dragon-ml-toolbox 14.7.0__py3-none-any.whl → 16.2.0__py3-none-any.whl

ml_tools/ML_datasetmaster.py

@@ -3,27 +3,25 @@ from torch.utils.data import Dataset
 import pandas
 import numpy
 from sklearn.model_selection import train_test_split
-from typing import Literal, Union, Tuple, List, Optional
-from abc import ABC, abstractmethod
-import matplotlib.pyplot as plt
+from typing import Literal, Union, List, Optional
+from abc import ABC
 from pathlib import Path
 
 from .path_manager import make_fullpath, sanitize_filename
 from ._logger import _LOGGER
 from ._script_info import _script_info
 from .custom_logger import save_list_strings
-from .ML_scaler import PytorchScaler
-from .keys import DatasetKeys
+from .ML_scaler import DragonScaler
+from ._keys import DatasetKeys, MLTaskKeys
 from ._schema import FeatureSchema
+from .custom_logger import custom_logger
 
 
 __all__ = [
-    "DatasetMaker",
-    "DatasetMakerMulti",
-    "SequenceMaker"
+    "DragonDataset",
+    "DragonDatasetMulti"
 ]
 
-
 # --- Internal Helper Class ---
 class _PytorchDataset(Dataset):
     """
@@ -57,6 +55,8 @@ class _PytorchDataset(Dataset):
             
         self._feature_names = feature_names
         self._target_names = target_names
+        self._classes: List[str] = []
+        self._class_map: dict[str,int] = dict()
 
     def __len__(self):
         return len(self.features)
@@ -78,6 +78,15 @@ class _PytorchDataset(Dataset):
             return self._target_names
         else:
             _LOGGER.error(f"Dataset {self.__class__} has not been initialized with any target names.")
+            raise ValueError()
+
+    @property
+    def classes(self):
+        return self._classes
+    
+    @property
+    def class_map(self):
+        return self._class_map
 
 
 # --- Abstract Base Class ---
@@ -88,23 +97,29 @@ class _BaseDatasetMaker(ABC):
     """
     def __init__(self):
         self._train_ds: Optional[Dataset] = None
+        self._val_ds: Optional[Dataset] = None
         self._test_ds: Optional[Dataset] = None
-        self.scaler: Optional[PytorchScaler] = None
+        self.scaler: Optional[DragonScaler] = None
         self._id: Optional[str] = None
         self._feature_names: List[str] = []
         self._target_names: List[str] = []
         self._X_train_shape = (0,0)
+        self._X_val_shape = (0,0)
         self._X_test_shape = (0,0)
         self._y_train_shape = (0,)
+        self._y_val_shape = (0,)
         self._y_test_shape = (0,)
+        self.class_map: dict[str, int] = dict()
+        self.classes: list[str] = list()
         
     def _prepare_scaler(self, 
                         X_train: pandas.DataFrame, 
                         y_train: Union[pandas.Series, pandas.DataFrame], 
+                        X_val: pandas.DataFrame,
                         X_test: pandas.DataFrame, 
                         label_dtype: torch.dtype, 
                         schema: FeatureSchema):
-        """Internal helper to fit and apply a PytorchScaler using a FeatureSchema."""
+        """Internal helper to fit and apply a DragonScaler using a FeatureSchema."""
         continuous_feature_indices: Optional[List[int]] = None
 
         # Get continuous feature indices *from the schema*
@@ -122,26 +137,33 @@ class _BaseDatasetMaker(ABC):
             _LOGGER.info("No continuous features listed in schema. Scaler will not be fitted.")
 
         X_train_values = X_train.to_numpy()
+        X_val_values = X_val.to_numpy()
         X_test_values = X_test.to_numpy()
 
         # continuous_feature_indices is derived
         if self.scaler is None and continuous_feature_indices:
-            _LOGGER.info("Fitting a new PytorchScaler on training data.")
+            _LOGGER.info("Fitting a new DragonScaler on training data.")
             temp_train_ds = _PytorchDataset(X_train_values, y_train, label_dtype) # type: ignore
-            self.scaler = PytorchScaler.fit(temp_train_ds, continuous_feature_indices)
+            self.scaler = DragonScaler.fit(temp_train_ds, continuous_feature_indices)
 
         if self.scaler and self.scaler.mean_ is not None:
-            _LOGGER.info("Applying scaler transformation to train and test feature sets.")
+            _LOGGER.info("Applying scaler transformation to train, validation, and test feature sets.")
             X_train_tensor = self.scaler.transform(torch.tensor(X_train_values, dtype=torch.float32))
+            X_val_tensor = self.scaler.transform(torch.tensor(X_val_values, dtype=torch.float32))
             X_test_tensor = self.scaler.transform(torch.tensor(X_test_values, dtype=torch.float32))
-            return X_train_tensor.numpy(), X_test_tensor.numpy()
+            return X_train_tensor.numpy(), X_val_tensor.numpy(), X_test_tensor.numpy()
 
