Author: Robert Hyatt
Date: 07:41:39 02/11/02
Go up one level in this thread
On February 11, 2002 at 01:49:54, Uri Blass wrote: >On February 10, 2002 at 23:40:06, Robert Hyatt wrote: > >>On February 09, 2002 at 22:49:22, Russell Reagan wrote: >> >>>On February 09, 2002 at 22:35:54, Robert Hyatt wrote: >>> >>>>Several issues: >>>> >>>>1. 6 piece files are going to require some large mate-in-N scores. IE they >>>>will need 16 bits at least, which doubles the size compared to a current 5 >>>>piece file where all scores are 8 bits max. >>> >>>Why would you need to use a full 16-bits? And why would you add on "at least" to >>>that? How many mates are going to take more moves than 65,536 moves without >>>reaching a 50-move rule point or 3-fold repitition? The answer is none. Maybe >>>you could explain why my reasoning doesn't work. >>> >> >> >>Simply because the _current_ way of building them uses distance to mate. >>And distance to mate is distance to mate, period... and it will definitely >>push beyond 8 bits, and using 9 bits is _not_ convenient for reasonable speed >>and accessing. > >What happens if programs use only 8 bits and decide that it is a draw when more >than 8 bits are needed? > >I suspect that it is better for chess programs because most long mates are draws >by the 50 move rule. With 6 pieces, I could easily see mates in beyond 150 moves. Say in KQ vs KNNP. 50 moves to win the pawn, 50 to win a knight, another 30 to win the last knight, then another few to mate. I think that the majority of the "deep mates" will probably work out ok, and if the tables are ever built using DTM and DTC, then things will change again. But at the moment, the tables are just absolute, assuming that a mate in 130 is a mate in 130 without regard for the 50 move rule, which is ok for the present. > >Is there statistics for an estimate for the probability for a draw when there is >a long mate? > >It is impossible to know because programs that use tablebases do not play >perfect when you consider the 50 move rule but it is possible to have some >estimate. > >If you want to calculate the probability that mate in 70 is winning then you can >take 2 programs that use tablebases and choose a lot of random tablebases >positions that are evaluated as mate in 70. > >You can get an estimate for the probnability to win by the proportion of games >that are won. > >You can do the same for mate in 71,72.... > >Uri
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.