Author: Robert Hyatt
Date: 21:32:20 12/21/99
Go up one level in this thread
On December 22, 1999 at 00:30:05, Robert Hyatt wrote: >On December 21, 1999 at 23:46:06, Greg Lindahl wrote: > >>>>>>How important is forcing in shallow plies verses deeper plies? That's easy >>>>>>to examine using a program. >> >>I'd still like to see an answer to this question -- why debate how much of an >>effect DB's algorithms are in the last 4 plies when you can do a few >>measurements? Isn't computer chess at least partially an experimental science? >> >>-- g > > >This doesn't seem like an experiment worth testing. In chess, we have a >root position, and we search to a terminal position where we apply an eval >and then back up a score. I don't see anything that would suggest that it >is better (or worse) to extend near the root or near the tips. Except that >extending near the tips causes the tree to grow quicker. But nothing >suggests that you can't get from the root position to the right tip position >by extending early, vs only by extending late. The issue is extending _right_. > >Which means that program A might do better with one approach, while B might >do better with the other. The chess program is the sum of its parts. And >all this comparing with DB is nonsense... I should add that I ran a huge experiment last year that is sort of connected to this. The question was "how to limit extensions?". I came to the conclusion, after a lot of testing, that the deeper I went, the more I wanted to throttle the extensions, to avoid tree explosions. I do all my normal stuff up to 2*iteration_depth (for 12 ply search, this means all extensions are done fully up to ply=24). Beyond that, I reduce the amount of each extension... And after a _lot_ of testing, I concluded that this fit "my" program best...
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.