dragon-ml-toolbox 14.3.1__py3-none-any.whl → 14.8.0__py3-none-any.whl

ml_tools/ML_evaluation_multi.py

@@ -34,6 +34,8 @@ __all__ = [
     "multi_target_shap_summary_plot",
 ]
 
+DPI_value = 250
+
 
 def multi_target_regression_metrics(
     y_true: np.ndarray,
@@ -90,7 +92,7 @@ def multi_target_regression_metrics(
 
         # --- Save Residual Plot ---
         residuals = true_i - pred_i
-        fig_res, ax_res = plt.subplots(figsize=(8, 6), dpi=100)
+        fig_res, ax_res = plt.subplots(figsize=(8, 6), dpi=DPI_value)
         ax_res.scatter(pred_i, residuals, alpha=0.6, edgecolors='k', s=50)
         ax_res.axhline(0, color='red', linestyle='--')
         ax_res.set_xlabel("Predicted Values")
@@ -103,7 +105,7 @@ def multi_target_regression_metrics(
         plt.close(fig_res)
 
         # --- Save True vs. Predicted Plot ---
-        fig_tvp, ax_tvp = plt.subplots(figsize=(8, 6), dpi=100)
+        fig_tvp, ax_tvp = plt.subplots(figsize=(8, 6), dpi=DPI_value)
         ax_tvp.scatter(true_i, pred_i, alpha=0.6, edgecolors='k', s=50)
         ax_tvp.plot([true_i.min(), true_i.max()], [true_i.min(), true_i.max()], 'k--', lw=2)
         ax_tvp.set_xlabel('True Values')
@@ -127,7 +129,10 @@ def multi_label_classification_metrics(
     y_prob: np.ndarray,
     target_names: List[str],
     save_dir: Union[str, Path],
-    threshold: float = 0.5
+    threshold: float = 0.5,
+    ROC_PR_line: str='darkorange',
+    cmap: str = "Blues",
+    font_size: int = 16
 ):
     """
     Calculates and saves classification metrics for each label individually.
@@ -158,6 +163,10 @@ def multi_label_classification_metrics(
     
     # Generate binary predictions from probabilities
     y_pred = (y_prob >= threshold).astype(int)
+    
+    # --- Save current RC params and update font size ---
+    original_rc_params = plt.rcParams.copy()
+    plt.rcParams.update({'font.size': font_size})
 
     _LOGGER.info("--- Multi-Label Classification Evaluation ---")
 
@@ -174,7 +183,7 @@ def multi_label_classification_metrics(
         f"Jaccard Score (macro): {j_score_macro:.4f}\n"
         f"--------------------------------------------------\n"
     )
-    print(overall_report)
+    # print(overall_report)
     overall_report_path = save_dir_path / "classification_report_overall.txt"
     overall_report_path.write_text(overall_report)
 
@@ -192,8 +201,26 @@ def multi_label_classification_metrics(
         report_path.write_text(report_text) # type: ignore
 
         # --- Save Confusion Matrix ---
-        fig_cm, ax_cm = plt.subplots(figsize=(6, 6), dpi=100)
-        ConfusionMatrixDisplay.from_predictions(true_i, pred_i, cmap="Blues", ax=ax_cm)
+        fig_cm, ax_cm = plt.subplots(figsize=(6, 6), dpi=DPI_value)
+        disp_ = ConfusionMatrixDisplay.from_predictions(true_i, 
+                                                pred_i, 
+                                                cmap=cmap, 
+                                                ax=ax_cm, 
+                                                normalize='true',
+                                                labels=[0, 1],
+                                                display_labels=["Negative", "Positive"])
+        
+        disp_.im_.set_clim(vmin=0.0, vmax=1.0)
+        
+        # Turn off gridlines
+        ax_cm.grid(False)
+        
+        # Manually update font size of cell texts
+        for text in ax_cm.texts:
+            text.set_fontsize(font_size)
+
+        fig_cm.tight_layout()
+        
         ax_cm.set_title(f"Confusion Matrix for '{name}'")
         cm_path = save_dir_path / f"confusion_matrix_{sanitized_name}.svg"
         plt.savefig(cm_path)
@@ -202,8 +229,8 @@ def multi_label_classification_metrics(
         # --- Save ROC Curve ---
         fpr, tpr, _ = roc_curve(true_i, prob_i)
         auc = roc_auc_score(true_i, prob_i)
-        fig_roc, ax_roc = plt.subplots(figsize=(6, 6), dpi=100)
-        ax_roc.plot(fpr, tpr, label=f'AUC = {auc:.2f}')
+        fig_roc, ax_roc = plt.subplots(figsize=(6, 6), dpi=DPI_value)
+        ax_roc.plot(fpr, tpr, label=f'AUC = {auc:.2f}', color=ROC_PR_line)
         ax_roc.plot([0, 1], [0, 1], 'k--')
         ax_roc.set_title(f'ROC Curve for "{name}"')
         ax_roc.set_xlabel('False Positive Rate'); ax_roc.set_ylabel('True Positive Rate')
@@ -215,14 +242,17 @@ def multi_label_classification_metrics(
         # --- Save Precision-Recall Curve ---
         precision, recall, _ = precision_recall_curve(true_i, prob_i)
         ap_score = average_precision_score(true_i, prob_i)
-        fig_pr, ax_pr = plt.subplots(figsize=(6, 6), dpi=100)
-        ax_pr.plot(recall, precision, label=f'AP = {ap_score:.2f}')
+        fig_pr, ax_pr = plt.subplots(figsize=(6, 6), dpi=DPI_value)
+        ax_pr.plot(recall, precision, label=f'AP = {ap_score:.2f}', color=ROC_PR_line)
         ax_pr.set_title(f'Precision-Recall Curve for "{name}"')
         ax_pr.set_xlabel('Recall'); ax_pr.set_ylabel('Precision')
         ax_pr.legend(loc='lower left'); ax_pr.grid(True, linestyle='--', alpha=0.6)
         pr_path = save_dir_path / f"pr_curve_{sanitized_name}.svg"
         plt.savefig(pr_path)
         plt.close(fig_pr)
+        
+    # restore RC params
+    plt.rcParams.update(original_rc_params)
 
