facebook's implementation

Tune.py

@@ -34,7 +34,7 @@ TRAINING_SMPLE = 27
 
 # learning rate
 LR_RATE = 0.0001
-EPOCHS = 20
+EPOCHS = 10
 
 # depends on model architecture
 # ResNet, DenseNet = 224
@@ -109,7 +109,7 @@ print(f'> Constants : Classes = {CLASS_SIZE}, Batch = {BATCH_SIZE}, epochs = {EP
 device = SetUp.get_device()
 
 
-for i in range(0,CLASS_SIZE):
+for i in range(0,1):#CLASS_SIZE):
     FORGET_CLASS_IDX = i
     # Create model using Factory
     model = Model.create(
@@ -118,13 +118,13 @@ for i in range(0,CLASS_SIZE):
         size = CLASS_SIZE)
 
     # we may need to load existing model or finetune
-    model.train(
-            epochs = EPOCHS, 
-            loader = train_loader, 
-            rate = LR_RATE)
+    #model.train(
+    #        epochs = EPOCHS, 
+    #        loader = train_loader, 
+    #        rate = LR_RATE)
 
         # save.
-    model.save(filename=arch.name.lower())
+    #model.save(filename=arch.name.lower())
 
 
     # done tuning
@@ -147,10 +147,10 @@ for i in range(0,CLASS_SIZE):
 
     # Evaluate
     current_mode = "Finetuned"
-    accuracy, report_dict = model.evaluate(
-        loader = test_loader,
-        mode=current_mode
-    )
+    #accuracy, report_dict = model.evaluate(
+    #    loader = test_loader,
+    #    mode=current_mode
+    #)
 
     Util._log_to_csv(
         arch=model.__class__.__name__, 
@@ -161,13 +161,13 @@ for i in range(0,CLASS_SIZE):
     )
 
     # unlearning algorithms
-    linear_filtration = LinearFiltration(target_class_idx=FORGET_CLASS_IDX)
+    #linear_filtration = LinearFiltration(target_class_index=FORGET_CLASS_IDX)
     #filtration.apply(reloaded.model)
 
-    weight_filtration = WeightFiltration(num_classes = CLASS_SIZE,target_class_idx=FORGET_CLASS_IDX)
+    #weight_filtration = WeightFiltration(num_classes = CLASS_SIZE,target_class_idx=FORGET_CLASS_IDX)
     #weight_filtration.apply(reloaded.model)
 
-    certified_removal = CertifiedRemoval(removal_bound=0.05, epsilon=0.5, l2_reg=0.1)
+    certified_removal = CertifiedRemoval(target_class_index=FORGET_CLASS_IDX,removal_bound=0.05, epsilon=0.5, l2_reg=15)
     #certified_removal.apply(reloaded.model)
 
     # to be unlearned
@@ -179,14 +179,14 @@ for i in range(0,CLASS_SIZE):
 
     # to evaluate
     forget_test_loader, retain_test_loader = get_unlearning_loaders(
-            dataset=test_data, 
-            forget_class_idx=FORGET_CLASS_IDX, 
-            batch_size=BATCH_SIZE
-        )
+        dataset=test_data, 
+        forget_class_idx=FORGET_CLASS_IDX, 
+        batch_size=BATCH_SIZE
+    )
 
 
-    strategies = [linear_filtration, weight_filtration, certified_removal]
-    #strategies = [linear_filtration]
+    #strategies = [linear_filtration, weight_filtration, certified_removal]
+    strategies = [certified_removal]
     for strategy in strategies:
         # test again
         reloaded = Model.create(