Author: Peter Fendrich
Date: 15:02:14 06/02/98
Go up one level in this thread
On June 02, 1998 at 17:29:21, Stuart Cracraft wrote: >On June 02, 1998 at 15:24:35, Robert Hyatt wrote: > >> >>my position file has 65536 entries, and is "wrap-around" once it fills >>up in a FIFO ring buffer. However, I *never* let it fill up, because I >>delete it regularly. >> > >Why do you delete it regularly? Seems to me there's valuable >"experience" >in that file. Why not add a field "frequency" that shows how frequently >in >real games the given position has been visited. Then, when the file >fills >up and becomes a burden to the regular hash table, clean out the ones >with >frequency=some_low_number? > >> >>this sort of learning was added so that if someone finds a quick exit >>out >>of book, book learning won't work. This at least gives it a chance to >>not >>play into the same lost game over and over. I only learn positions from >>the first 10 non-book moves before turning it off... > >Someone forcing the program to play the same lost game over and over >whenever the same opening moves are played is a significant problem >especially when having a program play on the Internet Chess Server or >in a computer vs. computer tournament. I've seen it a few times and am >sure you've seen it quite a bit. Kinda negates the value of the server >for testing research if some joe knows how to get free rating points >without really beating the program each time. > >--Stuart The "book learning" prevents a program from losing the same game over and over. The "position-based" learning may become irrellevant after a while. The program gets more knowledge built in by development and the scores stored as permanent aren't permanent any more... //Peter
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.