ins-pricing 0.4.2__tar.gz → 0.4.3__tar.gz

{ins_pricing-0.4.2 → ins_pricing-0.4.3}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: ins_pricing
-Version: 0.4.2
+Version: 0.4.3
 Summary: Reusable modelling, pricing, governance, and reporting utilities.
 Author: meishi125478
 License: Proprietary

{ins_pricing-0.4.2 → ins_pricing-0.4.3}/ins_pricing/modelling/core/bayesopt/config_preprocess.py

@@ -195,6 +195,7 @@ class BayesOptConfig:
     cache_predictions: bool = False
     prediction_cache_dir: Optional[str] = None
     prediction_cache_format: str = "parquet"
+    dataloader_workers: Optional[int] = None
 
     def __post_init__(self) -> None:
         """Validate configuration after initialization."""
@@ -210,6 +211,12 @@ class BayesOptConfig:
             errors.append(
                 f"task_type must be one of {valid_task_types}, got '{self.task_type}'"
             )
+        if self.dataloader_workers is not None:
+            try:
+                if int(self.dataloader_workers) < 0:
+                    errors.append("dataloader_workers must be >= 0 when provided.")
+            except (TypeError, ValueError):
+                errors.append("dataloader_workers must be an integer when provided.")
         # Validate loss_name
         try:
             normalized_loss = normalize_loss_name(self.loss_name, self.task_type)

{ins_pricing-0.4.2 → ins_pricing-0.4.3}/ins_pricing/modelling/core/bayesopt/trainers/trainer_base.py

@@ -306,6 +306,19 @@ class TrainerBase:
         self.enable_distributed_optuna: bool = False
         self._distributed_forced_params: Optional[Dict[str, Any]] = None
 
+    def _apply_dataloader_overrides(self, model: Any) -> Any:
+        """Apply dataloader-related overrides from config to a model."""
+        cfg = getattr(self.ctx, "config", None)
+        if cfg is None:
+            return model
+        workers = getattr(cfg, "dataloader_workers", None)
+        if workers is not None:
+            model.dataloader_workers = int(workers)
+        profile = getattr(cfg, "resource_profile", None)
+        if profile:
+            model.resource_profile = str(profile)
+        return model
+
     def _export_preprocess_artifacts(self) -> Dict[str, Any]:
         dummy_columns: List[str] = []
         if getattr(self.ctx, "train_oht_data", None) is not None:

{ins_pricing-0.4.2 → ins_pricing-0.4.3}/ins_pricing/modelling/core/bayesopt/trainers/trainer_ft.py

@@ -163,6 +163,7 @@ class FTTrainer(TrainerBase):
             num_numeric_tokens=num_numeric_tokens,
             loss_name=loss_name,
         )
+        model = self._apply_dataloader_overrides(model)
         model.set_params(model_params)
         try:
             return float(model.fit_unsupervised(
@@ -248,7 +249,7 @@ class FTTrainer(TrainerBase):
                 requested_heads=params.get("n_heads")
             )
 
-            return FTTransformerSklearn(
+            model = FTTransformerSklearn(
                 model_nme=self.ctx.model_nme,
                 num_cols=self.ctx.num_features,
                 cat_cols=self.ctx.cate_list,
@@ -266,7 +267,10 @@ class FTTrainer(TrainerBase):
                 use_ddp=self.ctx.config.use_ft_ddp,
                 num_numeric_tokens=num_numeric_tokens,
                 loss_name=loss_name,
-            ).set_params({"_geo_params": geo_params_local} if geo_enabled else {})
+            )
+            model = self._apply_dataloader_overrides(model)
+            model.set_params({"_geo_params": geo_params_local} if geo_enabled else {})
+            return model
 
         def fit_predict(model, X_train, y_train, w_train, X_val, y_val, w_val, trial_obj):
             geo_train = geo_val = None
@@ -353,6 +357,7 @@ class FTTrainer(TrainerBase):
                 weight_decay=float(resolved_params.get("weight_decay", 0.0)),
                 loss_name=loss_name,
             )
+            tmp_model = self._apply_dataloader_overrides(tmp_model)
             tmp_model.set_params(resolved_params)
             geo_train_full = self.ctx.train_geo_tokens
             geo_train = None if geo_train_full is None else geo_train_full.iloc[train_idx]
@@ -387,6 +392,7 @@ class FTTrainer(TrainerBase):
             weight_decay=float(resolved_params.get("weight_decay", 0.0)),
             loss_name=loss_name,
         )
+        self.model = self._apply_dataloader_overrides(self.model)
         if refit_epochs is not None:
             self.model.epochs = int(refit_epochs)
         self.model.set_params(resolved_params)
@@ -460,6 +466,7 @@ class FTTrainer(TrainerBase):
                 weight_decay=float(resolved_params.get("weight_decay", 0.0)),
                 loss_name=loss_name,
             )
+            model = self._apply_dataloader_overrides(model)
             model.set_params(resolved_params)
 
