Author: Stuart Cracraft
Date: 19:44:35 08/29/04
Go up one level in this thread
On August 29, 2004 at 11:39:18, Dan Honeycutt wrote: >On August 29, 2004 at 01:10:47, Stuart Cracraft wrote: > >>On August 28, 2004 at 11:03:44, Dan Honeycutt wrote: >> >>>On August 27, 2004 at 22:44:47, Stuart Cracraft wrote: >>> >>>> >>>>Computer vs. computer and especially computer against a version of itself >>>>is well-known for exaggerating the relative difference. >>>> >>> >>>Hi Stuart: >>>Could this explain that the stronger version could lose to the weaker version >>>because their play is so nearly identical? >>> >>> >>>>Why not setup your program for internet chess server chess and have it >>>>play there? >>>> >>> >>>I'm trying to do an A-B comparison where A and B are identical except for X to >>>see if X is an improvement. I'd have to let both play on the server long enough >>>to get a reliable rating which would seem slower than a controlled test - though >>>I would get a lot more variety of opponents. >> >>Use brute force: >> >> 1) run a very large test suite >> 2) do self play a very large number of games with slight randomization >> 3) run it for a large number of games on ICS >> >>It's a lot of work but you want to know whether X wins or not and the >>future of your program depends on it. >> >>Stuart > >Hi Stuart: >You are exactly right. Problem is, so many X's, so little time. I'm trying to >find a way to judge X in a reasonable amount of time. It's not easy - if it >were everybody's program would play like Junior or Shredder. > >Dan H. Dan -- can you consider some kind of mathematical/engineering/biological approach to balance all your terms? Have you done any auto-tuning? I had to write this entire 4 month 4000 line chess program just to get something going with auto-tuning and got kind of bogged-down in the implementation but it is all pretty much together for the next step. I want to auto-tune because a) I am a weak player and can't possibly balance the terms myself b) I don't have the time to balance/test/rebalance and retest for decades and c) it is more fun to watch the work being done than to do it myself. Stuart
This page took 0 seconds to execute
Last modified: Thu, 15 Apr 21 08:11:13 -0700
Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.