David Fagan

An analysis of genotype-phenotype maps in grammatical evolution

We present an analysis of the genotype-phenotype map in Grammatical Evolution (GE). The standard map adopted in GE is a depth-first expansion of the non-terminal symbols during the derivation sequence. Earlier studies have indicated that allowing the path of the expansion to be under the guidance of evolution as opposed to a deterministic process produced significant performance gains on all of the benchmark problems analysed. In this study we extend this analysis to include a breadth-first and random map, investigate additional benchmark problems, and take into consideration the implications of recent results on alternative grammar representations with this new evidence. We conclude that it is possible to improve the performance of grammar-based Genetic Programming by the manner in which a genotype-phenotype map is performed.

A Grammar Design Pattern for Arbitrary Program Synthesis Problems in Genetic Programming

Grammar Guided Genetic Programming has been applied to many problem domains. It is well suited to tackle program synthesis, as it has the capability to evolve code in arbitrary languages. Nevertheless, grammars designed to evolve code have always been tailored to specific problems resulting in bespoke grammars, which makes them difficult to reuse. In this study a more general approach to grammar design in the program synthesis domain is presented. The approach undertaken is to create a grammar for each data type of a language and combine these grammars for the problem at hand, without having to tailor a grammar for every single problem. The approach can be applied to arbitrary problem instances of program synthesis and can be used with any programming language. The approach is also extensible to use libraries available in a given language. The grammars presented can be applied to any grammar-based Genetic Programming approach and make it easy for researches to rerun experiments or test new problems. The approach is tested on a suite of benchmark problems and compared to PushGP, as it is the only GP system that has presented results on a wide range of benchmark problems. The object of this study is to match or outperform PushGP on these problems without tuning grammars to solve each specific problem.

PonyGE2: grammatical evolution in Python

Grammatical Evolution (GE) is a population-based evolutionary algorithm, where a formal grammar is used in the genotype to phenotype mapping process. PonyGE2 is an open source implementation of GE in Python, developed at UCDs Natural Computing Research and Applications group. It is intended as an advertisement and a starting-point for those new to GE, a reference for students and researchers, a rapid-prototyping medium for our own experiments, and a Python workout. As well as providing the characteristic genotype to phenotype mapping of GE, a search algorithm engine is also provided. A number of sample problems and tutorials on how to use and adapt PonyGE2 have been developed.

Comparing the performance of the evolvable πGrammatical Evolution genotype-phenotype map to Grammatical Evolution in the dynamic Ms. Pac-Man environment

In this work, we examine the capabilities of two forms of mappings by means of Grammatical Evolution (GE) to successfully generate controllers by combining high-level functions in a dynamic environment. In this work we adopted the Ms. Pac-Man game as a benchmark test bed. We show that the standard GE mapping and Position Independent GE (πGE) mapping achieve similar performance in terms of maximising the score. We also show that the controllers produced by both approaches have an overall better performance in terms of maximising the score compared to a hand-coded agent. There are, however, significant differences in the controllers produced by these two approaches: standard GE produces more controllers with invalid code, whereas the opposite is seen with πGE.

Exploring position independent initialisation in grammatical evolution

Initialisation in Grammatical Evolution (GE) is a topic that remains open to debate on many fronts. The literature falls between two mainstay approaches: random and sensible initialisation. These methods are not without their drawbacks with the type of trees generated. This paper tackles this problem by extending these traditional operators to incorporate position independence in the initialisation process in GE. This new approach to initialisation is shown to provide a viable alternative to the commonly used approaches, whilst avoiding the common pitfalls of traditional approaches to initialisation.

Towards adaptive mutation in grammatical evolution

Adaptive mutation operations have been proposed in Evolutionary Computation (EC) many times and in different varieties, but few have gained widespread use. In nature, mutation rates vary over time, however it has become common practice to use static, widely accepted, values for mutation, particularly in GP-like systems. In this study, an adaptive mutation operation is presented and applied to Grammatical Evolution (GE) over a variety of benchmark problems. The results are examined and it is determined that the new operators could replace the need to specify mutation rates in GE on the problem domains examined.

Introducing Semantic-Clustering Selection in Grammatical Evolution

Semantics has gained much attention in the last few years and new advanced crossover and mutation operations have been created which use semantic information to improve the quality and generalisability of individuals in genetic programming. In this paper we present a new selection operator in grammatical evolution which uses semantic information of individuals instead of just the fitness value. The semantic traits of an individual are stored in a vector. An unsupervised learning technique is used to cluster individuals based on their semantic vector. Individuals are only allowed to reproduce with individuals from the same cluster to preserve semantic locality and intensify the search in a certain semantic area. At the same time, multiple semantic areas are covered by the search as there exist multiple clusters which cover different areas and therefore preserve semantic diversity. This new selection operator is tested on several symbolic regression benchmark problems and compared to grammatical evolution with tournament selection to analyse its performance.

Understanding expansion order and phenotypic connectivity in πGE

Since its inception, πGE has used evolution to guide the order of how to construct derivation trees. It was hypothesised that this would allow evolution to adjust the order of expansion during the run and thus help with search. This research aims to identify if a specific order is reachable, how reachable it may be, and goes on to investigate what happens to the expansion order during a πGE run. It is concluded that within πGE we do not evolve towards a specific order but a rather distribution of orders. The added complexity that an evolvable order gives πGE can make it difficult to understand how it can effectively search, by examining the connectivity of the phenotypic landscape it is hoped to understand this. It is concluded that the addition of an evolvable derivation tree expansion order makes the phenotypic landscape associated with πGE very densely connected, with solutions now linked via a single mutation event that were not previously connected.

Investigation of the performance of different mapping orders for GE on the max problem

We present an analysis of how the genotype-phenotype map in Grammatical Evolution (GE) can effect performance on the Max Problem. Earlier studies have demonstrated a performance decrease for Position Independent Grammatical Evolution (πGE) in this problem domain. In πGE the genotype-phenotype map is changed so that the evolutionary algorithm controls not only what the next expansion will be but also the choice of what position in the derivation tree is expanded next. In this study we extend previous work and investigate whether the ability to change the order of expansion is responsible for the performance decrease or if the problem is simply that a certain order of expansion in the genotype-phenotype map is responsible. We conclude that the reduction of performance in the Max problem domain by πGE is rooted in the way the genotype-phenotype map and the genetic operators used with this mapping interact.

Dynamic ant: introducing a new benchmark for genetic programming in dynamic environments

In this paper we present a new variant of the Ant Problem in the Dynamic Problem Domain. This approach presents a functional dynamism to the problem landscape, where by the behaviour of the ant is driven by its ability to explore the search space being constrained. This restriction is designed in such a way as to ensure that no generalised solution to the problem is possible, thus providing a functional change in behaviour.