ins-pricing 0.3.3__tar.gz → 0.3.4__tar.gz

{ins_pricing-0.3.3 → ins_pricing-0.3.4}/PKG-INFO

@@ -1,162 +1,162 @@
-Metadata-Version: 2.4
-Name: ins_pricing
-Version: 0.3.3
-Summary: Reusable modelling, pricing, governance, and reporting utilities.
-Author: meishi125478
-License: Proprietary
-Keywords: pricing,insurance,bayesopt,ml
-Classifier: Programming Language :: Python :: 3
-Classifier: Programming Language :: Python :: 3 :: Only
-Classifier: Programming Language :: Python :: 3.9
-Classifier: License :: Other/Proprietary License
-Classifier: Operating System :: OS Independent
-Classifier: Intended Audience :: Developers
-Requires-Python: >=3.9
-Description-Content-Type: text/markdown
-Requires-Dist: numpy>=1.20
-Requires-Dist: pandas>=1.4
-Provides-Extra: bayesopt
-Requires-Dist: torch>=1.13; extra == "bayesopt"
-Requires-Dist: optuna>=3.0; extra == "bayesopt"
-Requires-Dist: xgboost>=1.6; extra == "bayesopt"
-Requires-Dist: scikit-learn>=1.1; extra == "bayesopt"
-Requires-Dist: statsmodels>=0.13; extra == "bayesopt"
-Requires-Dist: joblib>=1.2; extra == "bayesopt"
-Requires-Dist: matplotlib>=3.5; extra == "bayesopt"
-Provides-Extra: plotting
-Requires-Dist: matplotlib>=3.5; extra == "plotting"
-Requires-Dist: scikit-learn>=1.1; extra == "plotting"
-Provides-Extra: explain
-Requires-Dist: torch>=1.13; extra == "explain"
-Requires-Dist: shap>=0.41; extra == "explain"
-Requires-Dist: scikit-learn>=1.1; extra == "explain"
-Provides-Extra: geo
-Requires-Dist: contextily>=1.3; extra == "geo"
-Requires-Dist: matplotlib>=3.5; extra == "geo"
-Provides-Extra: gnn
-Requires-Dist: torch>=1.13; extra == "gnn"
-Requires-Dist: pynndescent>=0.5; extra == "gnn"
-Requires-Dist: torch-geometric>=2.3; extra == "gnn"
-Provides-Extra: all
-Requires-Dist: torch>=1.13; extra == "all"
-Requires-Dist: optuna>=3.0; extra == "all"
-Requires-Dist: xgboost>=1.6; extra == "all"
-Requires-Dist: scikit-learn>=1.1; extra == "all"
-Requires-Dist: statsmodels>=0.13; extra == "all"
-Requires-Dist: joblib>=1.2; extra == "all"
-Requires-Dist: matplotlib>=3.5; extra == "all"
-Requires-Dist: shap>=0.41; extra == "all"
-Requires-Dist: contextily>=1.3; extra == "all"
-Requires-Dist: pynndescent>=0.5; extra == "all"
-Requires-Dist: torch-geometric>=2.3; extra == "all"
-
-# Insurance-Pricing
-
-A reusable toolkit for insurance modeling, pricing, governance, and reporting.
-
-## Overview
-
-Insurance-Pricing (ins_pricing) is an enterprise-grade Python library designed for machine learning model training, pricing calculations, and model governance workflows in the insurance industry.
-
-### Core Modules
-
-| Module | Description |
-|--------|-------------|
-| **modelling** | ML model training (GLM, XGBoost, ResNet, FT-Transformer, GNN) and model interpretability (SHAP, permutation importance) |
-| **pricing** | Factor table construction, numeric binning, premium calibration, exposure calculation, PSI monitoring |
-| **production** | Model prediction, batch scoring, data drift detection, production metrics monitoring |
-| **governance** | Model registry, version management, approval workflows, audit logging |
-| **reporting** | Report generation (Markdown format), report scheduling |
-| **utils** | Data validation, performance profiling, device management, logging configuration |
-
-### Quick Start
-
-```python
-# Model training with Bayesian optimization
-from ins_pricing import bayesopt as ropt
-
-model = ropt.BayesOptModel(
-    train_data, test_data,
-    model_name='my_model',
-    resp_nme='target',
-    weight_nme='weight',
-    factor_nmes=feature_list,
-    cate_list=categorical_features,
-)
-model.bayesopt_xgb(max_evals=100)      # Train XGBoost
-model.bayesopt_resnet(max_evals=50)    # Train ResNet
-model.bayesopt_ft(max_evals=50)        # Train FT-Transformer
-
-# Pricing: build factor table
-from ins_pricing.pricing import build_factor_table
-factors = build_factor_table(
-    df,
-    factor_col='age_band',
-    loss_col='claim_amount',
-    exposure_col='exposure',
-)
-
-# Production: batch scoring
-from ins_pricing.production import batch_score
-scores = batch_score(model.trainers['xgb'].predict, df)
-
-# Model governance
