improve model training

scripts/run_rain_ml_worker.py

@@ -54,6 +54,7 @@ class WorkerConfig:
     tune_hyperparameters: bool
     max_hyperparam_trials: int
     calibration_methods: str
+    threshold_policy: str
     walk_forward_folds: int
     allow_empty_data: bool
     dataset_path_template: str
@@ -194,6 +195,8 @@ def run_training_cycle(cfg: WorkerConfig, env: dict[str, str]) -> None:
         str(cfg.max_hyperparam_trials),
         "--calibration-methods",
         cfg.calibration_methods,
+        "--threshold-policy",
+        cfg.threshold_policy,
         "--walk-forward-folds",
         str(cfg.walk_forward_folds),
         "--feature-set",
@@ -300,6 +303,7 @@ def load_config() -> WorkerConfig:
         tune_hyperparameters=read_env_bool("RAIN_TUNE_HYPERPARAMETERS", False),
         max_hyperparam_trials=read_env_int("RAIN_MAX_HYPERPARAM_TRIALS", 12),
         calibration_methods=read_env("RAIN_CALIBRATION_METHODS", "none,sigmoid,isotonic"),
+        threshold_policy=read_env("RAIN_THRESHOLD_POLICY", "validation"),
         walk_forward_folds=read_env_int("RAIN_WALK_FORWARD_FOLDS", 0),
         allow_empty_data=read_env_bool("RAIN_ALLOW_EMPTY_DATA", True),
         dataset_path_template=read_env(
@@ -341,6 +345,7 @@ def main() -> int:
         f"train_interval_hours={cfg.train_interval_hours} "
         f"predict_interval_minutes={cfg.predict_interval_minutes} "
         f"tune_hyperparameters={cfg.tune_hyperparameters} "
+        f"threshold_policy={cfg.threshold_policy} "
         f"walk_forward_folds={cfg.walk_forward_folds} "
         f"allow_empty_data={cfg.allow_empty_data} "
         f"model_backup_path={cfg.model_backup_path}",