I cannot create an instance of BeliefPropagation from a FactorGraph

[vicen]

Subject of the issue

I cannot create an instance of BeliefPropagation from a FactorGraph
AttributeError: module 'networkx' has no attribute 'cliques_containing_node'

The problem appears to be that pgmpy uses an outdated version of networkx in the triangulate method.

Your environment

standard Google colab environment

• pgmpy version (0.1.24)
• Python version 3.10.12 (main, Nov 20 2023, 15:14:05) [GCC 11.4.0]
• Operating System on linux

Steps to reproduce

from pgmpy.inference.ExactInference import BeliefPropagation

def create_factor(i, j, J):
  values = [np.exp(J * x_i * x_j) for x_i in [-1, 1] for x_j in [-1, 1]]
  return DiscreteFactor(variables=[i, j],
                        cardinality=[2, 2],
                        values=values,
                        state_names={0:[-1,1],1:[-1,1]})
f = dict()
f[(0,1)] = create_factor(0,1,2)
f[(1,2)] = create_factor(1,2,2)
f[(2,3)] = create_factor(2,3,2)
f[(3,0)] = create_factor(3,0,2)

G = FactorGraph()
G.add_nodes_from([0,1,2,3])
G.add_factors(*f.values())
G.add_edges_from([(v, f) for f in f.values() for v in f.variables])
bp = BeliefPropagation(G)

Expected behaviour

The BeliefPropagation object should be correctly created

Actual behaviour

Tell us what happens instead
AttributeError: module 'networkx' has no attribute 'cliques_containing_node'

[tristantreb]

Hi, I've plugged your code snippet on my IDE (VSCode). It worked. What's your networkx version (I've got 3.1)?

Here is my setting:
pgmpy 0.1.19
networkx 3.1
python 3.10.13
macOS Ventura 13.2

[ankurankan]

@vicen This has been fixed since the last release (0.1.24) but hasn't been released yet. Updating pgmpy to the github dev branch should fix the error:

pip install git+https://github.com/pgmpy/pgmpy.git@dev