Decide on an initial set of problems

[NicMcPhee]

We need to decide on a set of test problems. These two papers should be useful:

• Better GP Benchmarks: Community Survey Results and Proposals
• Genetic Programming Needs Better Benchmarks

[NicMcPhee]

We discussed this and skimmed the "Better GP Benchmarks" article and definitely want to do:

• At least the four "function defined" regression problems they recommend (e.g., Pagie-1).
• Order tree
• NK-landscapes – we need to decide on values of N,K

These either are already implemented in ECJ or will be easy to implement.

We also might do the two recommended regression problems that are based on empirical data. I think these will also be easy to do in ECJ if we have access to the data sets (which I assume we do given that they're listed in the article).

There are other interesting recommended problems in "Better GP Benchmarks" but it's not clear how hard it would be to implement those in ECJ:

• Multiple output multiplier
• Classification (PSP, CHIRP)
• Planning and control problems (Mario gameplay, physical TSP) My guess is the first two should be pretty easy to implement, but the last might be tricky, especially if we'd need to implement a simulator.

We should also do the sine regression problem (since we already have data for that), and probably do the Santa Fe trail ant problem since we already have data and since it looks like crossover bias may hurt slightly on that problem.

[NicMcPhee]

I think I've gone through all these and turned the relevant ones into issues, so I'm going to close this.