neuralRaven 1.1.0

Library of neural network with a evolutive algorithm of autolearned

NeuralRaven — Evolutionary Neural Network Library

NeuralRaven is a Python library that implements a fully customizable neural network architecture capable of training itself using an evolutionary algorithm (genetic optimization).
Unlike traditional machine-learning libraries, neuralRaven does not use gradient descent, backpropagation, or optimizers.
Instead, it evolves network weights over generations.

This makes RAVEN suitable for:

• Environments where gradient-based training performs poorly.
• Reinforcement learning tasks.
• Black-box optimization problems.
• Scenarios with non-differentiable fitness functions.

---

✨ Features

• Fully implemented neural network written in Python.
• Genetic/Evolutionary algorithm for self-training.
• Support for mutation, crossover, elitism, and population-based evolution.
• Save/Load functionality for evolved models.
• Small footprint and easy to integrate.
• Cross-platform wheel support (Windows, Linux, macOS).
• Compatible with Python 3.8–3.14.

---

📦 Installation

Once the package is published on PyPI:

pip install neuralRaven

If installing from source:

pip install .

🚀 Quick Example

# main.py

import neuralRaven

def main():
    print("=== Executing in neuralRaven.IO.fromFiles() ===")

    # Name files
    config_file = "configurationfile"
    data_file = "datatest"

    try:
        neuralRaven.IO.fromFiles(config_file, data_file)
        print("\n=== Execution complete without errors ===")
    except Exception as e:
        print("\n*** Error in the execution ***")
        print(e)

if __name__ == "__main__":
    main()

📁 Project Structure

raven/
    __init__.py
    neural_network.py
    evolutionary.py
test/
    test_basic.py
data/
    example_input.json
    example_output.json

🛠 Building Wheels (for PyPI)

This project uses cibuildwheel to build platform-specific wheels.

GitHub Actions workflows:

.github/workflows/build_wheels.yml — builds wheels for Windows/Linux/macOS.

.github/workflows/publish.yml — publishes wheels to PyPI.

📄 License

This project is released under the MIT License — see the LICENSE file for details.

👥 Authors

Carlos Gerardo Euresty Uribe
Lead author and primary developer of the neural network architecture and evolutionary algorithm.
Email: [e-mail]

Jesús Miguel Cerda González
Contributor responsible for packaging the project as a Python library, optimizing modules, and improving usability.
Email: [e-mail]