diff --git a/mantra/representations/simplicial_connectivity.py b/mantra/representations/simplicial_connectivity.py
index faa5c40..0e57ecb 100644
--- a/mantra/representations/simplicial_connectivity.py
+++ b/mantra/representations/simplicial_connectivity.py
@@ -31,7 +31,8 @@ def forward(self, data: Data):
         return data
 
 
-class AbstractSimplicialComplexConnectivity(BaseTransform, ABCMeta):
+class AbstractSimplicialComplexConnectivity(BaseTransform):
+    __metaclass__ = ABCMeta
     """Base class for connectivity transforms.
 
     Parent class for implementing a transform that adds a
@@ -108,28 +109,19 @@ def forward(self, data: Data):
                     self.generate_matrix(
                         data.simplex_trie, rank_idx, max_rank
                     ),
-                    device=data.triangulation.device,
                 )
             except ValueError:
                 idx_low_simp = rank_idx - 1 if rank_idx > 0 else rank_idx
                 if "incidence" in self.connectivity_name:
-                    data[connectivity_name] = (
-                        torch.zeros(
-                            [shape[idx_low_simp], shape[rank_idx]],
-                            layout=torch.sparse_coo,
-                        )
-                        .coalesce()
-                        .to(data.triangulations.device)
-                    )
+                    data[connectivity_name] = torch.zeros(
+                        [shape[idx_low_simp], shape[rank_idx]],
+                        layout=torch.sparse_coo,
+                    ).coalesce()
                 elif "adjacency" in self.connectivity_name:
-                    data[connectivity_name] = (
-                        torch.zeros(
-                            [shape[rank_idx], shape[rank_idx]],
-                            layout=torch.sparse_coo,
-                        )
-                        .coalesce()
-                        .to(data.triangulations.device)
-                    )
+                    data[connectivity_name] = torch.zeros(
+                        [shape[rank_idx], shape[rank_idx]],
+                        layout=torch.sparse_coo,
+                    ).coalesce()
 
         return data
 
@@ -328,7 +320,7 @@ def _from_sparse(data: scipy.sparse.csc_matrix, device=None) -> torch.Tensor:
         input data converted to tensor.
     """
     if device is None:
-        device = data.device("cpu")
+        device = torch.device("cpu")
     # cast from csc_matrix to coo format for compatibility
     coo = data.tocoo()