     _LOGGER.info(f"All individual label reports and plots saved to '{save_dir_path.name}'")
 

ml_tools/ML_trainer.py

@@ -13,11 +13,12 @@ from .keys import PyTorchLogKeys, PyTorchCheckpointKeys, DatasetKeys
 from ._logger import _LOGGER
 from .path_manager import make_fullpath
 from .ML_vision_evaluation import segmentation_metrics, object_detection_metrics
+from .ML_configuration import ClassificationMetricsFormat, MultiClassificationMetricsFormat
 
 
 __all__ = [
     "MLTrainer",
-    "ObjectDetectionTrainer"
+    "ObjectDetectionTrainer",
 ]
 
 
@@ -334,14 +335,16 @@ class MLTrainer:
 
                 yield y_pred_batch, y_prob_batch, y_true_batch
 
-    def evaluate(self, save_dir: Union[str, Path], data: Optional[Union[DataLoader, Dataset]] = None, classification_threshold: float = 0.5):
+    def evaluate(self, 
+                 save_dir: Union[str, Path], 
+                 data: Optional[Union[DataLoader, Dataset]] = None,
+                 format_configuration: Optional[Union[ClassificationMetricsFormat, MultiClassificationMetricsFormat]]=None):
         """
         Evaluates the model, routing to the correct evaluation function based on task `kind`.
 
         Args:
             save_dir (str | Path): Directory to save all reports and plots.
             data (DataLoader | Dataset | None): The data to evaluate on. If None, defaults to the trainer's internal test_dataset.
-            classification_threshold (float): Probability threshold for multi-label tasks.
         """
         dataset_for_names = None
         eval_loader = None
@@ -376,10 +379,10 @@ class MLTrainer:
             _LOGGER.error("Cannot evaluate. No valid data was provided or found.")
             raise ValueError()
 
-        print("\n--- Model Evaluation ---")
+        # print("\n--- Model Evaluation ---")
 
         all_preds, all_probs, all_true = [], [], []
-        for y_pred_b, y_prob_b, y_true_b in self._predict_for_eval(eval_loader, classification_threshold):
+        for y_pred_b, y_prob_b, y_true_b in self._predict_for_eval(eval_loader):
             if y_pred_b is not None: all_preds.append(y_pred_b)
             if y_prob_b is not None: all_probs.append(y_prob_b)
             if y_true_b is not None: all_true.append(y_true_b)
@@ -397,7 +400,19 @@ class MLTrainer:
             regression_metrics(y_true.flatten(), y_pred.flatten(), save_dir)
 
         elif self.kind == "classification":
-            classification_metrics(save_dir, y_true, y_pred, y_prob)
+            # Parse configuration
+            if format_configuration and isinstance(format_configuration, ClassificationMetricsFormat):
+                classification_metrics(save_dir=save_dir,
+                                       y_true=y_true, 
+                                       y_pred=y_pred, 
+                                       y_prob=y_prob,
+                                       cmap=format_configuration.cmap,
+                                       class_map=format_configuration.class_map,
+                                       ROC_PR_line=format_configuration.ROC_PR_line,
+                                       calibration_bins=format_configuration.calibration_bins,
+                                       font_size=format_configuration.font_size)
+            else:
+                classification_metrics(save_dir, y_true, y_pred, y_prob)
 
         elif self.kind == "multi_target_regression":
             try:
@@ -419,7 +434,18 @@ class MLTrainer:
             if y_prob is None:
                 _LOGGER.error("Evaluation for multi_label_classification requires probabilities (y_prob).")
                 return
-            multi_label_classification_metrics(y_true, y_prob, target_names, save_dir, classification_threshold)
+            
+            if format_configuration and isinstance(format_configuration, MultiClassificationMetricsFormat):
+                multi_label_classification_metrics(y_true=y_true,
+                                                   y_prob=y_prob,
+                                                   target_names=target_names,
+                                                   save_dir=save_dir,
+                                                   threshold=format_configuration.threshold,
+                                                   ROC_PR_line=format_configuration.ROC_PR_line,
+                                                   cmap=format_configuration.cmap,
+                                                   font_size=format_configuration.font_size)
+            else:
+                multi_label_classification_metrics(y_true, y_prob, target_names, save_dir)
             
         elif self.kind == "segmentation":
             class_names = None
@@ -445,7 +471,7 @@ class MLTrainer:
             
             segmentation_metrics(y_true, y_pred, save_dir, class_names=class_names)
         
