optimised

unlearning/Strategy.py

@@ -0,0 +1,62 @@
+
+import torch.nn as nn
+import time
+import os
+from pathlib import Path
+from torch.utils.data import DataLoader
+import Util
+
+class Strategy:
+    """Abstract base class for unlearning algorithms with automated, strategy-specific logging."""
+    
+    def __init__(self, target_class_index):
+        # Dynamically set file name based on the class name (e.g., 'NormalizingLinearFiltration.txt')
+        self.strategy_name = self.__class__.__name__
+        self.target_class_index = target_class_index
+
+    def set_target_class(self, target_class_index: int):
+        """Dynamically switch the unlearning target without retraining."""
+        self.target_class_index = target_class_index
+        
+
+    def apply(self, model: nn.Module, dataset) -> nn.Module:
+        log_file = Path(f"reports/{self.strategy_name}/{model.__class__.__name__}/time_metrics.txt")
+        Util._initialize_log_file(log_file=log_file)
+        """
+        Wraps the unlearning execution with automated timing and strategy-specific logging.
+        DO NOT override this method in subclasses. Override _run instead.
+        """
+    
+
+
+        retain_loader, forget_loader = self._split_data(dataset)
+        
+        # record start time to evaluate efficiency
+        start_time = time.perf_counter()
+        # Execute core unlearning logic
+        processed_model = self._run(model, forget_loader, retain_loader)
+        
+        end_time = time.perf_counter()
+        execution_time = end_time - start_time
+        
+        # Log to the strategy's specific file
+        Util.log_metric(
+            log_file=log_file, 
+            execution_time=execution_time
+        )
+
+        print(f"[{self.strategy_name}] Completed in {execution_time:.6f} seconds. Saved to {log_file}")
+        
+        return processed_model
+
+    def _run(self, model: nn.Module, forget_loader: DataLoader, retain_loader: DataLoader) -> nn.Module:
+        """Subclasses implement their core unlearning logic here."""
+        raise NotImplementedError
+    
+    '''
+    different strategies split data in to different partitions differently. 
+    So a strategy will implement its own and since this part is startegy specific. 
+    not all should compute it the same.
+    '''
+    def _split_data(self,dataset):
+        pass