-from ins_pricing.governance import ModelRegistry
-registry = ModelRegistry('models.json')
-registry.register(model_name, version, metrics=metrics)
-```
-
-### Project Structure
-
-```
-ins_pricing/
-├── cli/                    # Command-line entry points
-├── modelling/
-│   ├── core/bayesopt/     # ML model training core
-│   ├── explain/           # Model interpretability
-│   └── plotting/          # Model visualization
-├── pricing/               # Insurance pricing module
-├── production/            # Production deployment module
-├── governance/            # Model governance
-├── reporting/             # Report generation
-├── utils/                 # Utilities
-└── tests/                 # Test suite
-```
-
-### Installation
-
-```bash
-# Basic installation
-pip install ins_pricing
-
-# Full installation (all optional dependencies)
-pip install ins_pricing[all]
-
-# Install specific extras
-pip install ins_pricing[bayesopt]    # Model training
-pip install ins_pricing[explain]     # Model explanation
-pip install ins_pricing[plotting]    # Visualization
-pip install ins_pricing[gnn]         # Graph neural networks
-```
-
-#### Multi-platform & GPU installation notes
-
-- **PyTorch (CPU/GPU/MPS)**: Install the correct PyTorch build for your platform/GPU first (CUDA on
-  Linux/Windows, ROCm on supported AMD platforms, or MPS on Apple Silicon). Then install the
-  optional extras you need (e.g., `bayesopt`, `explain`, or `gnn`). This avoids pip pulling a
-  mismatched wheel.
-- **Torch Geometric (GNN)**: `torch-geometric` often requires platform-specific wheels (e.g.,
-  `torch-scatter`, `torch-sparse`). Follow the official PyG installation instructions for your
-  CUDA/ROCm/CPU environment, then install `ins_pricing[gnn]`.
-- **Multi-GPU**: Training code will use CUDA when available and can enable multi-GPU via
-  `torch.distributed`/`DataParallel` where supported. On Windows, CUDA DDP is not supported and will
-  fall back to single-GPU or DataParallel where possible.
-
-### Requirements
-
-- Python >= 3.9
-- Core dependencies: numpy >= 1.20, pandas >= 1.4
-
-### License
-
-Proprietary
+Metadata-Version: 2.4
+Name: ins_pricing
+Version: 0.3.4
+Summary: Reusable modelling, pricing, governance, and reporting utilities.
+Author: meishi125478
+License: Proprietary
+Keywords: pricing,insurance,bayesopt,ml
+Classifier: Programming Language :: Python :: 3
+Classifier: Programming Language :: Python :: 3 :: Only
+Classifier: Programming Language :: Python :: 3.9
+Classifier: License :: Other/Proprietary License
+Classifier: Operating System :: OS Independent
+Classifier: Intended Audience :: Developers
+Requires-Python: >=3.9
+Description-Content-Type: text/markdown
+Requires-Dist: numpy>=1.20
+Requires-Dist: pandas>=1.4
+Provides-Extra: bayesopt
+Requires-Dist: torch>=1.13; extra == "bayesopt"
+Requires-Dist: optuna>=3.0; extra == "bayesopt"
+Requires-Dist: xgboost>=1.6; extra == "bayesopt"
+Requires-Dist: scikit-learn>=1.1; extra == "bayesopt"
+Requires-Dist: statsmodels>=0.13; extra == "bayesopt"
+Requires-Dist: joblib>=1.2; extra == "bayesopt"
+Requires-Dist: matplotlib>=3.5; extra == "bayesopt"
+Provides-Extra: plotting
+Requires-Dist: matplotlib>=3.5; extra == "plotting"
+Requires-Dist: scikit-learn>=1.1; extra == "plotting"
+Provides-Extra: explain
+Requires-Dist: torch>=1.13; extra == "explain"
+Requires-Dist: shap>=0.41; extra == "explain"
+Requires-Dist: scikit-learn>=1.1; extra == "explain"
+Provides-Extra: geo
+Requires-Dist: contextily>=1.3; extra == "geo"
+Requires-Dist: matplotlib>=3.5; extra == "geo"
+Provides-Extra: gnn
+Requires-Dist: torch>=1.13; extra == "gnn"
+Requires-Dist: pynndescent>=0.5; extra == "gnn"
+Requires-Dist: torch-geometric>=2.3; extra == "gnn"
+Provides-Extra: all
+Requires-Dist: torch>=1.13; extra == "all"
+Requires-Dist: optuna>=3.0; extra == "all"
+Requires-Dist: xgboost>=1.6; extra == "all"
+Requires-Dist: scikit-learn>=1.1; extra == "all"
+Requires-Dist: statsmodels>=0.13; extra == "all"
+Requires-Dist: joblib>=1.2; extra == "all"
+Requires-Dist: matplotlib>=3.5; extra == "all"
+Requires-Dist: shap>=0.41; extra == "all"
