Generic inference engine using inference by enumeration with Bayesian networks. Implemented in Python using the Pomegranate framework.

• Daniela Vignau
• Héctor Reyes

• Overview
• Installation
  • Requirements
• Usage
  • Example
• License

A Bayesian network is a probabilistic model that represents a set of variables and their conditional dependencies through a directed acyclic graph. Given an event, the likelihood that such an event occurred as a result of one or more known causes can be inferred from a Bayesian network. For example, Bayesian networks could represent the probabilistic relationships between diseases and symptoms: given symptoms S₁, ..., S_n, a model can be used to calculate the probability of the presence of a disease D as a query Q = P(D | S₁, ..., S_n).

This project provides a generic inference engine using inference by enumeration with Bayesian networks. It receives a JSON file with the probabilities associated with each node on the network as input, and then writes the probability assigned to the query to an output JSON file.

Install from this repository:

git clone https://github.com/hreyesm/inference-engine-with-bayesian-networks

The script was tested in an Anaconda environment running on MacOS.

• Python (3.7 or above)
• Pomegranate

After cloning the repository, enter the following terminal command to run the Python script:

python3 engine.py <input JSON file>

Please note that <input JSON file> should be replaced with a JSON file that contains a valid Bayesian network topology (see example).

Consider a Bayesian network from which the probability assigned to the query Q = P(G|R-S) is to be inferred, that is, the probability that the grass is wet (G) given that it rained (R) and there was no sprinkler (-S):

For the above topology, a valid input JSON file would look like the following:

[
  {
    "R": 0.2
  },

  {
    "S|R": 0.01,
    "S|-R": 0.4
  },

  {
    "G|RS": 0.99,
    "G|R-S": 0.8,
    "G|-RS": 0.9,
    "G|-R-S": 0.0
  },

  {
    "BELIEF": "R"
  },

  {
    "BELIEF": "-S"
  },

  {
    "QUERY": "G"
  }
]

The example_input.json file already contains this topology, so when the command python3 engine.py example_input.json is run, the probability assigned to the query Q = P(G|R-S) would be written to an output JSON file "G|R-S.json" as follows:

{"-G": 0.20000000000000007, "G": 0.7999999999999999}

Thus, the probability that the grass is wet (G) given that it rained (R) and there was no sprinkler (-S) is 80%.

The code of this repository was implemented by Héctor Reyes and Daniela Vignau. Released under the MIT license.