Author: Owen Lyne
Date: 05:42:27 12/17/99
Go up one level in this thread
On December 17, 1999 at 06:13:40, guy haworth wrote: >Yes, the stats are a little misleading because the game may be 'on the way >through' that profile. In fact, the games averaged only 7 moves in the >nominated profile - compared to theoretical depths now going up to 243 (KRNKNN). In fact the stats are very misleading for many reasons. For example, many one-sided endgames don't occur, because the losing side has resigned first, so the games that actually occur are biased towards the drawn ones (or at least, not so obviously won ones). Also, what matters to analysing a game accurately is all the lines that might have occured, but didn't. So a computer is probing the 5 main tablebases while there are many more than 5 pieces on the board and these database stats tell you nothing about that. And the typical human-human game doesn't not necessarily have the same endgame properties as a comp-comp or comp-human one. What would be needed would be a representative sample of say, Crafty's games, with all currently available TB's activated, with logging of all probes (which might well kill Crafty's perfomance in itself and make people doubt the validity of the experiment). Even with this information, you still don't know what really matters, because winning won positions a little faster is rather less interesting than getting a win instead of a draw from the tablebase hits. And tracing through the search to see the crucial information to win the game.... eek... Owen
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.