ensured row major ordering

model/pytorch/dcrnn_cell.py

@@ -74,9 +74,10 @@ class DCGRUCell(torch.nn.Module):
     def _build_sparse_matrix(L):
         L = L.tocoo()
         indices = np.column_stack((L.row, L.col))
+        # this is to ensure row-major ordering to equal torch.sparse.sparse_reorder(L)
+        indices = indices[np.lexsort((indices[:, 0], indices[:, 1]))]
         L = torch.sparse_coo_tensor(indices.T, L.data, L.shape, device=device)
         return L
-        # return torch.sparse.sparse_reorder(L)
 
     def forward(self, inputs, hx):
         """Gated recurrent unit (GRU) with Graph Convolution.