dit/dit

What it solves

Information theory is often used to quantify dependencies in data, but modern scientific research frequently encounters complex, multivariate settings where standard tools are insufficient or ambiguous. dit provides a centralized, consistent library of information-theoretic measures, allowing researchers to compare different methods and select the most appropriate one for their specific context.

How it works

dit allows users to define probability distributions and then apply a wide array of information-theoretic calculations to them. The package handles the heavy lifting for entropies, mutual informations, divergences, and common informations. Beyond simple calculations, it includes modules for constructing source codes (like Huffman or Shannon-Fano) and channel codes (like Hamming or LDPC), as well as a catalog of canonical communication channels for testing and analysis.

Who it’s for

This package is designed for information theorists, data scientists, and researchers who need to quantify dependencies within complex systems or perform coding-theoretic analysis using Python.

Highlights

• Extensive library of multivariate and conditional information measures.
• Built-in tools for constructing and evaluating lossless source codes.
• Comprehensive catalog of discrete memoryless channels for coding experiments.
• Flexible distribution objects that support marginalization, sampling, and custom probability bases.
• Extensible architecture with optional support for JAX, PyTorch, and other optimization backends.