Hacker News new | comments | ask | show | jobs | submit login

You may need to consider a better way to rank players - percentage of games won is problematic (as was discovered during the Tron Google AI challenge) and something like TrueSkill or bayeselo would be more robust.

Yeah, we're seeing the first problems with that right now- newly uploaded bots progress faster than they should because they play against a more "diluted" pool of opponents than the oldtimers did. We'll think of something.

I also think your alpha-beta bot is objectively better than the (currently) #1 bot.

Easiest thing to throw together now is a script which just feeds the whole game history into bayeselo as a bunch of fake chess matches, with the white move advantage set to 0. That's what was done for Tron.


Make every bot play each other once or multiple times.

Applications are open for YC Summer 2019

Guidelines | FAQ | Support | API | Security | Lists | Bookmarklet | Legal | Apply to YC | Contact