Author: Robert Hyatt
Date: 14:17:32 11/13/98
Go up one level in this thread
On November 13, 1998 at 11:30:55, Larry Coon wrote: >On November 12, 1998 at 21:33:55, Robert Hyatt wrote: > >>you can't really cut off... ie you get to choose to capture the opponent's >>piece, but then you *must* give him the opportunity to stop or else capture >>yours... and so forth... > >Exactly. As I described it, each side has the option of >stopping if it's their turn and they are ahead in the >exchange. It's exactly the way you coded it when you >work backward and compare the current score to the >previous score, and keep the best of the two. > >>you can stop early at the cost of a little accuracy... and you can even cut >>the whole algorithm to something pretty simple, and give up some more... > >I'm wondering how much a loss of accuracy here can really >cost you. After all, by the time you're doing a static >evaluation you're so focused on one set of exchanges that >you're ignoring any other tactical possibility. My guess >would be that the loss of accuracy from ignoring all the >other tactical possibilities probably overwhelms the loss >of accuracy from cutting off the SEE a little early, so >the loss from cutting off the SEE is insignificant. Has >that been your experience? The "loss" occurs because this is used for move ordering. If you don't "play out" all the captures, you might assume you can win a rook, or you might assume you win the exchange, and quit early. If you use this score for ordering captures, you might look at the wrong one first... > >>there aren't any branches... you *always* capture with the least valuable >>piece... so that simplifies this... and it only considers the target square, >>without regard to pins, overloaded pieces and so forth... to do more may >>make it too expensive to use... > >Yeah, that's what I was getting at. In my second position >(1k6/8/3nq1p1/4np2/8/6N1/5R1Q/4RK2 w), the focus of the >exchange moved from f5 to e5 midway through. So in that >case there were two brances -- continue at f5 or move to >e5. But I understand that the whole point of a *static* >evaluation is that you don't consider alternatives like >that (as you said below). > >>SEE doesn't do this... there is a single target square... to do more >>requires a normal search... > > >Side note, and I asked this in another branch of this >thread -- are there criteria that determine when you do >and don't do a SEE, or do you always do one when you're >in evaluate() and there are pieces under attack? > >Larry Coon I don't do *any* of this in evaluate. I use SEE to order captures, and in the q-search, to cull captures that seem hopeless...
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.