def read_graph(graph_file, dataset_name, directed=False, weighted=False):
"""
Reads the input network in networkx.
:param graph_file: The directory where graph in EPGM format is stored
:param dataset_name: The name of the graph selected out of all the graph heads in EPGM file
:return: The graph in networkx format
"""
try: # assume args.input points to an EPGM graph
G_epgm = EPGM(graph_file)
graphs = G_epgm.G["graphs"]
if (
dataset_name is None
): # if dataset_name is not given, use the name of the 1st graph head
dataset_name = graphs[0]["meta"]["label"]
print(
"WARNING: dataset name not specified, using dataset '{}' in the 1st graph head"
.format(dataset_name))
graph_id = None
for g in graphs:
if g["meta"]["label"] == dataset_name:
graph_id = g["id"]
g = G_epgm.to_nx(graph_id, directed)
if weighted:
raise NotImplementedError
else:
# This is the correct way to set the edge weight in a MultiGraph.
edge_weights = {e: 1 for e in g.edges(keys=True)}
nx.set_edge_attributes(g, name="weight", values=edge_weights)
except: # otherwise, assume arg.input points to an edgelist file
if weighted:
g = nx.read_edgelist(
graph_file,
nodetype=int,
data=(("weight", float), ),
create_using=nx.DiGraph(),
)
else:
g = nx.read_edgelist(graph_file,
nodetype=int,
create_using=nx.DiGraph())
for edge in g.edges():
g[edge[0]][edge[1]]["weight"] = 1
if not directed:
g = g.to_undirected()
if not nx.is_connected(g):
print("Graph is not connected")
# take the largest connected component as the data
g_ccs = (g.subgraph(c).copy() for c in nx.connected_components(g))
g = max(g_ccs, key=len)
print("Largest subgraph statistics: {} nodes, {} edges".format(
g.number_of_nodes(), g.number_of_edges()))
print("Graph statistics: {} nodes, {} edges".format(
g.number_of_nodes(), g.number_of_edges()))
return g
def test_load_epgm(self):
"""Test that the EPGM is loaded correctly from epgm path"""
G_epgm = EPGM(self.input_dir)
print(self.input_dir)
assert "graphs" in G_epgm.G.keys()
assert "vertices" in G_epgm.G.keys()
assert "edges" in G_epgm.G.keys()
# check that G_epgm.G['graphs] has at least one graph head:
assert len(G_epgm.G["graphs"]) > 0
# cora nodes should have a subject attribute
graph_id = G_epgm.G["graphs"][0]["id"]
assert self.target_attribute in G_epgm.node_attributes(
graph_id, self.node_type)
# cora should have 2708 vertices
n_nodes = 2708
nodes = G_epgm.G["vertices"]
assert len(nodes) == n_nodes
# cora nodes should have 7 unique values for subject attribute:
assert sum(["data" in v for v in nodes]) == n_nodes
subjects = np.unique([v["data"][self.target_attribute] for v in nodes])
assert len(subjects) == 7
def test_load_epgm(self):
"""Test that the EPGM is loaded correctly from epgm path"""
G_epgm = EPGM(self.input_dir)
assert "graphs" in G_epgm.G.keys()
assert "vertices" in G_epgm.G.keys()
assert "edges" in G_epgm.G.keys()
# check that G_epgm.G['graphs] has at least one graph head:
assert len(G_epgm.G["graphs"]) > 0
# graph nodes of self.node_type type should have a self.target_attribute attribute
graph_id = G_epgm.G["graphs"][0]["id"]
assert self.target_attribute in G_epgm.node_attributes(
graph_id, self.node_type)
# graph should have 260 vertices
n_nodes = 260
nodes = G_epgm.G["vertices"]
assert len(nodes) == n_nodes
# 'user' nodes should have 3 unique values for 'elite' attribute:
# first make sure that all nodes have 'data' key
assert sum(["data" in v for v in nodes]) == n_nodes
labels_all = [v["data"].get(self.target_attribute) for v in nodes]
labels = list(filter(lambda l: l is not None, labels_all))
assert len(np.unique(labels)) == 3
def load_data(path, dataset_name=None, node_type=None, target_attribute=None):
"""
Loads the node data
:param path: Input filename or directory where graph in EPGM format is stored
:param node_type: For HINs, the node type to consider
:param target_attribute: For EPGM format, the target node attribute
:return: N x 2 numpy arrays where the first column is the node id and the second column is the node label.
"""
if os.path.isdir(path):
g_epgm = EPGM(path)
graphs = g_epgm.G["graphs"]
for g in graphs:
if g["meta"]["label"] == dataset_name:
g_id = g["id"]
g_vertices = g_epgm.G["vertices"] # retrieve all graph vertices
if node_type is None:
node_type = g_epgm.node_types(g_id)
if len(node_type) == 1:
node_type = node_type[0]
else:
raise Exception(
"Multiple node types detected in graph {}: {}.".format(
g_id, node_type
)
)
if target_attribute is None:
target_attribute = g_epgm.node_attributes(g_id, node_type)
if len(target_attribute) == 1:
target_attribute = target_attribute[0]
else:
raise Exception(
"Multiple node attributes detected for nodes of type {} in graph {}: {}.".format(
node_type, g_id, target_attribute
)
)
y = np.array(
get_nodes(
g_vertices, node_type=node_type, target_attribute=target_attribute
)
)
else:
y_df = pd.read_csv(path, delimiter=" ", header=None, dtype=str)
y_df.sort_values(by=[0], inplace=True)
y = y_df.values
return y
def test_node_types(self):
"""Test the .node_types() method"""
G_epgm = EPGM(self.input_dir)
graph_id = G_epgm.G["graphs"][0]["id"]
# cora has a single 'paper' node type:
node_types = G_epgm.node_types(graph_id)
assert len(node_types) == 1
assert self.node_type in node_types
with pytest.raises(Exception):
G_epgm.node_types("invalid_graph_id")
def test_node_types(self):
"""Test the .node_types() method"""
G_epgm = EPGM(self.input_dir)
graph_id = G_epgm.G["graphs"][0]["id"]
# dataset has multiple node types:
node_types = G_epgm.node_types(graph_id)
assert len(node_types) == 3
assert "person" in node_types
assert "paper" in node_types
assert "venue" in node_types
with pytest.raises(Exception):
G_epgm.node_types("invalid_graph_id")
def from_epgm(epgm_location, dataset_name=None, directed=False):
"""
Imports a graph stored in EPGM format to a NetworkX object
Args:
epgm_location (str): The directory containing the EPGM data
dataset_name (str), optional: The name of the dataset to import
directed (bool): If True, load as a directed graph, otherwise
load as an undirected graph
Returns:
A NetworkX graph containing the data for the EPGM-stored graph.
