Author: Peter Skinner
Date: 21:29:41 08/19/00
Go up one level in this thread
On August 19, 2000 at 19:26:32, Uri Blass wrote: >On August 19, 2000 at 12:35:39, Peter Skinner wrote: > >>On August 19, 2000 at 12:26:55, Jeroen van Dorp wrote: >> >>>Aren't both designed to play a hell of a lot better than me? >>> >>>Jeroen ;-} >> >>I beleive that Robert started making Crafty better for longer time controls, in >>the 17.xx versions. 16.19 was incredibly strong in blitz, and longer time >>controls. 17.11 is much better at longer games, than it is at blitz now. > >What are the results that you are based on when you say that 16.19 was better in >blitz. > >How many games? This was based on many games over ICS. Crafty 16.19 seems to do better vs the Rebel Tiger and Fritz/Junior clones. Where 17.11 is doing better in longer games vs these engines. I do not know exactly how many games, but over 300 I believe. Azkikr(C) on FICS in the bot I choose to test these findings. By no means is it a big hardware machine, only a PIII 450. > >I guess that bob did not decide only after 16.19 to think mainly about long time >control games. > >> >>I have found in the newer versions of Crafty since 17.5, that hash table have a >>dramatic influence on playing. I find that for blitz<=5 mins per side, that >>anything over 48M is useless, but also 48M is to much for games<=3 mins per >>side. If you play games anything >15 mins per side, I find that more is better. >> >>More isn't better in shorter games, this wasn't the case with version 16.19 to >>17.4. > >I am surprised to hear it because the only reason that I know that more hash is >worse in short time control is the fact that programs can lose on time but >crafty does not lose games on time. > >Uri This is what I have noticed in testing. I sometimes setup the program to only accept certain time controls, based on the hash settings I am using. The above settings are reflective of what I have found in games. I _never_ lose a game on time anyways, so that was never a concern. I am basing in on # of knps, and ply depth in these games. In shorter time control games, I have found that using a big hash table actually slows Crafty down and results in a 1-2 ply loss, where using more in game >5 mins per side, Crafty searches 1-2 ply deeper in all stages of the game.
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.