9i @SrSSKJr SSKJr SSKJr "SS5rg)zEnsemble model combining Elo ratings with ML predictions. The EnsembleModel takes Elo-based probabilities from the rating engine and blends them with additional ML predictions (from feature-engineered models) to produce a final probability estimate. ) annotations)Any)Sessionc\rSrSrSrS S SjjrS SjrSSjrS SjrSSjr SSjr S r g ) EnsembleModela+Combines Elo ratings with ML predictions for improved accuracy. Supports multiple blending strategies: - Weighted average (fixed weights learned offline) - Logistic calibration (train a logistic regression on top of Elo + ML features) - Stacking (use ML predictions as features for a meta-model) c*XlX l0Ulg)uInitialize the ensemble model. Args: session: Database session for loading data. blend_mode: Blending strategy — "weighted", "logistic", or "stacking". N)_session blend_mode_weights)selfsessionr s 9/root/tipsharks/tipsharks-elo-api/packages/ml/ensemble.py__init__EnsembleModel.__init__s $*, c[e)zTrain the ensemble (e.g., learn blend weights). Args: date_from: Start date string (YYYY-MM-DD). date_to: End date string (YYYY-MM-DD). Returns: Training metrics dict. NotImplementedErrorr date_fromdate_tos rtrainEnsembleModel.train)s "!rc[e)aiPredict win/place probabilities for all starters in a race. Combines Elo ratings with ML feature predictions. Args: race_id: The race ID. Returns: List of prediction dicts, one per starter, with keys: - starter_id: int - win_probability: float - place_probability: float - elo_win_prob: float - ml_win_prob: float (if applicable) r)r race_ids rpredictEnsembleModel.predict9s &"!rc[e)zEvaluate ensemble performance on a test window. Args: date_from: Start date string (YYYY-MM-DD). date_to: End date string (YYYY-MM-DD). Returns: Dict of evaluation metrics (accuracy, Brier score, log loss, top-3 rate, etc.). rrs revaluateEnsembleModel.evaluateNs "!rc[e)zdPersist learned blend weights to disk. Args: path: File path to save weights to (JSON format). rr paths r save_weightsEnsembleModel.save_weights_ "!rc[e)zeLoad learned blend weights from disk. Args: path: File path to load weights from (JSON format). rr#s r load_weightsEnsembleModel.load_weightsgr'r)r r r N)weighted)rrr strreturnNone)rr,rr,r-zdict[str, Any])rintr-zlist[dict[str, Any]])r$r,r-r.) __name__ __module__ __qualname____firstlineno____doc__rrrr r%r)__static_attributes__rrrrs%---  - """  " "" "*"""  """"rrN)r4 __future__rtypingrsqlalchemy.ormrrr6rrr:s!#"^"^"r