def make_jtree_from_tri_graph(G):
"""returns JT graph"""
# clique graph
CG = nx.Graph()
# maximal weight spanning tree of clique graph is guaranteed to be a junction tree
# (i.e., it satisfies running intersection property)
# where weight is the size of the intersection between adjacent cliques.
CG.add_weighted_edges_from(
(tuple(c1), tuple(c2), -c1c2)
for (c1, c2) in combinations(nx.find_cliques(G), 2)
for c1c2 in [len(set(c1).intersection(set(c2)))] if c1c2 > 0)
JT = nx.Graph(nx.mst(CG)) # Minimal weight spanning tree for CliqueGraph
for src, targ in JT.edges():
JT[src][targ]["sep"] = tuple(set(src).intersection(set(targ)))
return JT
y,
z,
visualize=True)
assert (x.shape == new_x.shape) # check triangulation got everything
x, y, z = new_x, new_y, new_z
if nx.__version__ < '0.99':
raise ImportError('The version of NetworkX must be at least '
'0.99 to run this example')
# Make a NetworkX graph out of our point and edge data
g = build_geometric_graph(x, y, z, edges)
# Compute minimum spanning tree using networkx
# nx.mst returns an edge generator
start_idx, end_idx, _ = np.array(list(nx.mst(g))).T
start_idx = start_idx.astype(np.int)
end_idx = end_idx.astype(np.int)
# Plot this with Mayavi
graph_plot(
x,
y,
z,
start_idx,
end_idx,
edge_scalars=z[start_idx],
opacity=0.8,
colormap='summer',
line_width=4,
)