             geo_train = geo_val = None
@@ -565,6 +572,7 @@ class FTTrainer(TrainerBase):
             num_numeric_tokens=self._resolve_numeric_tokens(),
             loss_name=loss_name,
         )
+        model = self._apply_dataloader_overrides(model)
         adaptive_heads, heads_adjusted = self._resolve_adaptive_heads(
             d_model=resolved_params.get("d_model", model.d_model),
             requested_heads=resolved_params.get("n_heads"),
@@ -728,6 +736,7 @@ class FTTrainer(TrainerBase):
             num_numeric_tokens=self._resolve_numeric_tokens(),
             loss_name=loss_name,
         )
+        self.model = self._apply_dataloader_overrides(self.model)
         resolved_params = dict(params or {})
         # Reuse supervised tuning structure params unless explicitly overridden.
         if not resolved_params and self.best_params:
@@ -797,8 +806,4 @@ class FTTrainer(TrainerBase):
             self.model,
             pred_prefix=pred_prefix,
             predict_kwargs_train=predict_kwargs_train,
-            predict_kwargs_test=predict_kwargs_test
-        )
-
-
-# =============================================================================
+          

{ins_pricing-0.4.2 → ins_pricing-0.4.3}/ins_pricing/modelling/core/bayesopt/trainers/trainer_gnn.py

@@ -59,7 +59,7 @@ class GNNTrainer(TrainerBase):
                 self.ctx.config.gnn_knn_gpu_mem_overhead),
             loss_name=loss_name,
         )
-        return model
+        return self._apply_dataloader_overrides(model)
 
     def cross_val(self, trial: optuna.trial.Trial) -> float:
         base_tw_power = self.ctx.default_tweedie_power()

{ins_pricing-0.4.2 → ins_pricing-0.4.3}/ins_pricing/modelling/core/bayesopt/trainers/trainer_resn.py

@@ -45,7 +45,7 @@ class ResNetTrainer(TrainerBase):
                 getattr(self.ctx.config, "resn_weight_decay", 1e-4),
             )
         )
-        return ResNetSklearn(
+        model = ResNetSklearn(
             model_nme=self.ctx.model_nme,
             input_dim=self._resolve_input_dim(),
             hidden_dim=int(params.get("hidden_dim", 64)),
@@ -64,6 +64,7 @@ class ResNetTrainer(TrainerBase):
             use_ddp=self.ctx.config.use_resn_ddp,
             loss_name=loss_name
         )
+        return self._apply_dataloader_overrides(model)
 
     # ========= Cross-validation (for BayesOpt) =========
     def cross_val(self, trial: optuna.trial.Trial) -> float:

{ins_pricing-0.4.2 → ins_pricing-0.4.3}/ins_pricing/modelling/core/bayesopt/utils/torch_trainer_mixin.py

@@ -232,6 +232,14 @@ class TorchTrainerMixin:
         """Determine number of DataLoader workers."""
         if os.name == 'nt':
             return 0
+        override = getattr(self, "dataloader_workers", None)
+        if override is None:
+            override = os.environ.get("BAYESOPT_DATALOADER_WORKERS")
+        if override is not None:
+            try:
+                return max(0, int(override))
+            except (TypeError, ValueError):
+                pass
         if getattr(self, "is_ddp_enabled", False):
             return 0
         profile = profile or self._resolve_resource_profile()

{ins_pricing-0.4.2 → ins_pricing-0.4.3}/ins_pricing/setup.py

@@ -3,7 +3,7 @@ from setuptools import setup, find_packages
 
 setup(
     name="ins_pricing",
-    version="0.4.2",
+    version="0.4.3",
     description="Reusable modelling, pricing, governance, and reporting utilities.",
     author="meishi125478",
     license="Proprietary",

{ins_pricing-0.4.2 → ins_pricing-0.4.3}/ins_pricing.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: ins_pricing
-Version: 0.4.2
+Version: 0.4.3
 Summary: Reusable modelling, pricing, governance, and reporting utilities.
 Author: meishi125478
 License: Proprietary

{ins_pricing-0.4.2 → ins_pricing-0.4.3}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "ins_pricing"
-version = "0.4.2"
+version = "0.4.3"
 description = "Reusable modelling, pricing, governance, and reporting utilities."
 readme = "README.md"
 requires-python = ">=3.9"