Author: Robert Hyatt
Date: 06:43:28 08/12/00
Go up one level in this thread
On August 12, 2000 at 02:59:01, Mark Young wrote: >On August 11, 2000 at 15:46:55, Robert Hyatt wrote: > >>On August 11, 2000 at 12:42:14, Vincent Diepeveen wrote: >> >>>On August 11, 2000 at 09:01:14, Robert Hyatt wrote: >>> >>>>On August 11, 2000 at 07:42:42, Vincent Diepeveen wrote: >>>> >>>>>That's not learning. crafty isn't using learning >>>>>in non-book positions. >>>>> >>>>>It's the killertables that speed it up then. >>>> >>>> >>>>Actually it does. It has a "position learning file" that I keep turned off >>>>in my test directory so that it doesn't produce odd results like this. >>>> >>>> >>> >>>Why is that silly thing in crafty anyway? >>>Now users can fool themselves! >>> >> >> >>It is there to prevent problems when the opponent plays 1. a3 and takes Crafty >>out of book. And then carefully plays the same game over and over until he >>finds a way to beat it. And then he repeats that won line over and over. With >>position learning, this is impossible to do. > >I for one like programs that have position learning, I wish all programs would >use them, just for the reason given. I know I cannot fool the program twice, and >as a consumer of chess programs I like this very much. It may just be human >nature, but when I know I can beat a chess program at will be exploiting this >weakness I lose interest in playing that chess program. > >The first chess program that I owned that used position learning was a tabletop >model the Fidelity 2265. I?m not sure who was the first to use position >learning, but I ?m still happy with my 2265 because of it. > > > I wish I had thought of this 30 years ago... In 1972, our Xerox customer engineer beat my program (named "blitz" at the time) 5-6 games in a row over several nights (standard time control games). When I got around to looking at the games, I discovered that he had played the _exact_ same game each time, and only varied if blitz varied due to a timing difference. I thought it a huge waste of time to spend 4 hours per game for 5-6 games, playing the _exact_ same moves each time. But it apparently salved his ego enough that he enjoyed this. flash forward to 1997. Chess servers. IM players. Same song, second verse. This time I had read a paper by Dave Slate and Tony Scherzer, so I implemented 'position learning' as they explained it and pretty well solved the problem. :)
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.