Author: Will Singleton
Date: 21:37:22 09/26/01
Go up one level in this thread
On September 27, 2001 at 00:29:18, Will Singleton wrote: >On September 27, 2001 at 00:06:35, Jeremiah Penery wrote: > >>On September 26, 2001 at 21:45:46, Robert Hyatt wrote: >> >>>On September 26, 2001 at 20:32:58, Dann Corbit wrote: >>> >>>>Or if the opponent move is forced, I ponder the response. >>>>Or if the opponent move is obvious or takes most of the search time etc., I >>>>ponder the response. >>> >>> >>>Here is the problem.. I had to explain this to Komputer Korner a few years ago >>>as well... >>> >>>If you correctly predict your opponent's move at least 50% of the time, or >>>more, then the way we currently ponder can _not_ be improved on. Because in >>>at least 50% of the cases, we will be correct, and we focus all our search time >>>on the _right_ move. If our target time is 3 minutes, and our opponent makes >>>a move after 3 minutes or longer, we can move instantly and use no time on our >>>clock. >>> >>>For any other scheme, you are going to split your search time among at _least_ >>>two moves and more likely more than that. That means that after your opponent >>>searches for 3 minutes and makes a move, you haven't searched more than 90 >>>seconds on any one move. You have to keep going for another 90 seconds if one >>>of the two moves you have been fiddling with is correct. And if not, you have >>>to spend a full 3 minutes. So best case is you save 90 seconds. If you could >>>narrow your pondering to 2 moves, and the opponent _always_ played one of those >>>two moves, you will save 90 seconds every move, where I save 180 seconds every >>>other move (assuming a 50% prediction rate). We are "even". But I know I am >>>going to be wrong one of every two moves (actually less, as against GM players >>>in long games I get 75-80% right generally). But if you fiddle with more than >>>2 moves, you will lose big time, because if you try 3, you will spend 60 seconds >>>on each, and when your opponent moves you only saved 60 seconds if he plays one >>>of your three possibilities. I save 3 minutes every other move. You save 180 >>>seconds every other move. It is easy to see which is better. >>> >>>If you can't predict correctly 50% of the time, then this changes of course. >>>But I have never seen that happen, at least in my case. If it does, my opponent >>>is losing badly. >> >> >>I think Dann's idea is a good one, and here's why: >> >>Probably most of the time you correctly predict the opponent's move are times >>when the move seems "forced", or at least clearly better (>0.3 pawns?) than the >>other moves. Obviously, in such positions you will predict near 100% against a >>good opponent, and these types of positions happen a lot. In every game I've >>ever seen, there are at least a few moves (or sequences of moves) that are >>almost totally forced. In these forcing lines, Dann's scheme of pondering will >>act the same way as it currently does, therefore causing you to lose nothing >>here. IMO, you can gain a lot, however, from not just picking some random move >>to ponder when a lot of moves are near-equal. >> >>Take the opening move, for an easy example. If you have no opening book, and >>you play 1. e4, what move are you going to ponder on? There are a bunch of >>viable moves here, so just because your evaluation happened to prefer 1. ...e5 >>by 0.01 over 1. ...d5, or Nf6, or Nc6, or some other move means you should ><cutoff> > >After scratching my head for awhile, I think I understand Dann's idea :) >I believe the fatal flaw is that you don't know the relative values of the >responses to e4, you just know that e5 (for example) is better than d5. So you >cannot decide that the anticipated responses are "about equal" until you >actually search them individually, which you would never do. > >Does that make sense? > >Will Oops, you were scoring the moves by time spent searching, not by score. Hmmm.. Actually, sounds like a good idea. Have you tried it? What are the results? Just to try it would be fairly trivial, it seems. Will
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.