+Requires-Dist: contextily>=1.3; extra == "all"
+Requires-Dist: pynndescent>=0.5; extra == "all"
+Requires-Dist: torch-geometric>=2.3; extra == "all"
+
+# Insurance-Pricing
+
+A reusable toolkit for insurance modeling, pricing, governance, and reporting.
+
+## Overview
+
+Insurance-Pricing (ins_pricing) is an enterprise-grade Python library designed for machine learning model training, pricing calculations, and model governance workflows in the insurance industry.
+
+### Core Modules
+
+| Module | Description |
+|--------|-------------|
+| **modelling** | ML model training (GLM, XGBoost, ResNet, FT-Transformer, GNN) and model interpretability (SHAP, permutation importance) |
+| **pricing** | Factor table construction, numeric binning, premium calibration, exposure calculation, PSI monitoring |
+| **production** | Model prediction, batch scoring, data drift detection, production metrics monitoring |
+| **governance** | Model registry, version management, approval workflows, audit logging |
+| **reporting** | Report generation (Markdown format), report scheduling |
+| **utils** | Data validation, performance profiling, device management, logging configuration |
+
+### Quick Start
+
+```python
+# Model training with Bayesian optimization
+from ins_pricing import bayesopt as ropt
+
+model = ropt.BayesOptModel(
+    train_data, test_data,
+    model_name='my_model',
+    resp_nme='target',
+    weight_nme='weight',
+    factor_nmes=feature_list,
+    cate_list=categorical_features,
+)
+model.bayesopt_xgb(max_evals=100)      # Train XGBoost
+model.bayesopt_resnet(max_evals=50)    # Train ResNet
+model.bayesopt_ft(max_evals=50)        # Train FT-Transformer
+
+# Pricing: build factor table
+from ins_pricing.pricing import build_factor_table
+factors = build_factor_table(
+    df,
+    factor_col='age_band',
+    loss_col='claim_amount',
+    exposure_col='exposure',
+)
+
+# Production: batch scoring
+from ins_pricing.production import batch_score
+scores = batch_score(model.trainers['xgb'].predict, df)
+
+# Model governance
+from ins_pricing.governance import ModelRegistry
+registry = ModelRegistry('models.json')
+registry.register(model_name, version, metrics=metrics)
+```
+
+### Project Structure
+
+```
+ins_pricing/
+├── cli/                    # Command-line entry points
+├── modelling/
+│   ├── core/bayesopt/     # ML model training core
+│   ├── explain/           # Model interpretability
+│   └── plotting/          # Model visualization
+├── pricing/               # Insurance pricing module
+├── production/            # Production deployment module
+├── governance/            # Model governance
+├── reporting/             # Report generation
+├── utils/                 # Utilities
+└── tests/                 # Test suite
+```
+
+### Installation
+
+```bash
+# Basic installation
+pip install ins_pricing
+
+# Full installation (all optional dependencies)
+pip install ins_pricing[all]
+
+# Install specific extras
+pip install ins_pricing[bayesopt]    # Model training
+pip install ins_pricing[explain]     # Model explanation
+pip install ins_pricing[plotting]    # Visualization
+pip install ins_pricing[gnn]         # Graph neural networks
+```
+
+#### Multi-platform & GPU installation notes
+
+- **PyTorch (CPU/GPU/MPS)**: Install the correct PyTorch build for your platform/GPU first (CUDA on
+  Linux/Windows, ROCm on supported AMD platforms, or MPS on Apple Silicon). Then install the
+  optional extras you need (e.g., `bayesopt`, `explain`, or `gnn`). This avoids pip pulling a
+  mismatched wheel.
+- **Torch Geometric (GNN)**: `torch-geometric` often requires platform-specific wheels (e.g.,
+  `torch-scatter`, `torch-sparse`). Follow the official PyG installation instructions for your
+  CUDA/ROCm/CPU environment, then install `ins_pricing[gnn]`.
+- **Multi-GPU**: Training code will use CUDA when available and can enable multi-GPU via
+  `torch.distributed`/`DataParallel` where supported. On Windows, CUDA DDP is not supported and will
+  fall back to single-GPU or DataParallel where possible.
+
+### Requirements
+
+- Python >= 3.9
+- Core dependencies: numpy >= 1.20, pandas >= 1.4
+
+### License
+
+Proprietary

