Genetic Algorithm Class Design
By ai-depot | June 30, 2002
Final Thoughts
Discussion
Offline learning is of course possible, all it takes is a slightly adapted main loop that simply takes a new candidate, evaluates it, and reports it back.
However, the great thing about this approach to genetic algorithms it is now possible to concurrently evolve multiple components of independent agents in an elegant fashion. This was not impossible beforehand, but this approach makes it easier.
This online learning scheme raises the issue of having deterministic or non-deterministic trials. In the first case, the same trial is repeated over and over. When this is not convenient, the second case tries to evaluate the solution without creating a special context: it respects the state of the creature in the environment and admits it cannot change that without cheating. This is discussed in much more detail in the Advanced Obstacle Avoidance Tutorial.
Conclusion
All that�s left for you to do now is to fill in the blanks! Pick your favourite compiler, and experiment with approaches to Genetic Algorithms. Of course, you could just use a library, but that wouldn�t be fun would it?
Written by Alex J. Champandard.
Tags: none
Category: tutorial |
