Author: Robert Hyatt
Date: 11:53:07 12/20/01
Go up one level in this thread
On December 19, 2001 at 23:00:52, Shane Hudson wrote: >Hi all, > >I don't know if anyone else here is interested, but if you want to see a >table showing how often each tablebase material configuration is reached >in practise (with statistics from a large database of master-level games), >I generated the results for a database of over 500,000 games and they >are available at the URL > >http://scid.sourceforge.net/tbstats.html > >You may find this helpful if you only have space or time to obtain a >limited number of tablebases -- but there is no guarantee that the endgames >most often reached in human play are also the endgames most useful to >a chess engine in deep searches, of course. > >It would be interesting to see if these correlate to the actual frequency >of use of each tablebase in a chess engine, but I don't think anyone has >hacked their tablebase-probing code to collect such stats yet. > >Cheers, >Shane One _important_ piece of data this misses is "how often are they important?" IE you are counting games that actually reach 6-5-4-3 pieces _on_ the board. But what about those 12 piece games where the program reaches an EGTB position deep in the search and discovers that it must avoid that position because it is a draw? IE the EGTBs are important _way_ before the game actually gets to 5 pieces or less. By that point the result is already clear. But when there are 16 pieces (or sometimes more) left, the "hits" begin to provide important information that may actually be far more important than the information actually found when the game _really_ reaches 6 or fewer pieces...
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.