Author: Dave Gomboc
Date: 07:34:19 01/28/02
Go up one level in this thread
On January 28, 2002 at 08:15:04, Ed Schröder wrote: >On January 28, 2002 at 07:42:39, Amir Ban wrote: > >> >>A white knight on a8/h8 is weak and often in trouble, and there's nothing wrong >>with factoring that in the evaluation function. The fact that search will often >>find this on its own is immaterial. >> >>Look at CCT4 ZarkovX vs. DJ for a nice demonstration. >> >>Amir > >There is nothing wrong having this kind of knowledge in your eval. The BIG >question is: IS IT NEEDED as search can do that job for you too? If a deep >search depth guarantees you to deal with the Na8/Nh8 cases why do you want to do >things twice and spoil valuable processor time? > >Ed What if you didn't have such terms in there? - various leaf nodes would be (more ;-) misassessed - the tree shape could change - your score windows could be different - your move ordering could be worse - your fail-highs/lows might take longer to resolve The better the guesses your evaluation function makes are, the more efficient your search will be, right? Your hash entries will have more accurate and useful data more often. You're likely to see the tactic sooner because your search tree is already pointing in the right direction. (In pure alpha-beta one would be guaranteed to find the tactic at depth x anyway, but everyone is pruning non-critical paths like crazy, so you're less likely to pick up the tactic if it's happening in a side line, or perhaps I should just say that it would take more iterations to do so.) (There's a lot of "could/might/should" stuff there because not having written a chess program, I haven't measured these things myself! These are things that I generally believe. I am certainly interested in being corrected by someone who disagrees, though!) I have seen the term "speculative" eval, but to me this is a bit of a misnomer. Was Tal speculating when he sacrificed a piece for every type of compensation imaginable? Well, maybe! ;-) But there's another way to look at it, which is to say that hey, the position is still good, so what if one has given up a piece for a pawn, that's not the most relevant factor. Why would I guess that the position is about -2, when some of the opponent's pieces are out of play and the king is under serious pressure? Surely, assigning such positions scores of roughly +1 instead must to lead to a serious improvement in play? Frankly, the only speculation that might be going on would be that the other program is scoring itself at +2 because it thinks it's going to beat off a strong initiative without giving back material. Dave
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.