reptarbot theDinobot
I hate bots with a passion. For trying to improve your game, I mean. Nothing wrong with a bot trying to improve though. It seems there was a recent boom in the go bot programming industry. I see a lot of new original bots popping up.
I've ran a bot on KGS before. It was called reptarbot theDinobot. It was just a modification of GNUGo. Lately I've been putting a lot of thought into the design of a bot, so I think I might just start writing one from scratch.
I'll probably start with a random bot, just so I can get the gtp (and kgs-gtp) protocol implemented properly. I'll probably keep the randomness and start off with a Monte Carlo implementation. They're very popular these days. Mogo and CrazyStone are based on MC and they seem to be performing better than the knowledge-based GNUGo. MC should be very simple to implement -- at least that's what I've read.
Once I get a reasonable bot, I'll probably write an interface for it, just to get the hang of Cocoa. They will be separate programs, so that the bot itself can be distributed to non-mac os x platforms. I probably won't use the gtp protocol between the bot and the interface, because I want to be able to see a little more than what the protocol allows.
Anyway, after that I'll develop multi-threading so it can take advantage of multi-core processors... or I might start off with a multi-threaded bot.
The main thing I want to do with the bot is to make it learn on its own from the rules I program in. I don't want to give hints, I want it to find patterns on its own. This is the part I'm putting the most thought into at the moment.
Should be fun. If I don't procrastinate, I expect the random bot to be up by next week.
No comments:
Post a Comment