-        return X_train_values, X_test_values
+        return X_train_values, X_val_values, X_test_values
 
     @property
     def train_dataset(self) -> Dataset:
         if self._train_ds is None: raise RuntimeError("Dataset not yet created.")
         return self._train_ds
+    
+    @property
+    def validation_dataset(self) -> Dataset:
+        if self._val_ds is None: raise RuntimeError("Dataset not yet created.")
+        return self._val_ds
 
     @property
     def test_dataset(self) -> Dataset:
@@ -176,6 +198,7 @@ class _BaseDatasetMaker(ABC):
     def dataframes_info(self) -> None:
         print("--- DataFrame Shapes After Split ---")
         print(f"  X_train shape: {self._X_train_shape}, y_train shape: {self._y_train_shape}")
+        print(f"  X_val shape:   {self._X_val_shape}, y_val shape:   {self._y_val_shape}")
         print(f"  X_test shape:  {self._X_test_shape}, y_test shape:  {self._y_test_shape}")
         print("------------------------------------")
     
@@ -195,7 +218,7 @@ class _BaseDatasetMaker(ABC):
 
     def save_scaler(self, directory: Union[str, Path], verbose: bool=True) -> None:
         """
-        Saves the fitted PytorchScaler's state to a .pth file.
+        Saves the fitted DragonScaler's state to a .pth file.
 
         The filename is automatically generated based on the dataset id.
         
@@ -215,6 +238,24 @@ class _BaseDatasetMaker(ABC):
         self.scaler.save(filepath, verbose=False)
         if verbose:
             _LOGGER.info(f"Scaler for dataset '{self.id}' saved as '{filepath.name}'.")
+            
+    def save_class_map(self, directory: Union[str,Path], verbose: bool=True) -> None:
+        """
+        Saves the class to index mapping {str: int} to a directory.
+        """
+        if not self.class_map:
+            _LOGGER.warning(f"No class_map defined. Skipping.")
+            return
+        
+        log_name = f"Class_to_Index_{self.id}" if self.id else "Class_to_Index"
+        
+        custom_logger(data=self.class_map,
+                      save_directory=directory,
+                      log_name=log_name,
+                      add_timestamp=False,
+                      dict_as="json")
+        if verbose:
+            _LOGGER.info(f"Class map for '{self.id}' saved as '{log_name}.json'.")
 
     def save_artifacts(self, directory: Union[str, Path], verbose: bool=True) -> None:
         """
@@ -224,19 +265,22 @@ class _BaseDatasetMaker(ABC):
         self.save_target_names(directory=directory, verbose=verbose)
         if self.scaler is not None:
             self.save_scaler(directory=directory, verbose=verbose)
+        if self.class_map:
+            self.save_class_map(directory=directory, verbose=verbose)
 
 
 # Single target dataset
-class DatasetMaker(_BaseDatasetMaker):
+class DragonDataset(_BaseDatasetMaker):
     """
     Dataset maker for pre-processed, numerical pandas DataFrames with a single target column.
 
     This class takes a DataFrame, and a FeatureSchema, automatically splits and converts them into PyTorch Datasets.
-    It can also create and apply a PytorchScaler using the schema.
+    It can also create and apply a DragonScaler using the schema.
     
