add support for isolated vertices when creating graph from numpy array

Author: jnke2016

python/cugraph/cugraph/structure/convert_matrix.py

@@ -434,7 +434,7 @@ def to_pandas_adjacency(G):
     return pdf
 
 
-def from_numpy_array(A, create_using=Graph):
+def from_numpy_array(A, create_using=Graph, vertices=None):
     """
     Initializes the graph from numpy array containing adjacency matrix.
 
@@ -446,6 +446,14 @@ def from_numpy_array(A, create_using=Graph):
     create_using: cugraph.Graph (instance or class), optional (default=Graph)
         Specify the type of Graph to create. Can pass in an instance to create
         a Graph instance with specified 'directed' attribute.
+    
+    vertices : cudf.Series or List, optional (default=None)
+        A cudf.Series or list containing all vertices of the graph. This is
+        optional, but must be used if the graph contains isolated vertices
+        which cannot be represented in the source and destination arrays.
+        If specified, this array must contain every vertex identifier,
+        including vertex identifiers that are already included in the
+        source and destination arrays.
     """
     if create_using is None:
         G = Graph()
@@ -461,7 +469,7 @@ def from_numpy_array(A, create_using=Graph):
             f"{type(create_using)}"
         )
 
-    G.from_numpy_array(A)
+    G.from_numpy_array(A, vertices=vertices)
     return G
 
 

python/cugraph/cugraph/structure/graph_classes.py

@@ -489,7 +489,7 @@ def from_pandas_adjacency(self, pdf):
         columns = pdf.columns
         self.from_numpy_array(np_array, columns)
 
-    def from_numpy_array(self, np_array, nodes=None):
+    def from_numpy_array(self, np_array, nodes=None, vertices=None):
         """
         Initializes the graph from numpy array containing adjacency matrix.
 
@@ -500,6 +500,14 @@ def from_numpy_array(self, np_array, nodes=None):
 
         nodes: array-like or None, optional (default=None)
             A list of column names, acting as labels for nodes
+        
+        vertices : cudf.Series or List, optional (default=None)
+            A cudf.Series or list containing all vertices of the graph. This is
+            optional, but must be used if the graph contains isolated vertices
+            which cannot be represented in the source and destination arrays.
+            If specified, this array must contain every vertex identifier,
+            including vertex identifiers that are already included in the
+            source and destination arrays.
         """
         np_array = np.asarray(np_array)
         if len(np_array.shape) != 2:
@@ -515,7 +523,7 @@ def from_numpy_array(self, np_array, nodes=None):
             df["src"] = src
             df["dst"] = dst
         df["weight"] = weight
-        self.from_cudf_edgelist(df, "src", "dst", edge_attr="weight")
+        self.from_cudf_edgelist(df, "src", "dst", edge_attr="weight", vertices=vertices)
 
     def from_numpy_matrix(self, np_matrix):
         """

python/cugraph/cugraph/tests/structure/test_graph.py

@@ -18,6 +18,7 @@
 import pandas as pd
 import scipy
 import networkx as nx
+import numpy as np
 
 import cupy
 import cudf
@@ -646,6 +647,35 @@ def test_number_of_edges():
     assert G_directed.number_of_edges() == G_undirected.number_of_edges()
 
 
+@pytest.mark.sg
+def test_vertex_list():
+    A = np.array([[0., 0., 0., 0., 0., 0., 0., 1., 0., 0.],
+       [0., 0., 0., 0., 0., 1., 0., 0., 1., 0.],
+       [0., 0., 0., 0., 0., 0., 0., 1., 0., 0.],
+       [0., 0., 0., 0., 0., 0., 0., 1., 0., 0.],
+       [0., 0., 0., 0., 0., 0., 0., 0., 0., 0.],
+       [0., 0., 0., 0., 0., 0., 0., 0., 0., 0.],
+       [0., 0., 0., 0., 0., 0., 0., 0., 0., 0.],
+       [0., 0., 0., 0., 0., 0., 0., 1., 0., 0.],
+       [0., 0., 0., 0., 0., 0., 0., 1., 0., 0.],
+       [0., 0., 0., 0., 0., 0., 0., 0., 0., 0.]])
+
+    G = nx.from_numpy_array(A)
+    cG = cugraph.from_numpy_array(nx.to_numpy_array(G))
+
+    assert len(G.nodes()) == len(cG.nodes())
+
+    nx_nodes = cudf.Series([n for n in G.nodes()])
+    cG_nodes = cG.nodes().sort_values(ignore_index=True)
+
+    assert_series_equal(
+        nx_nodes,
+        cG_nodes,
+        check_names=False,
+        check_dtype=False,
+    )
+
+
 # Test
 @pytest.mark.sg
 @pytest.mark.parametrize("graph_file", utils.DATASETS_SMALL)