bridge-mixed-data-distance 1.0.0

# BRIDGE

**Bayesian Redundancy-Aware Information Distance with Graph Ensembles for Mixed-Type Data**

BRIDGE is a methodological framework for constructing redundancy-aware distances for mixed-type data while accounting for conditional predictive novelty and uncertainty in graph structure.

The framework separates mixed-variable commensurability from redundancy and propagates representation and graph uncertainty through a generalized Bayesian (Gibbs) ensemble of directed acyclic graphs.

## Main features

- Supports mixed numerical and categorical data.

- Explicitly addresses redundant and near-redundant variables.

- Separates marginal similarity from conditional predictive novelty.

- Incorporates graph-structure uncertainty.

- Uses generalized Bayesian / Gibbs weighting over DAG ensembles.

- Supports simulation-based validation and real-data evaluation.

- Designed for reproducible methodological research.

## Installation

### From PyPI

After the first public release, BRIDGE will be installable with:

pip install bridge-mixed-data-distance

### From source

Clone the repository and install it locally:

git clone https://github.com/USERNAME/BRIDGE-mixed-data-distance.git

cd BRIDGE-mixed-data-distance

pip install .

The dgaliatsatos2010 placeholder will be replaced with the final GitHub repository URL before release.

## Basic Python usage

The public Python API will be available through:

import bridge\_mixed\_data\_distance

Additional usage examples will be provided in the examples/ directory.

## Repository structure

BRIDGE-mixed-data-distance/

├── src/

│   └── bridge\_mixed\_data\_distance/

│       └── \_\_init\_\_.py

├── examples/

├── tests/

├── README.md

├── LICENSE

├── CITATION.cff

├── pyproject.toml

└── .gitignore

Additional reproducibility scripts, simulation experiments, and validation materials are included in the research release.

## Methodological scope

BRIDGE was designed to study mixed-type distance construction under several forms of redundancy, including:

- exact linear redundancy,

- near-linear redundancy,

- nonlinear redundancy,

- cross-type redundancy,

- interaction-based redundancy such as XOR structures.

The methodology also evaluates uncertainty arising from alternative graphical representations rather than relying on a single estimated graph.

## Important interpretation

BRIDGE is not presented as the first redundancy-aware distance for mixed-type data.

The contribution is the integration of:

1. mixed-type commensurability,

2. conditional predictive novelty,

3. redundancy-aware distance construction, and

4. generalized Bayesian graph-ensemble uncertainty.

The graph weights used by BRIDGE should be interpreted as generalized Bayesian / Gibbs weights over candidate DAGs rather than as a conventional likelihood-based Bayesian-network posterior for the complete joint distribution.

## Reproducibility

The research repository contains materials for:

- Monte Carlo simulation experiments,

- graph-ensemble analyses,

- real-data validation,

- ablation studies,

- sensitivity analysis,

- runtime evaluation,

- reproducibility auditing.

Random seeds and saved result files are used where appropriate to facilitate independent verification of reported findings.

## Citation

If you use BRIDGE in academic work, please cite the software release and the associated methodological paper.

Citation metadata are provided in:

CITATION.cff

A permanent Zenodo DOI will be added after the first public GitHub release is archived.

## License

BRIDGE is released under the MIT License.

## Author

**Dimitrios Galiatsatos**

Hellenic Open University, Patras, Greece

ORCID: 0009-0005-5232-1425

Email: galiatsatos.dimitrios@ac.eap.gr

## Version

Current development release:

1.0.0