[BUG] Fix bug with 0s in sparse matrix for LCC calculation

graspologic/utils/utils.py

@@ -503,6 +503,13 @@ def largest_connected_component(
     inds: (optional)
         Indices/nodes from the original adjacency matrix that were kept after taking
         the largest connected component.
+
+    Notes
+    -----
+    For networks input in ``scipy.sparse.csr_matrix`` format, explicit zeros are removed
+    prior to finding the largest connected component, thus they are not treated as
+    edges. This differs from the convention in
+    :func:`scipy.sparse.csgraph.connected_components`.
     """
 
     if isinstance(graph, (nx.Graph, nx.DiGraph, nx.MultiGraph, nx.MultiDiGraph)):
@@ -539,6 +546,9 @@ def _largest_connected_component_adjacency(
     adjacency: Union[np.ndarray, csr_matrix],
     return_inds: bool = False,
 ):
+    if isinstance(adjacency, csr_matrix):
+        adjacency.eliminate_zeros()
+
     # If you treat an undirected graph as directed and take the largest weakly connected
     # component, you'll get the same answer as taking the largest connected component of
     # that undirected graph. So I wrote it this way to avoid the cost of checking for

tests/test_utils.py

@@ -269,6 +269,18 @@ def test_lcc_scipy(self):
         np.testing.assert_array_equal(lcc_matrix.toarray(), expected_lcc_matrix)
         np.testing.assert_array_equal(nodelist, expected_nodelist)
 
+    def test_lcc_scipy_empty(self):
+        adjacency = np.array([[0, 1], [1, 0]])
+        adjacency = csr_matrix(adjacency)
+
+        # remove the actual connecting edges. this is now a disconnected graph
+        # with two nodes. however, scipy still stores the entry that now has a 0 in it
+        # as having a 'nonempty' value, which is used in the lcc calculation
+        adjacency[0, 1] = 0
+        adjacency[1, 0] = 0
+        lcc_adjacency = gus.largest_connected_component(adjacency)
+        assert lcc_adjacency.shape[0] == 1
+
     def test_multigraph_lcc_numpystack(self):
         expected_g_matrix = np.array(
             [[0, 1, 0, 0], [0, 0, 1, 1], [0, 0, 0, 0], [0, 1, 0, 0]]