diff --git a/MARBLE/preprocessing.py b/MARBLE/preprocessing.py
index 5ab11957..d05844eb 100644
--- a/MARBLE/preprocessing.py
+++ b/MARBLE/preprocessing.py
@@ -24,6 +24,7 @@ def construct_dataset(
     local_gauges=False,
     seed=None,
     metric="euclidean",
+    eigendecomposition=True
 ):
     """Construct PyG dataset from node positions and features.
 
@@ -44,6 +45,8 @@ def construct_dataset(
             embedding dimension is > 2 or dim embedding is not dim of manifold)
         seed: Specify for reproducibility in the furthest point sampling.
               The default is None, which means a random starting vertex.
+        metric: metric used to fit proximity graph
+        eigendecomposition: perform eigendecomposition (needed for diffusion).
     """
 
     anchor = [torch.tensor(a).float() for a in utils.to_list(anchor)]
@@ -112,6 +115,7 @@ def construct_dataset(
         local_gauges=local_gauges,
         n_geodesic_nb=k * frac_geodesic_nb,
         var_explained=var_explained,
+        eigendecomposition=eigendecomposition
     )
 
 
@@ -120,6 +124,7 @@ def _compute_geometric_objects(
     n_geodesic_nb=10,
     var_explained=0.9,
     local_gauges=False,
+    eigendecomposition=True
 ):
     """
     Compute geometric objects used later: local gauges, Levi-Civita connections
@@ -189,9 +194,10 @@ def _compute_geometric_objects(
         kernels = g.gradient_op(data.pos, data.edge_index, gauges)
         Lc = None
 
-    print("\n---- Computing eigendecomposition ... ", end="")
-    L = g.compute_eigendecomposition(L)
-    Lc = g.compute_eigendecomposition(Lc)
+    if eigendecomposition:
+        print("\n---- Computing eigendecomposition ... ", end="")
+        L = g.compute_eigendecomposition(L)
+        Lc = g.compute_eigendecomposition(Lc)
 
     data.kernels = [
         utils.to_SparseTensor(K.coalesce().indices(), value=K.coalesce().values()) for K in kernels