     Attributes:
-        `scaler` -> PytorchScaler | None
+        `scaler` -> DragonScaler | None
         `train_dataset` -> PyTorch Dataset
+        `validation_dataset` -> PyTorch Dataset
         `test_dataset`  -> PyTorch Dataset
         `feature_names` -> list[str]
         `target_names`  -> list[str]
@@ -247,9 +291,11 @@ class DatasetMaker(_BaseDatasetMaker):
     def __init__(self,
                  pandas_df: pandas.DataFrame,
                  schema: FeatureSchema,
-                 kind: Literal["regression", "classification"],
-                 scaler: Union[Literal["fit"], Literal["none"], PytorchScaler],
-                 test_size: float = 0.2,
+                 kind: Literal["regression", "binary classification", "multiclass classification"],
+                 scaler: Union[Literal["fit"], Literal["none"], DragonScaler],
+                 validation_size: float = 0.2,
+                 test_size: float = 0.1,
+                 class_map: Optional[dict[str,int]]=None,
                  random_state: int = 42):
         """
         Args:
@@ -257,32 +303,46 @@ class DatasetMaker(_BaseDatasetMaker):
                 The pre-processed input DataFrame containing all columns. (features and single target).
             schema (FeatureSchema): 
                 The definitive schema object from data_exploration.
-            kind ("regression" | "classification"): 
-                The type of ML task. This determines the data type of the labels.
-            scaler ("fit" | "none" | PytorchScaler): 
+            kind (str): 
+                The type of ML task. Must be one of:
+                - "regression"
+                - "binary classification"
+                - "multiclass classification"
+            scaler ("fit" | "none" | DragonScaler): 
                 Strategy for data scaling:
-                - "fit": Fit a new PytorchScaler on continuous features.
+                - "fit": Fit a new DragonScaler on continuous features.
                 - "none": Do not scale data (e.g., for TabularTransformer).
-                - PytorchScaler instance: Use a pre-fitted scaler to transform data.
+                - DragonScaler instance: Use a pre-fitted scaler to transform data.
+            validation_size (float):
+                The proportion of the *original* dataset to allocate to the validation split.
             test_size (float): 
-                The proportion of the dataset to allocate to the test split.
+                The proportion of the dataset to allocate to the test split (can be 0).
+            class_map (dict[str,int] | None): Optional class map for the target classes in classification tasks. Can be set later using `.set_class_map()`.
             random_state (int): 
                 The seed for the random number of generator for reproducibility.
             
         """
         super().__init__()
         
+        # --- Validation for split sizes ---
+        if (validation_size + test_size) >= 1.0:
+            _LOGGER.error(f"The sum of validation_size ({validation_size}) and test_size ({test_size}) must be less than 1.0.")
+            raise ValueError()
+        elif validation_size <= 0.0:
+            _LOGGER.error(f"Invalid validation split of {validation_size}.")
+            raise ValueError()
+        
         _apply_scaling: bool = False
         if scaler == "fit":
             self.scaler = None # To be created
             _apply_scaling = True
         elif scaler == "none":
             self.scaler = None
-        elif isinstance(scaler, PytorchScaler):
+        elif isinstance(scaler, DragonScaler):
             self.scaler = scaler # Use the provided one
             _apply_scaling = True
         else:
-            _LOGGER.error(f"Invalid 'scaler' argument. Must be 'fit', 'none', or a PytorchScaler instance.")
+            _LOGGER.error(f"Invalid 'scaler' argument. Must be 'fit', 'none', or a DragonScaler instance.")
             raise ValueError()
         
         # --- 1. Identify features (from schema) ---
@@ -298,7 +358,7 @@ class DatasetMaker(_BaseDatasetMaker):
             _LOGGER.error("No target column found. The schema's features match the DataFrame's columns exactly.")
             raise ValueError("No target column found in DataFrame.")
         if len(target_cols_set) > 1:
-            _LOGGER.error(f"Ambiguous target. Found {len(target_cols_set)} columns not in the schema: {list(target_cols_set)}. DatasetMaker (single-target) requires exactly one.")
+            _LOGGER.error(f"Ambiguous target. Found {len(target_cols_set)} columns not in the schema: {list(target_cols_set)}. One target required.")
             raise ValueError("Ambiguous target: More than one non-feature column found.")
             
         target_name = list(target_cols_set)[0]
@@ -308,32 +368,112 @@ class DatasetMaker(_BaseDatasetMaker):
         # --- 3. Split Data ---
         features_df = pandas_df[self._feature_names]
         target_series = pandas_df[target_name]
-
-        X_train, X_test, y_train, y_test = train_test_split(
+        
+        # First split: (Train + Val) vs TesT
+        X_train_val, X_test, y_train_val, y_test = train_test_split(
             features_df, 
             target_series, 
             test_size=test_size, 
             random_state=random_state
         )
-        self._X_train_shape, self._X_test_shape = X_train.shape, X_test.shape
-        self._y_train_shape, self._y_test_shape = y_train.shape, y_test.shape
+        # Calculate validation split size relative to the (Train + Val) set
+        val_split_size = validation_size / (1.0 - test_size)
+        
+        # Second split: Train vs Val
+        X_train, X_val, y_train, y_val = train_test_split(
+            X_train_val, 
+            y_train_val, 
+            test_size=val_split_size, 
+            random_state=random_state
+        )
+        
+        self._X_train_shape, self._X_val_shape, self._X_test_shape = X_train.shape, X_val.shape, X_test.shape
+        self._y_train_shape, self._y_val_shape, self._y_test_shape = y_train.shape, y_val.shape, y_test.shape
         
