Text Classification
Joblib
Scikit-learn
English
scikit-learn
sklearn-logistic-regression
document-classification
binary-classification
legal-documents
hoa
property-management
ccr
declaration-of-covenants
logistic-regression
Eval Results (legacy)
Instructions to use GoverningDocs/ccr-binary-logreg with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Scikit-learn
How to use GoverningDocs/ccr-binary-logreg with Scikit-learn:
from huggingface_hub import hf_hub_download import joblib model = joblib.load( hf_hub_download("GoverningDocs/ccr-binary-logreg", "sklearn_model.joblib") ) # only load pickle files from sources you trust # read more about it here https://skops.readthedocs.io/en/stable/persistence.html - Notebooks
- Google Colab
- Kaggle
T18 Phase 1 Tier 1: config snapshot
Browse files- config.json +15 -4
config.json
CHANGED
|
@@ -6,9 +6,10 @@
|
|
| 6 |
"skip_boilerplate": true,
|
| 7 |
"operating_threshold": 0.436,
|
| 8 |
"decision_band": {
|
| 9 |
-
"reject_below": 0.
|
| 10 |
-
"fast_pass_at_or_above": 0.
|
| 11 |
-
"escalate_between": "(0.
|
|
|
|
| 12 |
},
|
| 13 |
"training_data": {
|
| 14 |
"source": "setfit_experiments PostgreSQL DB + multi-signal Phase 0 relabeling",
|
|
@@ -73,5 +74,15 @@
|
|
| 73 |
"mlp_05",
|
| 74 |
"mlp_tuned"
|
| 75 |
],
|
| 76 |
-
"winner_selection_rule": "max F1 across LogReg, LogReg+Platt, MLP at best threshold; simplicity tiebreak to LogReg"
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 77 |
}
|
|
|
|
| 6 |
"skip_boilerplate": true,
|
| 7 |
"operating_threshold": 0.436,
|
| 8 |
"decision_band": {
|
| 9 |
+
"reject_below": 0.25,
|
| 10 |
+
"fast_pass_at_or_above": 0.55,
|
| 11 |
+
"escalate_between": "(0.25, 0.55)",
|
| 12 |
+
"note": "Recalibrated empirically on production score distribution. Plan-time (0.30, 0.85) left FAST_PASS empty."
|
| 13 |
},
|
| 14 |
"training_data": {
|
| 15 |
"source": "setfit_experiments PostgreSQL DB + multi-signal Phase 0 relabeling",
|
|
|
|
| 74 |
"mlp_05",
|
| 75 |
"mlp_tuned"
|
| 76 |
],
|
| 77 |
+
"winner_selection_rule": "max F1 across LogReg, LogReg+Platt, MLP at best threshold; simplicity tiebreak to LogReg",
|
| 78 |
+
"calibrator": {
|
| 79 |
+
"filename": "ccr_binary_isotonic_calibrator.joblib",
|
| 80 |
+
"method": "isotonic_prefit",
|
| 81 |
+
"fit_split_seed": 42,
|
| 82 |
+
"fit_split_size": 64,
|
| 83 |
+
"test_ece_before": 0.27835753511850964,
|
| 84 |
+
"test_ece_after": 0.08663491157117499,
|
| 85 |
+
"shipped_model_filename": "ccr_binary_logreg_tuned.joblib",
|
| 86 |
+
"note": "Optional artifact. Produces approximately 3 plateau outputs (0.737, 0.833, 1.000) due to small (~70-doc) calibration set. Treat as 3-level confidence rather than fine-grained probability. See ISOTONIC_CALIBRATION_FINDINGS.md."
|
| 87 |
+
}
|
| 88 |
}
|