Author: Robert Hyatt
Date: 20:58:10 07/13/99
Go up one level in this thread
On July 13, 1999 at 07:42:28, Shep wrote: >On July 13, 1999 at 07:14:32, Terry Ripple wrote: > >>On July 13, 1999 at 07:03:27, Shep wrote: >> >>>On July 13, 1999 at 05:10:41, Didzis Cirulis wrote: >>> >>>>Hi, >>>> >>>>When playing on FICS (Computer(c)) Chess Tal lost a completely won game due to >>>>some mismanagement of time. It did run out of time. So how does this program >>>>learn in this situation? Will this line be considered good or bad next game? >>>> >>>>One more question: What happens if I start a new game when I am going to lose >>>>this one? Is program learning at all in this situation? Is this learning >>>>correct? >>> >>>Ed wrote that at least for Rebel, learning works OK in that case, even if the >>>game does not end by checkmate or resigning on your part. >>>IIRC Crafty also handles learning depending on the evaluation only, not "3rd >>>party effects" like running out of time. >>>This is reasonable; imagine the opposite situation where you have a won position >>>against the program and then lose on time or resign by accident... >>> >>>Another "out-of-time" learning problem is perceptible in the Fritz GUI: >>>If you play against another program (regardless if manually or eng-eng) and >>>allow the opponent's time to run out, the Fritz GUI will flag this game as won >>>in its book learning. >>>So you should always stop the opponent's clock with "Ctrl-Q" after the Fritz GUI >>>has moved. >>>And in eng-eng Blitz games, you should disable learning altogether because one >>>engine may lose on time in a won or drawn situation (seen this many times in >>>5-min Blitz). >>> >>>--- >>>Shep >>---------- >>Hi Shep, >> Are you saying to disable Book Learning and Position Learning or just one? > >I would only disable book learning as I suppose that position learning is done >during the game (thus based on the evaluations), not after the result of the >match has been determined. > >--- >Shep For crafty I would always leave position learning on. And book learning on. But I would turn result learning off if using one computer, because of the reasons already given. Normal book learning will probably be ok, as it happens over the first 10 moves out of book, which probably are done before any kind of time problem arises... Result learning is dangerous if games end with a flag falling...
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.