v2: bagged config

benchmark/tabarena/fela_ag_model.py

@@ -138,10 +138,21 @@ class FelaTabAGModel(AbstractModel):
             )
         n_classes = len(self.classes_) if hasattr(self, "classes_") else int(
             self._y_support.max() + 1)
-        proba = predict(
-            self._fela_model, self._X_support, self._y_support, X_np,
-            task="classification", n_classes=n_classes, support_cap=SUPPORT_CAP,
-        )
+        # n_bag>1: permutation-bagged ensemble (row + column order) — measured +1.4 AUC
+        # on adult at n_bag=8 vs single pass. Costs n_bag x inference latency.
+        n_bag = self._get_model_params().get("n_bag", 0)
+        if n_bag and n_bag > 1:
+            from modeling import predict_bagged
+
+            proba = predict_bagged(
+                self._fela_model, self._X_support, self._y_support, X_np,
+                "classification", n_classes, SUPPORT_CAP, n_bag=int(n_bag),
+            )
+        else:
+            proba = predict(
+                self._fela_model, self._X_support, self._y_support, X_np,
+                task="classification", n_classes=n_classes, support_cap=SUPPORT_CAP,
+            )
         # guard against NaN/inf or degenerate rows: AutoGluon drops NaN-pred rows when
         # scoring, which can leave a single class in y_true on small folds
         proba = np.asarray(proba, dtype=np.float64)
@@ -152,7 +163,7 @@ class FelaTabAGModel(AbstractModel):
         return self._convert_proba_to_unified_form(proba)
 
     def _set_default_params(self) -> None:
-        for param, val in {"tier": "small", "device": "cpu", "regression_backend": "lgbm"}.items():
+        for param, val in {"tier": "small", "device": "cpu", "regression_backend": "lgbm", "n_bag": 0}.items():
             self._set_default_param_value(param, val)
 
     def _get_default_auxiliary_params(self) -> dict:
@@ -170,6 +181,6 @@ class FelaTabAGModel(AbstractModel):
 
         return ConfigGenerator(
             model_cls=cls,
-            manual_configs=[{"tier": "small"}, {"tier": "big"}],
+            manual_configs=[{"tier": "small"}, {"tier": "big"}, {"tier": "big", "n_bag": 8}],
             search_space={},
         )