-        label_dtype = torch.float32 if kind == "regression" else torch.int64
+        # --- label_dtype logic ---
+        if kind == MLTaskKeys.REGRESSION or kind == MLTaskKeys.BINARY_CLASSIFICATION:
+            label_dtype = torch.float32
+        elif kind == MLTaskKeys.MULTICLASS_CLASSIFICATION:
+            label_dtype = torch.int64
+        else:
+            _LOGGER.error(f"Invalid 'kind' {kind}. Must be '{MLTaskKeys.REGRESSION}', '{MLTaskKeys.BINARY_CLASSIFICATION}', or '{MLTaskKeys.MULTICLASS_CLASSIFICATION}'.")
+            raise ValueError()
+        self.kind = kind
 
         # --- 4. Scale (using the schema) ---
         if _apply_scaling:
-            X_train_final, X_test_final = self._prepare_scaler(
-                X_train, y_train, X_test, label_dtype, schema
+            X_train_final, X_val_final, X_test_final = self._prepare_scaler(
+                X_train, y_train, X_val, X_test, label_dtype, schema
             )
         else:
             _LOGGER.info("Features have not been scaled as specified.")
             X_train_final = X_train.to_numpy()
+            X_val_final = X_val.to_numpy()
             X_test_final = X_test.to_numpy()
         
         # --- 5. Create Datasets ---
         self._train_ds = _PytorchDataset(X_train_final, y_train, labels_dtype=label_dtype, feature_names=self._feature_names, target_names=self._target_names)
+        self._val_ds = _PytorchDataset(X_val_final, y_val, labels_dtype=label_dtype, feature_names=self._feature_names, target_names=self._target_names)
         self._test_ds = _PytorchDataset(X_test_final, y_test, labels_dtype=label_dtype, feature_names=self._feature_names, target_names=self._target_names)
-    
+        
+        # --- 6. create class map if given ---
+        if self.kind != MLTaskKeys.REGRESSION:
+            if class_map is None:
+                self.class_map = dict()
+            else:
+                self.set_class_map(class_map)
+        else:
+            self.class_map = dict()
+
+    def set_class_map(self, class_map: dict[str, int], force_overwrite: bool=False) -> None:
+        """
+        Sets a map of class_name -> integer_label.
+        
+        This is used by the InferenceHandler and to finalize the model after training.
+
+        Args:
+            class_map (Dict[str, int]): A dictionary mapping the integer label
+                to its string name.
+                Example: {'cat': 0, 'dog': 1, 'bird': 2}
+            force_overwrite (bool): Required to overwrite a previously set class map.
+        """
+        if self.kind == MLTaskKeys.REGRESSION:
+            _LOGGER.warning(f"Class Map is for classifications tasks only.")
+            return
+        
+        if self.class_map:
+            warning_message = f"Class map was previously set."
+            if not force_overwrite:
+                warning_message += " Use `force_overwrite=True` to set new values."
+                _LOGGER.warning(warning_message)
+                return
+            else:
+                warning_message += ". Setting new values..."
+                _LOGGER.warning(warning_message)
+        
+        self.class_map = class_map
+        
+        try:
+            sorted_items = sorted(class_map.items(), key=lambda item: item[1])
+            class_list = [item[0] for item in sorted_items]
+        except Exception as e:
+            _LOGGER.error(f"Could not sort class map. Ensure it is a dict of {str: int}. Error: {e}")
+            raise TypeError()
+        else:
+            self.classes = class_list
+        
+        if self._train_ds:
+            self._train_ds._classes = class_list # type: ignore
+            self._train_ds._class_map = class_map # type: ignore
+        if self._val_ds:
+            self._val_ds._classes = class_list # type: ignore
+            self._val_ds._class_map = class_map # type: ignore
+        if self._test_ds:
+            self._test_ds._classes = class_list # type: ignore
+            self._test_ds._class_map = class_map # type: ignore
+            
+        _LOGGER.info(f"Class map set for dataset '{self.id}' and its subsets:\n{class_map}")
+
     def __repr__(self) -> str:
         s = f"<{self.__class__.__name__} (ID: '{self.id}')>\n"
         s += f"  Target: {self.target_names[0]}\n"
@@ -342,6 +482,8 @@ class DatasetMaker(_BaseDatasetMaker):
         
         if self._train_ds:
             s += f"  Train Samples: {len(self._train_ds)}\n" # type: ignore
+        if self._val_ds:
+            s += f"  Validation Samples: {len(self._val_ds)}\n" # type: ignore
         if self._test_ds:
             s += f"  Test Samples: {len(self._test_ds)}\n" # type: ignore
             
