Post by steadyeddie on Jan 31, 2017 12:48:58 GMT -8
Jan 2017:
I tried writing a genetic algorithm to work out what was a good evaluation function. I had a population and trained them using the rollouts, with most wins surviving to the next generation.
Some modest success, but it turns out the nudging or plain tracking of wins for state changes is much better.
I tried writing a genetic algorithm to work out what was a good evaluation function. I had a population and trained them using the rollouts, with most wins surviving to the next generation.
Some modest success, but it turns out the nudging or plain tracking of wins for state changes is much better.