Author: Tony Werten
Date: 05:53:49 02/08/01
Go up one level in this thread
On February 08, 2001 at 08:27:21, martin fierz wrote: >On February 08, 2001 at 06:30:14, Tony Werten wrote: > >>On February 08, 2001 at 04:24:11, David Blackman wrote: >> >>>On February 07, 2001 at 16:41:28, Tanya Deborah wrote: >>> >>>> >>>> >>>>Hi! >>>> >>>>I am playing a new match in checkers between the 2 strongest Spanish checkers >>>>programs of the world... >>> >>>Just curious, is "Spanish checkers" the same game as "Polish Draughts", >>>"International Draughts", "Damen" etc? >> >>She could have meant 2 spanish programs playing polish checkers :) >> >>I think it's true for the other coutries as well, but the game played in Holland >>is international draughts. >>> >>>http://www.multimania.com/nic55/dames/dames2.htm >>> >>>They say it is played in >>>"most French-speaking countries (France, Belgium, Switzerland, Canada, and the >>>African continent) and also in the Netherlands, and in the ex-soviet union >>>countries." >> >>If I remember correctly, in the SU they let the children start with checkers to >>get some ideas about the tactics, before switching to draughts. >> >>> >>>This is the game on the 10x10 board. >>> >>>According to people who have tried, it is a bit harder to write a strong program >>>for it than for chess. Perhaps it should be the next big board-game programming >>>challenge, now that chess programs are more or less in reach of the top human >>>players, and Go still seems much too hard. >> >>Depends on what you call difficult. In checkers ( and draught ) there seems to >>be no additional strength from searching deeper anymore, only from better >>evaluation. Which is a bit harder to achieve. It also means that a faster >>computer doesn't give you a stronger computer automaticly. > >this is not true. i performed some self-play experiments with my computer >checkers program, i used a version searching to a fixed depth n to play another >version searching to depth n-2, in 282 game-matches with the results > >depth win% >5-3 78.9% (+196=53-33) >7-5 72.0% (+153=100-29) >9-7 77.5% (+181=75-26) >11-9 65.8% (+130=111-41) >13-11 68.1% (+134=116-32) >15-13 66.3% (+119=136-27) >17-15 60.8% (+89=165-28) >19-17 58.9% (+78=176-28) >21-19 54.8% (+60=189-33) Off course it depends on the strength of the program. The weaker a program, the more it gains from additional search. Second, this is self play. The 21 ply version sees everything the 19 ply does plus 2 ply more. If you then still only score 54.8 % you gained very little from this additional search. ( Ok, not nothing but close to it) cheers, Tony > >yes, these are diminishing returns for deeper searches, but no, it is not true >that searching deeper doesnt help. i also tried taking out some very important >knowledge from the eval of the deeper searching program, expecting the shallower >program to beat it from some search depth on - which was not the case. the >result for the deeper program did get worse, but it always won. the statistical >errors on the above numbers are about 2%, so 78.9+-2.0%. >there was a chinook paper once, which reported diminishing returns too, but that >only had 20-game matches i think and huge error margins - too large to claim >anything. > >cheers > martin
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.