@@ -349,7 +491,7 @@ class DatasetMaker(_BaseDatasetMaker):
 
 
 # --- Multi-Target Class ---
-class DatasetMakerMulti(_BaseDatasetMaker):
+class DragonDatasetMulti(_BaseDatasetMaker):
     """
     Dataset maker for pre-processed, numerical pandas DataFrames with 
     multiple target columns.
@@ -358,15 +500,15 @@ class DatasetMakerMulti(_BaseDatasetMaker):
     *target_columns*. It validates that the schema's features and the
     target columns are mutually exclusive and together account for all
     columns in the DataFrame.
-    
-    Targets dtype is torch.float32
     """
     def __init__(self,
                  pandas_df: pandas.DataFrame,
                  target_columns: List[str],
                  schema: FeatureSchema,
-                 scaler: Union[Literal["fit"], Literal["none"], PytorchScaler],
-                 test_size: float = 0.2,
+                 kind: Literal["multitarget regression", "multilabel binary classification"],
+                 scaler: Union[Literal["fit"], Literal["none"], DragonScaler],
+                 validation_size: float = 0.2,
+                 test_size: float = 0.1,
                  random_state: int = 42):
         """
         Args:
@@ -377,11 +519,17 @@ class DatasetMakerMulti(_BaseDatasetMaker):
                 List of target column names.
             schema (FeatureSchema): 
                 The definitive schema object from data_exploration.
-            scaler ("fit" | "none" | PytorchScaler): 
+            kind (str):
+                The type of multi-target ML task. Must be one of:
+                - "multitarget regression"
+                - "multilabel binary classification"
+            scaler ("fit" | "none" | DragonScaler): 
                 Strategy for data scaling:
-                - "fit": Fit a new PytorchScaler on continuous features.
+                - "fit": Fit a new DragonScaler on continuous features.
                 - "none": Do not scale data (e.g., for TabularTransformer).
-                - PytorchScaler instance: Use a pre-fitted scaler to transform data.
+                - DragonScaler instance: Use a pre-fitted scaler to transform data.
+            validation_size (float):
+                The proportion of the dataset to allocate to the validation split.
             test_size (float): 
                 The proportion of the dataset to allocate to the test split.
             random_state (int): 
@@ -389,21 +537,34 @@ class DatasetMakerMulti(_BaseDatasetMaker):
                 
         ## Note:
         For multi-binary classification, the most common PyTorch loss function is nn.BCEWithLogitsLoss. 
-        This loss function requires the labels to be torch.float32 which is the same type required for regression (multi-regression) tasks.
+        This loss function requires the labels to be torch.float32 which is the same type required for multi-regression tasks.
         """
         super().__init__()
         
+        # --- Validation for split sizes ---
+        if (validation_size + test_size) >= 1.0:
+            _LOGGER.error(f"The sum of validation_size ({validation_size}) and test_size ({test_size}) must be less than 1.0.")
+            raise ValueError("validation_size and test_size sum must be < 1.0")
+        elif validation_size <= 0.0:
+            _LOGGER.error(f"Invalid validation split of {validation_size}.")
+            raise ValueError()
+            
+        # --- Validate kind parameter ---
+        if kind not in [MLTaskKeys.MULTITARGET_REGRESSION, MLTaskKeys.MULTILABEL_BINARY_CLASSIFICATION]:
+            _LOGGER.error(f"Invalid 'kind' {kind}. Must be '{MLTaskKeys.MULTITARGET_REGRESSION}' or '{MLTaskKeys.MULTILABEL_BINARY_CLASSIFICATION}'.")
+            raise ValueError()
+        
         _apply_scaling: bool = False
         if scaler == "fit":
             self.scaler = None
             _apply_scaling = True
         elif scaler == "none":
             self.scaler = None
-        elif isinstance(scaler, PytorchScaler):
+        elif isinstance(scaler, DragonScaler):
             self.scaler = scaler # Use the provided one
             _apply_scaling = True
         else:
-            _LOGGER.error(f"Invalid 'scaler' argument. Must be 'fit', 'none', or a PytorchScaler instance.")
+            _LOGGER.error(f"Invalid 'scaler' argument. Must be 'fit', 'none', or a DragonScaler instance.")
             raise ValueError()
         
