- Created Saturday, May 21st 2016 @ 05:00:33
How good would a "voted" tetris bot be if we combined the top 25 bots and then asked each of them to return a move, then the move we choose to make is the move sequence that was returned by the majority of the bots.
- Created Saturday, May 21st 2016 @ 13:53:58
I think it would be worse than currently best bots. The problem is systematically incorrect decisions by some bots. These would override good decisions by such bots as hogeris. IMHO. Voting mechanism works ok, when you can expect statistical averaging. In this situation IMHO voting would not work.
- Created Saturday, May 21st 2016 @ 16:25:58
Furthermore, different bots have different strategies. Bots like hogeris, TaroKong and Kubisme play a lot of T-spins, Holol hardly does. Some bots are really good in removing garbage, others in gaining as much points as possible. If you try to average that, most of those strategies will work less optimal.
- Created Saturday, May 21st 2016 @ 18:22:01
Using the modal vote would avoid errors that are idiosyncratic to certain bots, but by the same token would also avoid awesome moves that are idiosyncratic to certain bots. I reckon the result would be better than the average of the bots sampled, but not as good as the best.
- Updated Sunday, May 22nd 2016 @ 07:23:18
Make linear combination of top engines decisions with some coeff ki. And learn this meta-bot by ki. I'm sure it leads to 100% for hogeris and 0% for others)