Elastic-Net Logistic
Elastic-net logistic regressionA regularized linear classifier optimized for stable weights, interpretability, and calibrated probabilities.
Active V1 forecasts remain stable while V2 candidates face rolling, season-by-season evaluation.
A regularized linear classifier optimized for stable weights, interpretability, and calibrated probabilities.
A boosted-tree classifier designed to learn thresholds and nonlinear feature interactions.
A bagged-tree ensemble that provides robust nonlinear estimates and model diversity.
No V2.7 lock exists for this slate yet. Rally will show the selected team and probability here after the official cutoff; it will never backfill a pick after first pitch.
Pregame features and expanding windows.
Brier, log loss, calibration, stability.
Improvement across multiple windows.
Bounded derivatives; preserved lineage.