Author: Robert Hyatt
Date: 06:14:51 05/24/00
Go up one level in this thread
On May 24, 2000 at 02:25:40, Peter Kappler wrote: >On May 24, 2000 at 00:17:35, Robert Hyatt wrote: > >> >>The problem is that the only "old" evidence was self-play with varied depth. >>Which is probably not a good test, since small changes are often magnified in >>self-test play, while at other times small changes have no effect at all. >> > >This part confuses me. Why would self-play be inferior to a test suite? I >don't understand what you mean by "small-changes" above... The problem is this: If you add one small piece of information to one program, and that is the _only_ difference in the two programs, then you will see that piece of information influence the games _frequently_. Because one uses it and the other has no idea about it. > >I always thought the "go deep" experiments were testsuite-based because of the >enormous time reuiqred to play a meaningful number of games with 15-16 ply >searches. Not true? No. The "go deep" experiments were positions from real games (in fact, they were every 4th move from the first deep blue vs kasparov match.) We were only interested in how often a deeper ply produced a different move to play otb. > >--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.