I'm using the Grammatical Evolution (GE) package in ECJ (ec.gp.ge.*) to evolve the control structure of an agent in a multi-agent system. This approach seems to be working fairly well. However, my choice for using GE was rather arbitrary and I would like to compare my current implementation using GE to other GGGP (Grammar Guided GP) solutions.
The reason that I want to conduct this comparison is because I'm in doubt if GE is the best representation for my problem. For example, in  GE is outperformed by GP on the Santa Fe Ant Trail problem. Because the Santa Fe Ant Trail problem is in the agent domain, I suspect that GP might be better for my agent problem as well.
There are two GGGP representations which I would like to use in my implementation, and therefore in ECJ. One is 'plain' grammar guided GP and the other is tree adjoining grammars (TAG). According to , one of the benefits of using TAGs is: "The TAG transformation permits local dependencies in the genotype space to map to long-distance dependencies in the intermediate phenotype space in a controlled way, corresponding to the structure of the grammar."
In short my questions are:
- Does the above reasoning for researching other grammar based GP representations make sense, or am I missing the point (or some recent literature)?
- Are there other GGGP implementations in ECJ besides GE, such as plain GGGP and TAG? If not, why?
If other GGGP implementations do not exist, I'm thinking of developing them as extensions to ECJ in the future.
1. OʼNeill, M. and Ryan, C. Grammatical Evolution. Evolutionary Computation 5, 4 (2001), 349-358.
2. McKay, R.I., Hoai, N.X., Whigham, P.A., Shan, Y., and O’Neill, M. Grammar-based Genetic Programming: a survey. Genetic Programming and Evolvable Machines 11, 3-4 (2010), 365-396.
Rinde van Lon
Disclaimer: http://www.kuleuven.be/cwis/email_disclaimer.htm for more information.