Genetic Engine

A hybrid between strongly-typed (STGP) and grammar-guided genetic programming (GGGP).

About GeneticEngine

GeneticEngine is a Genetic Programming framework for single- and multi-objective optimization. GeneticEngine allows the user to provide domain knowledge about the shape of the solution (using type annotations) and by defining the fitness function.

Documentation

• Documentation
• Examples

class MyExpr(ABC):
	def eval(self):
		...

@dataclass
class Plus(MyExpr):
	left: MyExpr
	right: MyExpr

	def eval(self):
		return self.left.eval() + self.right.eval()

@dataclass
class Literal(MyExpr):
	value: int

	def eval(self):
		return self.value

In this small example, we are defining the language that supports the plus operator and integer literals. GeneticEngine will be able to automatically generate all possible expressions, such as Plus(left=Plus(left=Literal(12), right=Literal(12)), right=Literal(15)), and guide the search towards your goal (e.g., lambda x: abs(x-2022)). For this very simple toy problem, it will find an expression that computes 2022, ideally as small as possible. And this is a very uninteresting example. But if you introduce variables into the mix, you have a very powerful symbolic regression toolkit for arbitrariy complex expressions.

Contributing

After cloning the repo, please run source setup_dev.sh to install virtualenv, all dependencies and setup all pre-commit hooks.

Pull Requests are more than welcome!

Authors

GeneticEngine has been developed at LASIGE, University of Lisbon by:

• Alcides Fonseca
• Leon Ingelse
• Guilherme Espada
• Paulo Santos
• Pedro Barbosa
• Eduardo Madeira

Acknowledgements

This work was supported by Fundação para a Ciência e Tecnologia (FCT) through:

• the LASIGE Research Unit (ref. UIDB/00408/2020 and UIDP/00408/2020)
• Pedro Barbosa PhD fellowship (SFRH/BD/137062/2018)
• Guilherme Espada PhD fellowship (UI/BD/151179/2021)
• Paulo Santos CMU|Portugal PhD fellowship (SFRH/BD/151469/2021)
• the CMU|Portugal CAMELOT project (LISBOA-01-0247-FEDER-045915)
• the FCT Exploratory project RAP (EXPL/CCI-COM/1306/2021)
• the FCT Advanced Computing projects (CPCA/A1/395424/2021, CPCA/A1/5613/2020, CPCA/A2/6009/2020)

Publications

• Domain-Aware Feature Learning with Grammar-Guided Genetic Programming at EuroGP'23
• Data types as a more ergonomic frontend for Grammar-Guided Genetic Programming at GPCE'22
• Benchmarking Individual Representation in Grammar-Guided Genetic Programming at Evo*'22
• Grammatical Evolution Mapping for Semantically-Constrained Genetic Programming at GPTP'21
• The Usability Argument for Refinement Typed Genetic Programming at PPSN'20

Please cite as:

Espada, Guilherme, et al. "Data types as a more ergonomic frontend for Grammar-Guided Genetic Programming.", GPCE '22: Concepts and Experiences, 2022

Bibtex:

@inproceedings{espada2022data,
  author={Guilherme Espada and Leon Ingelse and Paulo Canelas and Pedro Barbosa and Alcides Fonseca},
  editor    = {Bernhard Scholz and Yukiyoshi Kameyama},
  title={Datatypes as a More Ergonomic Frontend for Grammar-Guided Genetic Programming},
  booktitle = {{GPCE} '22: Concepts and Experiences, Auckland, NZ, December 6 - 7, 2022},
  pages     = {1},
  publisher = {{ACM}},
  year      = {2022},
}