Trim the artifact-gap note from the drift README
Browse files
evaluation/drift/CADE/SVM_validation/README.md
CHANGED
|
@@ -51,16 +51,6 @@ All four columns — `ip, vendor, device_type, class_drift_score` — are
|
|
| 51 |
outputs. The correct/incorrect split itself is the only judgment column
|
| 52 |
(implied by the file membership).
|
| 53 |
|
| 54 |
-
## Known artifact gap (honest note)
|
| 55 |
-
|
| 56 |
-
The per-IP *predicted* labels produced during Stage 2 were not retained in
|
| 57 |
-
this artifact, and the script that generated the correct/incorrect split was
|
| 58 |
-
not committed. The split reproduces the paper's reported percentage
|
| 59 |
-
(773/814), and its criterion is as documented above, but reviewers wishing to
|
| 60 |
-
audit Stage 2 per-IP need the predictions. Regenerating `correct_ips.csv` /
|
| 61 |
-
`incorrect_ips.csv` with an explicit `label_pred` column (and committing the
|
| 62 |
-
generator) is the planned follow-up.
|
| 63 |
-
|
| 64 |
## Quick reproduction
|
| 65 |
|
| 66 |
```python
|
|
|
|
| 51 |
outputs. The correct/incorrect split itself is the only judgment column
|
| 52 |
(implied by the file membership).
|
| 53 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 54 |
## Quick reproduction
|
| 55 |
|
| 56 |
```python
|