{ins_pricing-0.3.3 → ins_pricing-0.3.4}/ins_pricing/docs/modelling/BayesOpt_USAGE.md

@@ -75,13 +75,13 @@ Under `ins_pricing/modelling/core/bayesopt/`:
 
 1) **Tools and utilities**
 
-- `IOUtils / TrainingUtils / PlotUtils`: I/O, training utilities (batch size, tweedie loss, free_cuda), plotting helpers
+- `IOUtils / TrainingUtils / PlotUtils`: I/O, training utilities (batch size, loss functions, free_cuda), plotting helpers
 - `DistributedUtils`: DDP init, rank/world_size helpers
 
 2) **TorchTrainerMixin (common components for torch tabular training)**
 
 - DataLoader: `_build_dataloader()` / `_build_val_dataloader()` (prints batch/accum/workers)
-- Loss: `_compute_losses()` / `_compute_weighted_loss()` (regression uses tweedie; classification uses BCEWithLogits)
+- Loss: `_compute_losses()` / `_compute_weighted_loss()` (regression supports tweedie/poisson/gamma/mse/mae; classification uses BCEWithLogits)
 - Early stop: `_early_stop_update()`
 
 3) **Sklearn-style model classes (core training objects)**
@@ -292,7 +292,7 @@ FT role is controlled by `ft_role` (from config or CLI `--ft-role`):
 ### 4.1 Supervised models (GLM/XGB/ResNet/FT-as-model)
 
 - `TrainerBase.tune()` calls each trainer's `cross_val()` and minimizes validation metric (default direction `minimize`)
-- Regression typically uses Tweedie deviance or related loss; classification uses logloss
+- Regression loss is configurable (tweedie/poisson/gamma/mse/mae); classification uses logloss
 
 ### 4.2 FT self-supervised (`unsupervised_embedding`)
 

{ins_pricing-0.3.3 → ins_pricing-0.3.4}/ins_pricing/modelling/core/bayesopt/config_preprocess.py

@@ -12,6 +12,7 @@ import pandas as pd
 from sklearn.preprocessing import StandardScaler
 
 from .utils import IOUtils
+from .utils.losses import normalize_loss_name
 from ....exceptions import ConfigurationError, DataValidationError
 
 # NOTE: Some CSV exports may contain invisible BOM characters or leading/trailing
@@ -81,6 +82,7 @@ class BayesOptConfig:
         task_type: Either 'regression' or 'classification'
         binary_resp_nme: Column name for binary response (optional)
         cate_list: List of categorical feature column names
+        loss_name: Regression loss ('auto', 'tweedie', 'poisson', 'gamma', 'mse', 'mae')
         prop_test: Proportion of data for validation (0.0-1.0)
         rand_seed: Random seed for reproducibility
         epochs: Number of training epochs
@@ -117,6 +119,7 @@ class BayesOptConfig:
     task_type: str = 'regression'
     binary_resp_nme: Optional[str] = None
     cate_list: Optional[List[str]] = None
+    loss_name: str = "auto"
 
     # Training configuration
     prop_test: float = 0.25
@@ -207,6 +210,15 @@ class BayesOptConfig:
             errors.append(
                 f"task_type must be one of {valid_task_types}, got '{self.task_type}'"
             )
+        # Validate loss_name
+        try:
+            normalized_loss = normalize_loss_name(self.loss_name, self.task_type)
+            if self.task_type == "classification" and normalized_loss not in {"auto", "logloss", "bce"}:
+                errors.append(
+                    "loss_name must be 'auto', 'logloss', or 'bce' for classification tasks."
+                )
+        except ValueError as exc:
+            errors.append(str(exc))
 
         # Validate prop_test
         if not 0.0 < self.prop_test < 1.0:

{ins_pricing-0.3.3 → ins_pricing-0.3.4}/ins_pricing/modelling/core/bayesopt/core.py

@@ -17,6 +17,12 @@ from .model_plotting_mixin import BayesOptPlottingMixin
 from .models import GraphNeuralNetSklearn
 from .trainers import FTTrainer, GLMTrainer, GNNTrainer, ResNetTrainer, XGBTrainer
 from .utils import EPS, infer_factor_and_cate_list, set_global_seed
+from .utils.losses import (
+    infer_loss_name_from_model_name,
+    normalize_loss_name,
+    resolve_tweedie_power,
+    resolve_xgb_objective,
+)
 
 
 class _CVSplitter:
@@ -293,6 +299,14 @@ class BayesOptModel(BayesOptPlottingMixin, BayesOptExplainMixin):
         self.config = cfg
         self.model_nme = cfg.model_nme
         self.task_type = cfg.task_type
+        normalized_loss = normalize_loss_name(getattr(cfg, "loss_name", None), self.task_type)
+        if self.task_type == "classification":
+            self.loss_name = "logloss" if normalized_loss == "auto" else normalized_loss
+        else:
+            if normalized_loss == "auto":
+                self.loss_name = infer_loss_name_from_model_name(self.model_nme)
+            else:
+                self.loss_name = normalized_loss
         self.resp_nme = cfg.resp_nme
         self.weight_nme = cfg.weight_nme
         self.factor_nmes = cfg.factor_nmes
@@ -339,14 +353,7 @@ class BayesOptModel(BayesOptPlottingMixin, BayesOptExplainMixin):
         if self.task_type == 'classification':
             self.obj = 'binary:logistic'
         else:  # regression task
-            if 'f' in self.model_nme:
-                self.obj = 'count:poisson'
-            elif 's' in self.model_nme:
-                self.obj = 'reg:gamma'
-            elif 'bc' in self.model_nme:
-                self.obj = 'reg:tweedie'
-            else:
-                self.obj = 'reg:tweedie'
+            self.obj = resolve_xgb_objective(self.loss_name)
         self.fit_params = {
             'sample_weight': self.train_data[self.weight_nme].values
         }
@@ -426,6 +433,11 @@ class BayesOptModel(BayesOptPlottingMixin, BayesOptExplainMixin):
     def default_tweedie_power(self, obj: Optional[str] = None) -> Optional[float]:
         if self.task_type == 'classification':
             return None
+        loss_name = getattr(self, "loss_name", None)
+        if loss_name:
+            resolved = resolve_tweedie_power(str(loss_name), default=1.5)
+            if resolved is not None:
+                return resolved
         objective = obj or getattr(self, "obj", None)
         if objective == 'count:poisson':
             return 1.0
@@ -503,6 +515,7 @@ class BayesOptModel(BayesOptPlottingMixin, BayesOptExplainMixin):
                 patience=5,
                 task_type=self.task_type,
                 tweedie_power=tw_power,
+                loss_name=self.loss_name,
                 use_data_parallel=False,
                 use_ddp=False,
                 use_approx_knn=self.config.gnn_use_approx_knn,

