Upload FROZEN_ARTIFACTS_README.md
Browse files- FROZEN_ARTIFACTS_README.md +21 -0
FROZEN_ARTIFACTS_README.md
ADDED
|
@@ -0,0 +1,21 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
# Frozen JSON Artifacts for `make reproduce-paper`
|
| 2 |
+
|
| 3 |
+
This directory contains the canonical aggregated JSON files that the paper's
|
| 4 |
+
tables and figures are computed from. Bundling these allows reviewers to
|
| 5 |
+
regenerate paper-facing artifacts without re-running 40 GPU-hours of experiments.
|
| 6 |
+
|
| 7 |
+
| File | Source | Used by |
|
| 8 |
+
|---|---|---|
|
| 9 |
+
| `rd5_fullscale_aggregated.json` | `results/rd5_fullscale/aggregated/aggregated_results.json` | Table 1 (3 strategies), Fig 1 |
|
| 10 |
+
| `tabdpt_probing_3seed.json` | `results/phase7/tabdpt_probing/aggregated_3seed.json` | Table 1 TabDPT row, Fig 1 |
|
| 11 |
+
| `tabdpt_causal_3seed.json` | `results/phase7/tabdpt_causal/aggregated_3seed.json` | §4.4 TabDPT causal claim |
|
| 12 |
+
| `nam_holdout.json` | `results/phase9_nam_holdout/nam_holdout_results.json` | §4.3 NAM holdout |
|
| 13 |
+
| `lofo_primary_endpoint.json` | `results/phase8a/nonlinear_probing/lofo_primary_endpoint.json` | Appendix LOFO table |
|
| 14 |
+
| `c1c2_baselines.json` | `results/neurips/c1c2_baselines.json` | §4.5 shuffled-label control |
|
| 15 |
+
| `scale_10k_multiseed.json` | `results/scale_10k_multiseed/results.json` | Appendix G N=10K scaling |
|
| 16 |
+
| `tabpfn25_fullscale_aggregated.json` | `results/rd6/tabpfn25_fullscale/aggregated/*.json` | Table 7 (v2 vs v2.5) |
|
| 17 |
+
|
| 18 |
+
To regenerate paper figures/tables from these artifacts without GPU:
|
| 19 |
+
```bash
|
| 20 |
+
make reproduce-paper
|
| 21 |
+
```
|