         # --- 1. Get features and targets from schema/args ---
@@ -433,32 +594,47 @@ class DatasetMakerMulti(_BaseDatasetMaker):
         # --- 3. Split Data ---
         features_df = pandas_df[self._feature_names]
         target_df = pandas_df[self._target_names]
-
-        X_train, X_test, y_train, y_test = train_test_split(
+        
+        # First split: (Train + Val) vs Test
+        X_train_val, X_test, y_train_val, y_test = train_test_split(
             features_df,
             target_df, 
             test_size=test_size, 
             random_state=random_state
         )
-        self._X_train_shape, self._X_test_shape = X_train.shape, X_test.shape
-        self._y_train_shape, self._y_test_shape = y_train.shape, y_test.shape
+
+        # Calculate validation split size relative to the (Train + Val) set
+        val_split_size = validation_size / (1.0 - test_size)
+            
+        # Second split: Train vs Val
+        X_train, X_val, y_train, y_val = train_test_split(
+            X_train_val, 
+            y_train_val, 
+            test_size=val_split_size, 
+            random_state=random_state
+        )
+
+        self._X_train_shape, self._X_val_shape, self._X_test_shape = X_train.shape, X_val.shape, X_test.shape
+        self._y_train_shape, self._y_val_shape, self._y_test_shape = y_train.shape, y_val.shape, y_test.shape
         
         # Multi-target for regression or multi-binary
         label_dtype = torch.float32 
 
         # --- 4. Scale (using the schema) ---
         if _apply_scaling:
-            X_train_final, X_test_final = self._prepare_scaler(
-                X_train, y_train, X_test, label_dtype, schema
+            X_train_final, X_val_final, X_test_final = self._prepare_scaler(
+                X_train, y_train, X_val, X_test, label_dtype, schema
             )
         else:
             _LOGGER.info("Features have not been scaled as specified.")
             X_train_final = X_train.to_numpy()
+            X_val_final = X_val.to_numpy()
             X_test_final = X_test.to_numpy()
         
         # --- 5. Create Datasets ---
         # _PytorchDataset now correctly handles y_train (a DataFrame)
         self._train_ds = _PytorchDataset(X_train_final, y_train, labels_dtype=label_dtype, feature_names=self._feature_names, target_names=self._target_names)
+        self._val_ds = _PytorchDataset(X_val_final, y_val, labels_dtype=label_dtype, feature_names=self._feature_names, target_names=self._target_names)
         self._test_ds = _PytorchDataset(X_test_final, y_test, labels_dtype=label_dtype, feature_names=self._feature_names, target_names=self._target_names)
 
     def __repr__(self) -> str:
@@ -469,234 +645,13 @@ class DatasetMakerMulti(_BaseDatasetMaker):
         
         if self._train_ds:
             s += f"  Train Samples: {len(self._train_ds)}\n" # type: ignore
+        if self._val_ds:
+            s += f"  Validation Samples: {len(self._val_ds)}\n" # type: ignore
         if self._test_ds:
             s += f"  Test Samples: {len(self._test_ds)}\n" # type: ignore
             
