https://github.com/codereclaimers/neat-python
Python implementation of the NEAT neuroevolution algorithm
https://github.com/codereclaimers/neat-python
neuroevolution python
Last synced: 23 days ago
JSON representation
Python implementation of the NEAT neuroevolution algorithm
- Host: GitHub
- URL: https://github.com/codereclaimers/neat-python
- Owner: CodeReclaimers
- License: bsd-3-clause
- Created: 2015-09-26T22:59:53.000Z (over 10 years ago)
- Default Branch: master
- Last Pushed: 2026-03-02T17:16:59.000Z (28 days ago)
- Last Synced: 2026-03-02T19:50:21.376Z (28 days ago)
- Topics: neuroevolution, python
- Language: Python
- Homepage:
- Size: 2.53 MB
- Stars: 1,541
- Watchers: 68
- Forks: 509
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
- Changelog: CHANGELOG.md
- Funding: .github/FUNDING.yml
- License: LICENSE
- Code of conduct: CODE_OF_CONDUCT.md
- Citation: CITATION.cff
Awesome Lists containing this project
README
[](https://github.com/CodeReclaimers/neat-python/actions/workflows/tests.yml)
[](http://neat-python.readthedocs.io)
[](https://coveralls.io/github/CodeReclaimers/neat-python?branch=master)
[](https://pepy.tech/project/neat-python)
## About ##
NEAT (NeuroEvolution of Augmenting Topologies) is a method developed by Kenneth O. Stanley for evolving arbitrary neural
networks. This project is a pure-Python implementation of NEAT with no dependencies beyond the standard library. It was
forked from the excellent project by @MattKallada.
For further information regarding general concepts and theory, please see the [publications page](https://www.kenstanley.net/papers) of Stanley's current website.
`neat-python` is licensed under the [3-clause BSD license](https://opensource.org/licenses/BSD-3-Clause). It is
currently only supported on Python 3.8 through 3.14, and pypy3.
## What's New in 2.0 ##
The CTRNN (Continuous-Time Recurrent Neural Network) implementation now supports **per-node evolvable time constants**. In v1.x, all nodes shared a single fixed time constant passed at network creation time. In v2.0, each node carries its own time constant as an evolved gene attribute, allowing the network to operate across multiple timescales simultaneously.
This is a breaking API change: `CTRNN.create(genome, config, time_constant)` is now `CTRNN.create(genome, config)`. Existing feedforward and discrete-time recurrent configurations require no changes.
For details on the change, its motivation, quantitative impact, and migration guide, see [CTRNN-CHANGES.pdf](examples/lorenz-ctrnn/docs/CTRNN-CHANGES.pdf).
## Features ##
* Pure Python implementation with no dependencies beyond the standard library
* Supports Python 3.8-3.14 and PyPy 3
* Reproducible evolution - Set random seeds for deterministic, repeatable experiments
* Parallel fitness evaluation using multiprocessing
* Network export to JSON format for interoperability
* Comprehensive documentation and examples
## Getting Started ##
If you want to try neat-python, please check out the repository, start playing with the examples (`examples/xor` is
a good place to start) and then try creating your own experiment.
The documentation is available on [Read The Docs](http://neat-python.readthedocs.io).
You can also ask questions via the [experimental support agent](https://neat-python.recursive.support)!
## Network Export ##
neat-python supports exporting trained networks to a JSON format that is framework-agnostic and human-readable. This allows you to:
- Convert networks to other formats (ONNX, TensorFlow, PyTorch, etc.) using third-party tools (the beginnings of a conversion system can be found in the `examples/export` directory)
- Inspect and debug network structure
- Share networks across platforms and languages
- Archive trained networks independently of neat-python
Example:
```python
import neat
from neat.export import export_network_json
# After training...
winner_net = neat.nn.FeedForwardNetwork.create(winner, config)
# Export to JSON
export_network_json(
winner_net,
filepath='my_network.json',
metadata={'fitness': winner.fitness, 'generation': 42}
)
```
See [`docs/network-json-format.md`](docs/network-json-format.md) for complete format documentation and guidance for creating converters to other frameworks.
## Citing ##
Here are APA and Bibtex entries you can use to cite this project in a publication. The listed authors are the originators
and/or maintainers of all iterations of the project up to this point. If you have contributed and would like your name added
to the citation, please submit an issue.
APA
```
McIntyre, A., Kallada, M., Miguel, C. G., Feher de Silva, C., & Netto, M. L. neat-python [Computer software]
```
Bibtex
```
@software{McIntyre_neat-python,
author = {McIntyre, Alan and Kallada, Matt and Miguel, Cesar G. and Feher de Silva, Carolina and Netto, Marcio Lobo},
title = {{neat-python}}
}
```
## Thank you! ##
Many thanks to the folks who have [cited this repository](https://scholar.google.com/scholar?start=0&hl=en&as_sdt=5,34&sciodt=0,34&cites=15315010889003730796&scipsc=) in their own work.