Lossless Bayesian Network Implementation

Copyright (C) 2025, Shyamal Chandra

Overview

This project implements a lossless Bayesian network in C++ with exact inference capabilities. The implementation maintains all probability information without approximation, providing precise probabilistic reasoning.

Features

• Lossless Representation: All probabilities stored and computed exactly
• Exact Inference: Variable elimination algorithm for precise inference
• DAG Validation: Automatic cycle detection and topological sorting
• Flexible Structure: Support for arbitrary DAG structures
• CPT Management: Efficient storage and access of conditional probability tables
• File I/O: Network serialization and loading capabilities

Building

Requirements

• C++17 compatible compiler (g++, clang++)
• Make

Build Instructions

make

This will compile the project and create the bayesian_network executable.

Running

make run

Or directly:

./bayesian_network

Project Structure

lossless_bayesian_networks/
├── node.hpp                    # Node class definition
├── cpt.hpp                     # Conditional Probability Table class
├── bayesian_network.hpp        # Main Bayesian network class
├── main.cpp                    # Example usage and demonstrations
├── Makefile                    # Build configuration
├── README.md                   # This file
└── docs/
    ├── lossless_bayesian_networks.tex    # Main documentation
    ├── presentation.tex                   # Beamer presentation
    └── reference_manual.tex              # Complete API reference

Documentation

The project includes comprehensive documentation:

1. Main Documentation (docs/lossless_bayesian_networks.tex): Complete guide covering architecture, algorithms, and usage
2. Presentation (docs/presentation.tex): Beamer slides for presentations
3. Reference Manual (docs/reference_manual.tex): Complete API documentation

To compile the LaTeX documents:

cd docs
pdflatex lossless_bayesian_networks.tex
pdflatex presentation.tex
pdflatex reference_manual.tex

Quick Start

#include "bayesian_network.hpp"

// Create network
BayesianNetwork network;

// Add nodes
network.addNode("Disease", "Disease", {"None", "Cold", "Flu"});
network.addNode("Symptom", "Fever", {"No", "Yes"});

// Add edge
network.addEdge("Disease", "Symptom");

// Create and set CPT
std::vector<size_t> dims = {3, 2};
ConditionalProbabilityTable cpt(dims);
cpt.setProbability({0}, 0, 0.9);  // P(Fever=No | Disease=None)
cpt.setProbability({0}, 1, 0.1);  // P(Fever=Yes | Disease=None)
// ... set other probabilities
cpt.normalize();
network.setCPT("Symptom", cpt);

// Perform inference
std::map<std::string, std::string> evidence;
evidence["Symptom"] = "Yes";
std::vector<std::string> query = {"Disease"};
auto results = network.variableElimination(query, evidence);

Examples

The main.cpp file includes two complete examples:

1. Medical Diagnosis Network: Demonstrates disease diagnosis based on symptoms
2. Alarm Network: Classic Bayesian network example with burglary, earthquake, and alarm

License

Copyright (C) 2025, Shyamal Chandra

Contact

For questions or issues, please refer to the documentation or contact the author.