DBNInference causing errors when fitting the parameter to DBN

[Rajaram1604]

Subject of the issue

DBNInference causing errors when fitting the parameter to DBN

Your environment

• pgmpy version - 3.11.0
• Python version - 0.1.24
• Operating System - Windows 11 Pro

Steps to reproduce

1. Fitting the parameter to DBN by using the fit method
2. Making the inference by using DBNInference
   Example code:
   `from pgmpy.models import DynamicBayesianNetwork as DBN
   from pgmpy.inference import DBNInference
   import numpy as np
   import pandas as pd

Create a DBN

dbn = DBN()

Add Edges

dbn.add_edges_from([
(("A", 0), ("B", 0)),
(("A", 0), ("C", 0)),
(("B", 0), ("D", 0)),
(("C", 0), ("D", 0)),
(("A", 0), ("A", 1)),
(("B", 0), ("B", 1)),
(("C", 0), ("C", 1)),
(("D", 0), ("D", 1)),
])

Generating Random data

data = np.random.randint(low=0, high=2, size=(1000, 20))
print(data)

col_names = []
for t in range(5):
...
col_names.extend([("A", t), ("B", t), ("C", t), ("D", t)])

df = pd.DataFrame(data, columns=col_names)
print(df)

print("Edges: ", dbn.edges)

fit the data into model, Currently only Maximum Likelihood Estimator is supported.

dbn.fit(df, estimator="MLE")
dbn.initialize_initial_state()
assert dbn.check_model()

for cpd in dbn.get_cpds():
print(cpd)

making inference

inference = DBNInference(dbn)
results = inference.query(variables=[("A", 0)], evidence= {("B", 0): 1})
print(results[("A", 0)])

`

Expected behaviour

It should give the inference results

Actual behaviour

Traceback (most recent call last):
File "C:\Users\Rajarampandian.A\PycharmProjects\pythonProject1\dbn_parameter_learning.py", line 43, in
inference = DBNInference(dbn)
^^^^^^^^^^^^^^^^^
File "C:\Users\Rajarampandian.A\PycharmProjects\pythonProject1.venv\Lib\site-packages\pgmpy\inference\dbn_inference.py", line 75, in init
self.one_and_half_junction_tree = one_and_half_markov_model.to_junction_tree()
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "C:\Users\Rajarampandian.A\PycharmProjects\pythonProject1.venv\Lib\site-packages\pgmpy\models\MarkovNetwork.py", line 515, in to_junction_tree
triangulated_graph = self.triangulate()
^^^^^^^^^^^^^^^^^^
File "C:\Users\Rajarampandian.A\PycharmProjects\pythonProject1.venv\Lib\site-packages\pgmpy\models\MarkovNetwork.py", line 433, in triangulate
clique_dict, clique_dict_removed = _get_cliques_dict(node)
^^^^^^^^^^^^^^^^^^^^^^^
File "C:\Users\Rajarampandian.A\PycharmProjects\pythonProject1.venv\Lib\site-packages\pgmpy\models\MarkovNetwork.py", line 409, in _get_cliques_dict
clique_dict = nx.cliques_containing_node(
^^^^^^^^^^^^^^^^^^^^^^^^^^
AttributeError: module 'networkx' has no attribute 'cliques_containing_node'

[ankurankan]

@Rajaram1604 The error is because of changes in the lastest version of networkx. This has already been fixed in pgmpy but hasn't been released yet. Could you give it another try using the dev branch of pgmpy? You can install it using:

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

[Rajaram1604]

Thanks @ankurankan for the quick response, let me try with dev branch installation and update you.

[Rajaram1604]

Yes @ankurankan , its working after installing pgmpy from the dev branch. Thanks for the update.

Also, just wanted to know currently , Is pgmpy supporting the online learning for continuous streaming data for the DBN as i don't find fit_update method in the DynamicBayesianNetwork class as well as is there any algorithm for learning the structure from the data's for DBN by using HillClimbSearch or some other algorithm?

[ankurankan]

@Rajaram1604 Sorry for the late reply. Unfortunately, there's no support for either of those for DBNs yet.

[Rajaram1604]

Will it be incorporated these features with pgmpy in the near future or do you have any timeline?

[ankurankan]

@Rajaram1604 Sorry for the late reply. Unfortunately, we don't have a timeline currently for adding these.