Spaces:
Runtime error
Runtime error
Commit History
-
Improve mule classifier: native-NaN + missingness (CV PR-AUC 0.88->0.91, recall 13->15/16)
Aryan Singh +1: Claude Opus 4.8 (1M context)- builder.py: numerics kept as NaN for LightGBM native handling; add missingness indicators + row missing-count; IsolationForest uses a median-imputed copy - train.py: logistic baseline imputes before scaling (NaN-safe) - scoring.py: direction-aware narrative (cleared accounts explain what kept risk low) - rings.py: median-impute features before k-NN similarity graph (NaN-safe) - dashboard: center gauge value; remove fabricated channel/feed-source panels; add real risk-tier breakdown - regenerate model/pipeline/metadata/threshold/feature-list/rings/alert-feed Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>67eae2d
-
MuleGuard: end-to-end mule-account detection + HF Space deploy
MuleGuard +1: Claude Opus 4.8- ML pipeline: leakage-safe feature selection (3924->103) + anomaly fusion - Calibrated LightGBM, CV PR-AUC 0.882, test recall 0.81 @ precision 0.87 - FastAPI scoring backend + Streamlit analyst console (live scoring tab) - Real-time feed simulator, SHAP explainability, pitch deck, arch diagrams - Docker Space (backend+frontend) + GitHub Action auto-sync to HF Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>af879c2