Datasets:
Cross-link to bootstrap_with_dense_weekly/ sibling
Browse files
imputation/bootstrap/README.md
CHANGED
|
@@ -81,6 +81,14 @@ print(meta["seed"], meta["n_boot"], len(meta["methods"]))
|
|
| 81 |
|
| 82 |
See [`SCHEMA.md`](SCHEMA.md) for the full column spec.
|
| 83 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 84 |
## Uploaded with
|
| 85 |
|
| 86 |
`tools/upload_leaderboard_bootstrap.py` in the
|
|
|
|
| 81 |
|
| 82 |
See [`SCHEMA.md`](SCHEMA.md) for the full column spec.
|
| 83 |
|
| 84 |
+
## Sibling: 17-method variant
|
| 85 |
+
|
| 86 |
+
`imputation/bootstrap_with_dense_weekly/` holds the same reducer applied
|
| 87 |
+
to a 17-method pool that adds `lsm2_weekly` (dense 7-day LSM-2). Skill /
|
| 88 |
+
fairness numbers per method are identical to this canonical variant
|
| 89 |
+
(they're pairwise vs `locf`); only average-rank values shift because the
|
| 90 |
+
comparison pool grew. See that dir's README for when to use which.
|
| 91 |
+
|
| 92 |
## Uploaded with
|
| 93 |
|
| 94 |
`tools/upload_leaderboard_bootstrap.py` in the
|