Generalize and parametrize distance functions to compute anomaly scores

baselines/flanders/flanders/main.py

@@ -42,7 +42,7 @@
     set_sklearn_model_params, 
     get_sklearn_model_params
 )
-from .utils import save_results
+from .utils import save_results, l2_norm
 from .server import EnhancedServer
 
 from torch.utils.data import DataLoader
@@ -124,14 +124,15 @@ def client_fn(cid: int, pool_size: int = 10, dataset_name: str = cfg.dataset.nam
         evaluate_fn=evaluate_fn,
         on_fit_config_fn=fit_config,
         fraction_fit=1,
-        fraction_evaluate=0,                # no federated evaluation
+        fraction_evaluate=0,
         min_fit_clients=2,
         min_evaluate_clients=0,
         warmup_rounds=2,
-        to_keep=2,                                    # Used in Flanders, MultiKrum, TrimmedMean (in Bulyan it is forced to 1)
-        min_available_clients=2,                    # All clients should be available
-        window=2,                                      # Used in Flanders
-        sampling=1,                                  # Used in Flanders
+        to_keep=2,
+        min_available_clients=2,
+        window=2,
+        sampling=1,
+        distance_function=l2_norm
     )
 
 

baselines/flanders/flanders/strategy.py

@@ -63,7 +63,8 @@ def __init__(
         maxiter: int = 100,
         sampling: str = None,
         alpha: float = 1,
-        beta: float = 1
+        beta: float = 1,
+        distance_function: Callable = None,
     ) -> None:
         """
         Parameters
@@ -99,6 +100,7 @@ def __init__(
         self.beta = beta
         self.params_indexes = None
         self.malicious_selected = False
+        self.distance_function = distance_function
 
 
     def aggregate_fit(
@@ -109,7 +111,7 @@ def aggregate_fit(
         clients_state: List[bool],
     ) -> Tuple[Optional[Parameters], Dict[str, Scalar]]:
         """
-        Apply MAR forecasting to exclude malicious clients from the average.
+        Apply MAR forecasting to exclude malicious clients from FedAvg.
 
         Parameters
         ----------
@@ -140,18 +142,16 @@ def aggregate_fit(
             print(f"aggregate_fit - Computing MAR on M {M.shape}")
             Mr = mar(M[:,:,:-1], pred_step, maxiter=self.maxiter, alpha=self.alpha, beta=self.beta)
 
-            # TODO: generalize this to user-selected distance functions
             print("aggregate_fit - Computing anomaly scores")
-            delta = np.subtract(M_hat, Mr[:,:,0])
-            anomaly_scores = np.sum(delta**2,axis=-1)**(1./2)
+            anomaly_scores = self.distance_function(M_hat, Mr[:,:,0])
             print(f"aggregate_fit - Anomaly scores: {anomaly_scores}")
 
             print("aggregate_fit - Selecting good clients")
             good_clients_idx = sorted(np.argsort(anomaly_scores)[:self.to_keep])
             malicious_clients_idx = sorted(np.argsort(anomaly_scores)[self.to_keep:])
             results = np.array(results)[good_clients_idx].tolist()
             print(f"aggregate_fit - Good clients: {good_clients_idx}")
-            
+
             print(f"aggregate_fit - clients_state: {clients_state}")
             for idx in good_clients_idx:
                 if clients_state[str(idx)]:

baselines/flanders/flanders/utils.py

@@ -17,6 +17,26 @@
 
 lock = Lock()            # if the script is run on multiple processors we need a lock to save the results
 
+def l2_norm(true_matrix, predicted_matrix):
+    """
+    Compute the l2 norm between two matrices.
+
+    Parameters
+    ----------
+    true_matrix : ndarray
+        The true matrix.
+    predicted_matrix : ndarray
+        The predicted matrix by MAR.
+
+    Returns
+    -------
+    anomaly_scores : ndarray
+        1-d array of anomaly scores.
+    """
+    delta = np.subtract(true_matrix, predicted_matrix)
+    anomaly_scores = np.sum(delta**2,axis=-1)**(1./2)
+    return anomaly_scores
+
 def save_params(parameters, cid, dir="clients_params", remove_last=False, rrl=False):
     """
     Args: