Author: Robert Hyatt
Date: 13:33:44 10/04/00
Go up one level in this thread
On October 04, 2000 at 12:50:35, Uri Blass wrote: >On October 04, 2000 at 10:23:48, Robert Hyatt wrote: > >>On October 04, 2000 at 08:44:44, Mike Adams wrote: >> >>>I just want to point out what i'm told, correct me if i'm wrong. A better >>>evaluate helps an engine search deeper because it begins to distinguish more >>>between good and bad moves when no obvious material gaining moves are available. >>> The more it can differentiate moves the more alpha beta cutoffs it can get >>>which means less nodes searched for a given depth and consequently more time >>>spent getting more depth. Also as in the case of this game were pulsar made bad >>>decisions a good evaluate can both defend against the opponents tactics and >>>create tactical possiblities. >> >> >>I don't buy that. Evaluation slows you down. No evaluation at all will >>search the perfect game tree, node-wise. Adding evaluation will begin to >>make move ordering get worse. Which will expand the tree, not contract >>it. Matrial-only will search the smallest tree, assuming you do a decent >>job of ordering captures. > >I agree that evaluation will slow you down if you are interested only in depth >but it does not say that it will slow you down if you are interested in seeing >tactics faster. > >Evaluation can help you to see tactics because it can help you to search the >right lines. It depends on how you define "evaluation". Traditionally this is the thing called _after_ the search has reached what it considers to be a "quiet position". Some use the eval and let it return a "search deeper" indicator. But this doesn't have anything to do with the question asked. It simply said "as I increase knowledge will I also search deeper?" and the answer is a simple "no". In the context of the question, it seems that he has been given the opinion that more knowledge somehow makes the program more efficient and lets alpha/beta work more efficiently. That doesn't happen. > >With bad evaluation you may prune good lines because of null move pruning when >good evaluation is not going to prune the same lines because the good evaluation >will tell you that there is a threat. > >In most cases with material only evaluation there is no problem of losing >material so the order is perfect because of the bad evaluation but the question >is what happens in cases when there is a tactical idea. Knowledge generally has nothing to do with tactics. Tactics is handled by the search and the extensions. Knowledge is used to choose between lines that the tactical search has proven to be equal in terms of material. There are some vague exceptions, perhaps, but not in a beginner's program. > >The fact that you can get bigger depth with material only evaluation in most of >the positions does not prove that material only evaluation does the program >better in tactics and the question is what happens in cases that there is a win >of material for one side. > > >Uri
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.