def get_trivial_graph(n_node_features=0):
"""Generate a trivial graph for internal purposes."""
nodes = pd.DataFrame([0, 1, 2])
if n_node_features > 0:
nodes["features"] = 3 * [n_node_features * [0.0]]
edges = pd.DataFrame()
edges["start_node"] = [0, 1, 2]
edges["end_node"] = [1, 2, 0]
return Graph(nodes, edges, 0)
def make(folder="./datasets",
add_features=False,
write_to_file=True,
n_graphs=5):
""" Makes pickle with graphs that test robustness of hcga """
graphs = []
# one, two and three node graphs
for _ in range(n_graphs):
graphs.append(
_add_graph_desc(nx.grid_graph([1]).copy(), "one-node graph"))
graphs.append(
_add_graph_desc(nx.grid_graph([2]).copy(), "two-node graph"))
graphs.append(
_add_graph_desc(nx.grid_graph([3]).copy(), "three-node graph"))
# no edges
G = nx.Graph()
G.add_node(0)
G.add_node(1, weight=2)
G.add_node(2, weight=3)
for _ in range(n_graphs):
graphs.append(_add_graph_desc(G.copy(), "graph without edges"))
# directed graph no weights
G = nx.DiGraph()
G.add_nodes_from(range(100, 110))
for _ in range(n_graphs):
graphs.append(
_add_graph_desc(G.copy(), "directed graph with no weights"))
# directed graph weighted
G = nx.DiGraph()
H = nx.path_graph(10)
G.add_nodes_from(H)
G.add_edges_from(H.edges)
for _ in range(n_graphs):
graphs.append(_add_graph_desc(G.copy(), "directed graph weighted"))
# adding features to all
if add_features:
graphs = [add_dummy_node_features(graph) for graph in graphs]
graphs_coll = GraphCollection()
for graph in graphs:
nodes = pd.DataFrame(list(graph.nodes))
edges = pd.DataFrame(columns=["start_node", "end_node"])
for i, edge in enumerate(graph.edges):
edges.loc[i, "start_node"] = edge[0]
edges.loc[i, "end_node"] = edge[1]
graphs_coll.add_graph(Graph(nodes, edges, np.random.randint(0, 2)))
if write_to_file:
save_dataset(graphs_coll, "TESTDATA", folder=folder)
return graphs_coll
def extract_benchmark_graphs(datadir, dataname): # pylint: disable=too-many-locals
""" Read data from https://ls11-www.cs.tu-dortmund.de/staff/morris/graphkerneldatasets
graph index starts with 1 in file
Returns:
List of networkx objects with graph and node labels
"""
prefix = str(Path(datadir) / dataname)
with open(prefix + "_graph_indicator.txt") as f:
nodes_df = pd.read_csv(f, dtype=np.int, header=None) - 1
nodes_df.columns = ["graph_id"]
with open(prefix + "_graph_labels.txt") as f:
graph_labels = pd.read_csv(f, header=None)
edges_df = pd.DataFrame()
with open(prefix + "_A.txt") as f:
for edges_df_next in pd.read_csv(
f, sep=",", delimiter=None, dtype=np.int, header=None, chunksize=1e6
):
edges_df = edges_df.append(edges_df_next - 1)
edges_df.columns = ["start_node", "end_node"]
edges_df["graph_id"] = nodes_df["graph_id"][
edges_df["start_node"].to_list()
].to_list()
columns = []
if Path(prefix + "_node_labels.txt").exists():
with open(prefix + "_node_labels.txt") as f:
nodes_df["labels_value"] = pd.read_csv(f, header=None)
nodes_df["labels"] = list(
pd.get_dummies(nodes_df["labels_value"]).to_numpy(dtype=float)
)
columns.append("labels")
if Path(prefix + "_node_attributes.txt").exists():
with open(prefix + "_node_attributes.txt") as f:
nodes_df["attributes"] = list(pd.read_csv(f, header=None).to_numpy())
columns.append("attributes")
graph_ids = list(set(nodes_df["graph_id"]))
graphs = GraphCollection()
for graph_id in graph_ids:
nodes = nodes_df.loc[nodes_df["graph_id"] == graph_id][columns]
edges = edges_df.loc[edges_df["graph_id"] == graph_id][
["start_node", "end_node"]
]
graphs.add_graph(Graph(nodes, edges, int(graph_labels.loc[graph_id][0])))
return graphs
def _add_graph(label):
for _ in range(n_graphs):
graph = nx.stochastic_block_model(sizes, probs)
graphs.add_graph(Graph(list(graph.nodes), list(graph.edges),
label))