         return s
 
 
-# --- Private Base Class ---
-class _BaseMaker(ABC):
-    """
-    Abstract Base Class for extra dataset makers.
-    """
-    def __init__(self):
-        self._train_dataset = None
-        self._test_dataset = None
-        self._val_dataset = None
-
-    @abstractmethod
-    def get_datasets(self) -> Tuple[Dataset, ...]:
-        """
-        The primary method to retrieve the final, processed PyTorch datasets.
-        Must be implemented by all subclasses.
-        """
-        pass
-
-
-# --- SequenceMaker ---
-class SequenceMaker(_BaseMaker):
-    """
-    Creates windowed PyTorch datasets from time-series data.
-    
-    Pipeline:
-    
-    1. `.split_data()`: Separate time series into training and testing portions.
-    2. `.normalize_data()`: Normalize the data. The scaler will be fitted on the training portion.
-    3. `.generate_windows()`: Create the windowed sequences from the split and normalized data.
-    4. `.get_datasets()`: Return Pytorch train and test datasets.
-    """
-    def __init__(self, data: Union[pandas.DataFrame, pandas.Series, numpy.ndarray], sequence_length: int):
-        super().__init__()
-        self.sequence_length = sequence_length
-        self.scaler = None
-        
-        if isinstance(data, pandas.DataFrame):
-            self.time_axis = data.index.values
-            self.sequence = data.iloc[:, 0].values.astype(numpy.float32)
-        elif isinstance(data, pandas.Series):
-            self.time_axis = data.index.values
-            self.sequence = data.values.astype(numpy.float32)
-        elif isinstance(data, numpy.ndarray):
-            self.time_axis = numpy.arange(len(data))
-            self.sequence = data.astype(numpy.float32)
-        else:
-            _LOGGER.error("Data must be a pandas DataFrame/Series or a numpy array.")
-            raise TypeError()
-            
-        self.train_sequence = None
-        self.test_sequence = None
-        
-        self._is_split = False
-        self._is_normalized = False
-        self._are_windows_generated = False
-
-    def normalize_data(self) -> 'SequenceMaker':
-        """
-        Normalizes the sequence data using PytorchScaler. Must be called AFTER 
-        splitting to prevent data leakage from the test set.
-        """
-        if not self._is_split:
-            _LOGGER.error("Data must be split BEFORE normalizing. Call .split_data() first.")
-            raise RuntimeError()
-
-        if self.scaler:
-            _LOGGER.warning("Data has already been normalized.")
-            return self
-
-        # 1. PytorchScaler requires a Dataset to fit. Create a temporary one.
-        # The scaler expects 2D data [n_samples, n_features].
-        train_features = self.train_sequence.reshape(-1, 1) # type: ignore
-
-        # _PytorchDataset needs labels, so we create dummy ones.
-        dummy_labels = numpy.zeros(len(train_features))
-        temp_train_ds = _PytorchDataset(train_features, dummy_labels, labels_dtype=torch.float32)
-
-        # 2. Fit the PytorchScaler on the temporary training dataset.
-        # The sequence is a single feature, so its index is [0].
-        _LOGGER.info("Fitting PytorchScaler on the training data...")
-        self.scaler = PytorchScaler.fit(temp_train_ds, continuous_feature_indices=[0])
-
-        # 3. Transform sequences using the fitted scaler.
-        # The transform method requires a tensor, so we convert, transform, and convert back.
-        train_tensor = torch.tensor(self.train_sequence.reshape(-1, 1), dtype=torch.float32) # type: ignore
-        test_tensor = torch.tensor(self.test_sequence.reshape(-1, 1), dtype=torch.float32) # type: ignore
-
-        self.train_sequence = self.scaler.transform(train_tensor).numpy().flatten()
-        self.test_sequence = self.scaler.transform(test_tensor).numpy().flatten()
-
-        self._is_normalized = True
-        _LOGGER.info("Sequence data normalized using PytorchScaler.")
-        return self
-
-    def split_data(self, test_size: float = 0.2) -> 'SequenceMaker':
-        """Splits the sequence into training and testing portions."""
-        if self._is_split:
-            _LOGGER.warning("Data has already been split.")
-            return self
-
-        split_idx = int(len(self.sequence) * (1 - test_size))
-        self.train_sequence = self.sequence[:split_idx]
-        self.test_sequence = self.sequence[split_idx - self.sequence_length:]
-        
-        self.train_time_axis = self.time_axis[:split_idx]
-        self.test_time_axis = self.time_axis[split_idx:]
-
-        self._is_split = True
-        _LOGGER.info(f"Sequence split into training ({len(self.train_sequence)} points) and testing ({len(self.test_sequence)} points).")
-        return self
-
-    def generate_windows(self, sequence_to_sequence: bool = False) -> 'SequenceMaker':
-        """
-        Generates overlapping windows for features and labels.
-        
-        "sequence-to-sequence": Label vectors are of the same size as the feature vectors instead of a single future prediction.
-        """
-        if not self._is_split:
