diff --git a/fedeca/strategies/bootstraper.py b/fedeca/strategies/bootstraper.py
index d93affa8..6eb1b746 100644
--- a/fedeca/strategies/bootstraper.py
+++ b/fedeca/strategies/bootstraper.py
@@ -186,10 +186,17 @@ def load_local_state(self, path: Path) -> "TorchAlgo":
         @remote_data
         def predict(self, datasamples, shared_state = None, predictions_path: os.PathLike = None):
             predictions = []
-            for _, chckpt in enumerate(self.checkpoints_list):
-                self._update_from_checkpoint(chckpt)
-                pred_btst = super(strategy.algo.__class__, self).predict(datasamples=datasamples, shared_state=shared_state, predictions_path=predictions_path, _skip=True)
-                predictions.append(pred_btst)
+            with tempfile.TemporaryDirectory() as tmpdirname:
+                paths_to_preds = []
+                for idx, chckpt in enumerate(self.checkpoints_list):
+                    self._update_from_checkpoint(chckpt)
+                    path_to_pred = Path(tmpdirname) / f"bootstrap_{idx}"
+                    super(strategy.algo.__class__, self).predict(datasamples=datasamples, shared_state=shared_state, predictions_path=path_to_pred, _skip=True)
+                    paths_to_preds.append(path_to_pred)
+                with zipfile.ZipFile(predictions_path, 'w') as f:        
+                    for pred in paths_to_preds:
+                        f.write(pred, compress_type=zipfile.ZIP_DEFLATED)
+                
             return predictions
 
     btst_algo = BtstAlgo(**strategy.algo.kwargs)
@@ -443,6 +450,10 @@ def __init__(self):
         data_path="./data",
         backend_type="subprocess",
     )
+
+    for node in test_data_nodes:
+        node.metric_functions = {accuracy.__name__: accuracy}
+
     first_key = list(clients.keys())[0]
 
     aggregation_node = AggregationNode(clients[first_key].organization_info().organization_id)