Push Genetic Programming in Python
Push Genetic Programming in Python
What is PushGP?
Push is programming language that plays nice with evolutionary computing / genetic programming. It is a stack-based language that features 1 stack per data type, including code. Programs are represented by lists of instructions, which modify the values on the stacks. Instructions are executed in order.
Why use PushGP?
PushGP is a leading software synthesis (sometimes called "programming by example") system. It utilized stochastic (typically evolutionary) search methods to produce programs that are capable of manipulating all the common data types, control structures, and data structures. It is easily extendable to specific use cases and has seen impressive human-competitive coding results. PushGP has discovered novel quantum computer programs previously unknown to human programers, and has achieved human competitive results in finding algebraic terms in the study of finite algebras.
In contrast to the majority of other ML/AI methods, PushGP does not require the transformation of data into numeric structures. PushGP does not optimize a set of numeric parameters using a gradient, but rather attempts to intelligently search the space of programs. The result is a system where the primary output is a program written in the Turing complete Push language.
PushGP has proven itself to be one of the most power "general program synthesis" frameworks. Like most evolutionary search frameworks, it usually requires an extremely high runtime, however it can solve problems that few other programming-by-example system can solve.
Additional references on the successes of PushGP:
- On the difficulty of benchmarking inductive program synthesis methods
- General Program Synthesis Benchmark Suite
- The Push3 execution stack and the evolution of control
Previous PushGP frameworks have focused on supporting genetic programming and software synthesis research. One of the leading PushGP projects is Clojush, which is written in Clojure and heavily focused on the experimentation needed to further the research field.
Pyshgp aims to bring PushGP to a wider range of users and use cases. Many popular ML/AI frameworks are written in Python, and with
pyshgp it is much easier to compare PushGP with other methods or build ML pipelines that contain PushGP and other models together.
Although PushGP is constantly changing through research and publication,
pyshgp is meant to be a slowly changing, more stable, PushGP framework. It is still possible to use
pyshgp for research and development, however accepted contributions to the main repository will be extensively benchmarked, tested, and documented.
pyshgp is compatible with python 3.5 and up.
Install from pip
pip install pyshgp
Build Frome source
- Clone the repo
- cd into the
pip install . --upgrade
- Thats it! Check out the examples and documentation.
Run the following command from project root directory. Make sure all the packages from
requirements-with-dev.txt are installed in the instance of python you are using.
python -m pytest
Example usages of
pyshgp can be found in the
examples/ folder of the Github repository.
pyshgp API can be found on official website.
Pysh Roadmap / Contributing
PyshGP isn't quite ready for its 1.0 release. It still has a few key features that need implementing. More details can be found in
ROADMAPS.md and in the projects tab on Github.
For information about contributing, see the Contributing Guide.
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.
|Filename, size||File type||Python version||Upload date||Hashes|
|Filename, size pyshgp-0.1.5-py3-none-any.whl (56.5 kB)||File type Wheel||Python version py3||Upload date||Hashes View hashes|
|Filename, size pyshgp-0.1.5.tar.gz (45.2 kB)||File type Source||Python version None||Upload date||Hashes View hashes|