-            _LOGGER.error("Cannot generate windows before splitting data. Call .split_data() first.")
-            raise RuntimeError()
-
-        self._train_dataset = self._create_windowed_dataset(self.train_sequence, sequence_to_sequence) # type: ignore
-        self._test_dataset = self._create_windowed_dataset(self.test_sequence, sequence_to_sequence) # type: ignore
-        
-        self._are_windows_generated = True
-        _LOGGER.info("Feature and label windows generated for train and test sets.")
-        return self
-
-    def _create_windowed_dataset(self, data: numpy.ndarray, use_sequence_labels: bool) -> Dataset:
-        """Efficiently creates windowed features and labels using numpy."""
-        if len(data) <= self.sequence_length:
-            _LOGGER.error("Data length must be greater than the sequence_length to create at least one window.")
-            raise ValueError()
-            
-        if not use_sequence_labels:
-            features = data[:-1]
-            labels = data[self.sequence_length:]
-            
-            n_windows = len(features) - self.sequence_length + 1
-            bytes_per_item = features.strides[0]
-            strided_features = numpy.lib.stride_tricks.as_strided(
-                features, shape=(n_windows, self.sequence_length), strides=(bytes_per_item, bytes_per_item)
-            )
-            return _PytorchDataset(strided_features, labels, labels_dtype=torch.float32)
-        
-        else:
-            x_data = data[:-1]
-            y_data = data[1:]
-            
-            n_windows = len(x_data) - self.sequence_length + 1
-            bytes_per_item = x_data.strides[0]
-            
-            strided_x = numpy.lib.stride_tricks.as_strided(x_data, shape=(n_windows, self.sequence_length), strides=(bytes_per_item, bytes_per_item))
-            strided_y = numpy.lib.stride_tricks.as_strided(y_data, shape=(n_windows, self.sequence_length), strides=(bytes_per_item, bytes_per_item))
-            
-            return _PytorchDataset(strided_x, strided_y, labels_dtype=torch.float32)
-
-    def denormalize(self, data: Union[torch.Tensor, numpy.ndarray]) -> numpy.ndarray:
-        """Applies inverse transformation using the stored PytorchScaler."""
-        if self.scaler is None:
-            _LOGGER.error("Data was not normalized. Cannot denormalize.")
-            raise RuntimeError()
-
-        # Ensure data is a torch.Tensor
-        if isinstance(data, numpy.ndarray):
-            tensor_data = torch.tensor(data, dtype=torch.float32)
-        else:
-            tensor_data = data
-
-        # Reshape for the scaler [n_samples, n_features]
-        if tensor_data.ndim == 1:
-            tensor_data = tensor_data.view(-1, 1)
-
-        # Apply inverse transform and convert back to a flat numpy array
-        original_scale_tensor = self.scaler.inverse_transform(tensor_data)
-        return original_scale_tensor.cpu().numpy().flatten()
-
-    def plot(self, predictions: Optional[numpy.ndarray] = None):
-        """Plots the original training and testing data, with optional predictions."""
-        if not self._is_split:
-            _LOGGER.error("Cannot plot before splitting data. Call .split_data() first.")
-            raise RuntimeError()
-        
-        plt.figure(figsize=(15, 6))
-        plt.title("Time Series Data")
-        plt.grid(True)
-        plt.xlabel("Time")
-        plt.ylabel("Value")
-        
-        plt.plot(self.train_time_axis, self.scaler.inverse_transform(self.train_sequence.reshape(-1, 1)), label='Train Data') # type: ignore
-        plt.plot(self.test_time_axis, self.scaler.inverse_transform(self.test_sequence[self.sequence_length-1:].reshape(-1, 1)), label='Test Data') # type: ignore
-
-        if predictions is not None:
-            pred_time_axis = self.test_time_axis[:len(predictions)]
-            plt.plot(pred_time_axis, predictions, label='Predictions', c='red')
-
-        plt.legend()
-        plt.show()
-
-    def get_datasets(self) -> Tuple[Dataset, Dataset]:
-        """Returns the final train and test datasets."""
-        if not self._are_windows_generated:
-            _LOGGER.error("Windows have not been generated. Call .generate_windows() first.")
-            raise RuntimeError()
-        return self._train_dataset, self._test_dataset
-    
-    def __repr__(self) -> str:
-        s = f"<{self.__class__.__name__}>:\n"
-        s += f"  Sequence Length (Window): {self.sequence_length}\n"
-        s += f"  Total Data Points: {len(self.sequence)}\n"
-        s += "  --- Status ---\n"
-        s += f"  Split: {self._is_split}\n"
-        s += f"  Normalized: {self._is_normalized}\n"
-        s += f"  Windows Generated: {self._are_windows_generated}\n"
-        
-        if self._are_windows_generated:
-            train_len = len(self._train_dataset) if self._train_dataset else 0 # type: ignore
-            test_len = len(self._test_dataset) if self._test_dataset else 0 # type: ignore
-            s += f"  Datasets (Train/Test): {train_len} / {test_len} windows\n"
-            
-        return s
-
-
 def info():
     _script_info(__all__)