{ins_pricing-0.3.3 → ins_pricing-0.3.4}/ins_pricing/modelling/core/bayesopt/models/model_ft_trainer.py

@@ -16,6 +16,11 @@ from torch.nn.parallel import DistributedDataParallel as DDP
 from torch.nn.utils import clip_grad_norm_
 
 from ..utils import DistributedUtils, EPS, TorchTrainerMixin
+from ..utils.losses import (
+    infer_loss_name_from_model_name,
+    normalize_loss_name,
+    resolve_tweedie_power,
+)
 from .model_ft_components import FTTransformerCore, MaskedTabularDataset, TabularDataset
 
 
@@ -159,7 +164,8 @@ class FTTransformerSklearn(TorchTrainerMixin, nn.Module):
                  weight_decay: float = 0.0,
                  use_data_parallel: bool = True,
                  use_ddp: bool = False,
-                 num_numeric_tokens: Optional[int] = None
+                 num_numeric_tokens: Optional[int] = None,
+                 loss_name: Optional[str] = None
                  ):
         super().__init__()
 
@@ -187,14 +193,18 @@ class FTTransformerSklearn(TorchTrainerMixin, nn.Module):
         self.weight_decay = weight_decay
         self.task_type = task_type
         self.patience = patience
+        resolved_loss = normalize_loss_name(loss_name, self.task_type)
         if self.task_type == 'classification':
+            self.loss_name = "logloss"
             self.tw_power = None  # No Tweedie power for classification.
-        elif 'f' in self.model_nme:
-            self.tw_power = 1.0
-        elif 's' in self.model_nme:
-            self.tw_power = 2.0
         else:
-            self.tw_power = tweedie_power
+            if resolved_loss == "auto":
+                resolved_loss = infer_loss_name_from_model_name(self.model_nme)
+            self.loss_name = resolved_loss
+            if self.loss_name == "tweedie":
+                self.tw_power = float(tweedie_power) if tweedie_power is not None else 1.5
+            else:
+                self.tw_power = resolve_tweedie_power(self.loss_name, default=1.5)
 
         if self.is_ddp_enabled:
             self.device = torch.device(f"cuda:{self.local_rank}")

{ins_pricing-0.3.3 → ins_pricing-0.3.4}/ins_pricing/modelling/core/bayesopt/models/model_gnn.py

@@ -17,6 +17,11 @@ from torch.nn.parallel import DistributedDataParallel as DDP
 from torch.nn.utils import clip_grad_norm_
 
 from ..utils import DistributedUtils, EPS, IOUtils, TorchTrainerMixin
+from ..utils.losses import (
+    infer_loss_name_from_model_name,
+    normalize_loss_name,
+    resolve_tweedie_power,
+)
 
 try:
     from torch_geometric.nn import knn_graph
@@ -109,7 +114,8 @@ class GraphNeuralNetSklearn(TorchTrainerMixin, nn.Module):
                  max_gpu_knn_nodes: Optional[int] = None,
                  knn_gpu_mem_ratio: float = 0.9,
                  knn_gpu_mem_overhead: float = 2.0,
-                 knn_cpu_jobs: Optional[int] = -1) -> None:
+                 knn_cpu_jobs: Optional[int] = -1,
+                 loss_name: Optional[str] = None) -> None:
         super().__init__()
         self.model_nme = model_nme
         self.input_dim = input_dim
@@ -139,14 +145,18 @@ class GraphNeuralNetSklearn(TorchTrainerMixin, nn.Module):
         self._adj_cache_key: Optional[Tuple[Any, ...]] = None
         self._adj_cache_tensor: Optional[torch.Tensor] = None
 
