POD RACING
Genetic algorithm for pod racing game, ranking 242nd in the legend league.
Followed Magus' blog for the basis of my algorithm.
Developed in python 2.7 on Ubuntu.
Post Mortem
- Eugenics
- Elitism (keep best pod in every iteration, so that there is no backward progress).
- Breed smartly (bred the individual with the best runner score with the individual with the best bruiser score)
- Don't change too much in a single mutation
- The more you change at once, the less likely you are to find improvement (but the potential for improvement is larger).
- Making smaller mutations creates more consistent (but more incremental) improvement, especially important when there isn't enough time to run long simulations.
- Start with a good base population (use basic logic to create a decent starting place instead of random if possible)
- Potential downside if base population is close to ok solutions, but far from the optimal solutions
- A good simulation is necessary