"""
G_epgm = EPGM(epgm_location)
graphs = G_epgm.G["graphs"]
# if dataset_name is not given, use the name of the 1st graph head
if not dataset_name:
dataset_name = graphs[0]["meta"]["label"]
warnings.warn(
"dataset name not specified, using dataset '{}' in the 1st graph head".format(
dataset_name
),
RuntimeWarning,
stacklevel=2,
)
# Select graph using dataset_name
for g in graphs:
if g["meta"]["label"] == dataset_name:
graph_id = g["id"]
# Convert to StellarGraph (via nx)
Gnx = G_epgm.to_nx(graph_id, directed=directed)
print(
"Graph statistics: {} nodes, {} edges".format(
Gnx.number_of_nodes(), Gnx.number_of_edges()
)
)
return Gnx
def test_node_attributes(self):
"""Test the .node_attributes() method"""
G_epgm = EPGM(self.input_dir)
graph_id = G_epgm.G["graphs"][0]["id"]
# cora has 1433 unique node attributes, including 'subject'
node_attributes = G_epgm.node_attributes(graph_id, self.node_type)
assert self.target_attribute in node_attributes
# after the predictions cora has 1434 attributes, including subject and subject_PREDICTED
if self.epgm_input:
assert (
len(node_attributes) == 1433
), "There should be 1433 unique node attributes; found {}".format(
len(node_attributes))
else:
assert (
len(node_attributes) == 1434
), "There should be 1434 unique node attributes; found {}".format(
len(node_attributes))
# passing a non-existent node type should return an empty array of node attributes:
assert len(G_epgm.node_attributes(graph_id, "person")) == 0
# if node_type is not supplied, a TypeError should be raised:
with pytest.raises(TypeError):
G_epgm.node_attributes(graph_id)
def test_node_attributes(self):
"""Test the .node_attributes() method"""
G_epgm = EPGM(self.input_dir)
graph_id = G_epgm.G["graphs"][0]["id"]
# dataset has 1 unique 'user' node attribute, 'elite'
node_attributes = G_epgm.node_attributes(graph_id, self.node_type)
assert self.target_attribute in node_attributes
assert (len(node_attributes) == 1
), "There should be 1 unique node attribute; found {}".format(
len(node_attributes))
# passing a non-existent node type should return an empty array of node attributes:
assert len(G_epgm.node_attributes(graph_id, "business")) == 0
# if node_type is not supplied, a TypeError should be raised:
with pytest.raises(TypeError):
G_epgm.node_attributes(graph_id)
def read_graph(graph_file, dataset_name, is_directed=False, is_weighted=False):
"""
Reads the input network in networkx.
Args:
graph_file: The directory where graph in EPGM format is stored.
dataset_name: The name of the graph selected out of all the graph heads in EPGM file.
Returns:
The graph in networkx format
"""
if graph_file.split('.')[-1] == 'gpickle':
g = nx.read_gpickle(graph_file)
for edge in g.edges():
g[edge[0]][edge[1]]["weight"] = 1 # {'weight': 1}
if not is_directed:
g = g.to_undirected()
return g
try: # assume args.input points to an EPGM graph
G_epgm = EPGM(graph_file)
graphs = G_epgm.G["graphs"]
if (
dataset_name is None
): # if dataset_name is not given, use the name of the 1st graph head
dataset_name = graphs[0]["meta"]["label"]
print(
"WARNING: dataset name not specified, using dataset '{}' in the 1st graph head"
.format(dataset_name))
graph_id = None
for g in graphs:
if g["meta"]["label"] == dataset_name:
graph_id = g["id"]
g = G_epgm.to_nx(graph_id, is_directed)
if is_weighted:
raise NotImplementedError
else:
# This is the correct way to set the edge weight in a MultiGraph.
edge_weights = {e: 1 for e in g.edges(keys=True)}
nx.set_edge_attributes(g, name="weight", values=edge_weights)
except: # otherwise, assume arg.input points to an edgelist file
if is_weighted:
g = nx.read_edgelist(
graph_file,
nodetype=int,
data=(("weight", float), ),
create_using=nx.DiGraph(),
)
else:
g = nx.read_edgelist(graph_file,
nodetype=int,
create_using=nx.DiGraph())
for edge in g.edges():
g[edge[0]][edge[1]]["weight"] = 1 # {'weight': 1}
if not is_directed:
g = g.to_undirected()
print("Graph statistics: {} nodes, {} edges".format(
g.number_of_nodes(), g.number_of_edges()))
return g