+        resolved_loss = normalize_loss_name(loss_name, self.task_type)
         if self.task_type == 'classification':
+            self.loss_name = "logloss"
             self.tw_power = None
-        elif 'f' in self.model_nme:
-            self.tw_power = 1.0
-        elif 's' in self.model_nme:
-            self.tw_power = 2.0
         else:
-            self.tw_power = tweedie_power
+            if resolved_loss == "auto":
+                resolved_loss = infer_loss_name_from_model_name(self.model_nme)
+            self.loss_name = resolved_loss
+            if self.loss_name == "tweedie":
+                self.tw_power = float(tweedie_power) if tweedie_power is not None else 1.5
+            else:
+                self.tw_power = resolve_tweedie_power(self.loss_name, default=1.5)
 
         self.ddp_enabled = False
         self.local_rank = int(os.environ.get("LOCAL_RANK", 0))

{ins_pricing-0.3.3 → ins_pricing-0.3.4}/ins_pricing/modelling/core/bayesopt/models/model_resn.py

@@ -12,6 +12,11 @@ from torch.nn.utils import clip_grad_norm_
 from torch.utils.data import TensorDataset
 
 from ..utils import DistributedUtils, EPS, TorchTrainerMixin
+from ..utils.losses import (
+    infer_loss_name_from_model_name,
+    normalize_loss_name,
+    resolve_tweedie_power,
+)
 
 
 # =============================================================================
@@ -140,7 +145,8 @@ class ResNetSklearn(TorchTrainerMixin, nn.Module):
                  stochastic_depth: float = 0.0,
                  weight_decay: float = 1e-4,
                  use_data_parallel: bool = True,
-                 use_ddp: bool = False):
+                 use_ddp: bool = False,
+                 loss_name: Optional[str] = None):
         super(ResNetSklearn, self).__init__()
 
         self.use_ddp = use_ddp
@@ -179,15 +185,18 @@ class ResNetSklearn(TorchTrainerMixin, nn.Module):
         else:
             self.device = torch.device('cpu')
 
-        # Tweedie power (unused for classification)
+        resolved_loss = normalize_loss_name(loss_name, self.task_type)
         if self.task_type == 'classification':
+            self.loss_name = "logloss"
             self.tw_power = None
-        elif 'f' in self.model_nme:
-            self.tw_power = 1
-        elif 's' in self.model_nme:
-            self.tw_power = 2
         else:
-            self.tw_power = tweedie_power
+            if resolved_loss == "auto":
+                resolved_loss = infer_loss_name_from_model_name(self.model_nme)
+            self.loss_name = resolved_loss
+            if self.loss_name == "tweedie":
+                self.tw_power = float(tweedie_power) if tweedie_power is not None else 1.5
+            else:
+                self.tw_power = resolve_tweedie_power(self.loss_name, default=1.5)
 
         # Build network (construct on CPU first)
         core = ResNetSequential(

{ins_pricing-0.3.3 → ins_pricing-0.3.4}/ins_pricing/modelling/core/bayesopt/trainers/trainer_base.py

@@ -578,6 +578,7 @@ class TrainerBase:
                     "n_layers": getattr(self.model, "n_layers", 4),
                     "dropout": getattr(self.model, "dropout", 0.1),
                     "task_type": getattr(self.model, "task_type", "regression"),
+                    "loss_name": getattr(self.model, "loss_name", None),
                     "tw_power": getattr(self.model, "tw_power", 1.5),
                     "num_geo": getattr(self.model, "num_geo", 0),
                     "num_numeric_tokens": getattr(self.model, "num_numeric_tokens", None),
@@ -638,6 +639,7 @@ class TrainerBase:
                             n_layers=model_config.get("n_layers", 4),
                             dropout=model_config.get("dropout", 0.1),
                             task_type=model_config.get("task_type", "regression"),
+                            loss_name=model_config.get("loss_name", None),
                             tweedie_power=model_config.get("tw_power", 1.5),
                             num_numeric_tokens=model_config.get("num_numeric_tokens"),
                             use_data_parallel=False,