Author: Robert Hyatt
Date: 13:08:00 01/24/00
Go up one level in this thread
On January 24, 2000 at 14:18:25, Ernst A. Heinz wrote: >>Berliner reported in a HiTech paper that when he tried testing at very shallow >>depths it broke his program because his eval assumed a certain basic search >>depth to find some simple tactics... and when he ran the "hitech vs lotech" >>tests to try to predict rating per ply increases, he saw this. I don't think >>it is possible to just re-do DB in a PC disguise. I think Hsu would start over >>and end up with something pretty similar to what everybody else has. Evolution >>has not brought us all to the same 'neighborhood' accidentally... > >Bob, > >The "Hitech vs. Lotech" experiment is totally inconclusive because >Berliner et al. only played 16 games for each iteration depth that >they considered. > >I have extensively analyzed all published self-play experiments in >computer chess, computer checkers, and computer Othello. None of >them present any conclusive, i.e., statistically confident evidence >for anything because nobody ever played enough games. > >For further details please see Chapter 9 of my book "Scalable Search >in Computer Chess" and my paper "Self-Play experiments in computer >chess revisited" which I presented at ACC-9/ACG-9, June 1999 in >Paderborn during the WCCC. > >Cheers, > >=Ernst= > >P.S. >"Scalable Search in Computer Chess" now at Amazon.de and MKP online. >http://www.amazon.de/exec/obidos/ASIN/3528057327/ >http://www.mkp.com/books_catalog/3-52805-732-7.asp :-) > >Please visit http://supertech.lcs.mit.edu/~heinz/node1.html for more >information about the book. I didn't mean to imply the result was important.. just that he recognized that his eval required some minimum search depth to function well, otherwise it would make mistakes. IE if you only search 2 plies deep, you need to evaluate forks statically or you get forked. Once you go deep enough, forks don't matter as the search picks them up and frees the eval from that responsibility. I think that was what he was getting at... An eval that behaves like this has some error rate that is known... pushing the endpoints far enough away from the root makes this error rate less harmful...
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.