootCV) | arfs | MIT | Python | ✅ | ? | ↗ | |
| Boruta | BorutaPy | MIT | Python | ✅ | tree-model dependent | ? | ↗ |
| Conformal Feature Selection (Jin-Candes 2023) | crepes / conformalInference | open | Python / R | ✅ | ? | ↗ | |
| Group Knockoffs | knockpy (set groups=) | MIT | Python | ✅ | ? | ↗ | |
| KOPI | aggregated_knockoffs | open | Python | ✅ | ? | ↗ | |
| Model-X knockoffs | knockpy | MIT | Python | ✅ | O(N^3) + stat | yes | ↗ |
| e-BH (Wang-Ramdas 2022) | e-BH | open | Python | ✅ | O(K log K) | no | ↗ |
| powershap | powershap | MIT | Python | ✅ | ? | ↗ | |
| TSKI | no PyPI — compose from knockpy + custom | n/a | Python | ⚠️ | O(B N^3) | yes | ↗ |