- Created Wednesday, August 10th 2016 @ 17:37:20
I'm still pretty new here and I've been enjoying the site very much! I created a four in a row bot for my first attempt, and I've really enjoyed the process of discovering new strategies and patching in defenses for various attacks.
My bot (Chinless) has been doing about as well as I have expected for him playing a largely reactionary game. But as he's gaining elo, he's starting to face more and more opponents who just simply outplay him with deeper thinking.
I'm new to game AI and have seen lots of posts about minimax, alpha beta pruning, monte carlo tree searches, etc. I'm trying to learn as much as I can so I can build better bots, but it kind of had me thinking, if everyone is using the same algorithms, it seems like things would start getting pretty boring pretty quickly.
So I have some questions. The first is, are pretty much all the top bots in all the games using known algorithms like the ones I mentioned above? Is it pretty much mandatory to implement stuff like that to be competitive? Are there any games on the site in which most known algorithms wouldn't work or be applicable? If everyone IS using the same known processes, what distinguishes the top bots from each other? Optimizations for faster play, or what?
- Created Thursday, August 11th 2016 @ 03:43:56
In perfect information games with a branching factor that's not too huge, like four in a row, you won't be able to compete with the best without some kind of search algorithm. In these kind of competitions the bots are distinguished by how optimized this search is and how they evaluate non-terminal positions.
- Updated Sunday, August 14th 2016 @ 19:49:46
Yeah, I also don't like when the bots distinguish themself on how smart they can brute force through the search tree (or throw away non pomising paths). However you can still have some fun when it comes to the evaluation function since if you have knowledge about a certain game, you can use your domain knowledge to gain an edge over the other bots. As for myself I have little idea about how to evaluate Go positions, so even if I brute force more effectively through the search tree, a bot of a solid player should easily beat my bot.
As for the WarLight AI challenges, pretty much no bots use some sort of minimax (apart from the picking stage), since this would be way to time consuming and would probably still give very bad results. Imo those were the best competitions so far, but I may be biased ;)