-        print("\n--- Training History ---")
+        # print("\n--- Training History ---")
         plot_losses(self.history, save_dir=save_dir)
     
     def explain(self,
@@ -502,7 +528,7 @@ class MLTrainer:
             rand_indices = torch.randperm(full_data.size(0))[:num_samples]
             return full_data[rand_indices]
 
-        print(f"\n--- Preparing SHAP Data (sampling up to {n_samples} instances) ---")
+        # print(f"\n--- Preparing SHAP Data (sampling up to {n_samples} instances) ---")
 
         # 1. Get background data from the trainer's train_dataset
         background_data = _get_random_sample(self.train_dataset, n_samples)
@@ -610,7 +636,7 @@ class MLTrainer:
             plot_n_features (int): Number of top features to plot.
         """
         
-        print("\n--- Attention Analysis ---")
+        # print("\n--- Attention Analysis ---")
         
         # --- Step 1: Check if the model supports this explanation ---
         if not getattr(self.model, 'has_interpretable_attention', False):
@@ -994,7 +1020,7 @@ class ObjectDetectionTrainer:
             _LOGGER.error("Cannot evaluate. No valid data was provided or found.")
             raise ValueError()
 
-        print("\n--- Model Evaluation ---")
+        # print("\n--- Model Evaluation ---")
 
         all_predictions = []
         all_targets = []
@@ -1043,7 +1069,7 @@ class ObjectDetectionTrainer:
             print_output=False
         )
         
-        print("\n--- Training History ---")
+        # print("\n--- Training History ---")
         plot_losses(self.history, save_dir=save_dir)
     
     def _callbacks_hook(self, method_name: str, *args, **kwargs):

ml_tools/ML_utilities.py

@@ -10,6 +10,7 @@ from ._logger import _LOGGER
 from .keys import DatasetKeys, PytorchModelArchitectureKeys, PytorchArtifactPathKeys, SHAPKeys, UtilityKeys, PyTorchCheckpointKeys
 from .utilities import load_dataframe
 from .custom_logger import save_list_strings, custom_logger
+from .serde import serialize_object_filename
 
 
 __all__ = [
@@ -18,7 +19,8 @@ __all__ = [
     "get_model_parameters",
     "inspect_model_architecture",
     "inspect_pth_file",
-    "set_parameter_requires_grad"
+    "set_parameter_requires_grad",
+    "save_pretrained_transforms"
 ]
 
 
@@ -524,5 +526,52 @@ def _set_params_grad(
     return params_changed
 
 
+def save_pretrained_transforms(model: nn.Module, output_dir: Union[str, Path]):
+    """
+    Checks a model for the 'self._pretrained_default_transforms' attribute, if found,
+    serializes the returned transform object as a .joblib file.
+
+    This saves the callable transform object itself for
+    later use, such as passing it directly to the 'transform_source'
+    argument of the PyTorchVisionInferenceHandler.
+
+    Args:
+        model (nn.Module): The model instance to check.
+        output_dir (str | Path): The directory where the transform file will be saved.
+    """
+    output_filename = "pretrained_model_transformations"
+
+    # 1. Check for the "secret attribute"
+    if not hasattr(model, '_pretrained_default_transforms'):
+        _LOGGER.warning(f"Model of type {type(model).__name__} does not have the required attribute. No transformations saved.")
+        return
+
+    # 2. Get the transform object
+    try:
+        transform_obj = model._pretrained_default_transforms
+    except Exception as e:
+        _LOGGER.error(f"Error calling the required attribute on model: {e}")
+        return
+
+    # 3. Check if the object is actually there
+    if transform_obj is None:
+        _LOGGER.warning(f"Model {type(model).__name__} has the required attribute but returned None. No transforms saved.")
+        return
+
+    # 4. Serialize and save using serde
+    try:
+        serialize_object_filename(
+            obj=transform_obj,
+            save_dir=output_dir,
+            filename=output_filename,
+            verbose=True,
+            raise_on_error=True
+        )
+        # _LOGGER.info(f"Successfully saved pretrained transforms to '{output_dir}'.")
+    except Exception as e:
+        _LOGGER.error(f"Failed to serialize transformations: {e}")
+        raise
+
+
 def info():
     _script_info(__all__)