Author: Uri Blass
Date: 11:23:07 01/05/01
Go up one level in this thread
On January 05, 2001 at 13:10:50, Vincent Diepeveen wrote: >On January 04, 2001 at 20:50:53, Jorge Pichard wrote: > >>If Chess Tiger 13.0 plus Rebel Century 3 play against GM Van Der Wiel, allowing >>the best move by either program, Will they score better than Rebel Century 3 >>alone. Mr Schroedder can you ask GM Van Der Wiel if he accept the challenge for >>the remaining of the match? >> >>Pichard. > >Ok let's look to 2 things. >First the practical aspect. That's easy to describe. > >First of all both chesstiger and rebel don't run at the same time, >tiger is windows, rebel is DOS. > >Even if they would run in the future (i doubt it) at the same time >then still they get only half of the system time, all branch prediction >tables are each time the other process gets system time again worthless >so practical it's all worthless to do it. > >But let's now assume you do hell of a lot of effort and you get >3 pc's. one running rebel one running tiger. the third pc is carrying >a decision program which move to select. > >But now let's look to the theoretic problem. > >You say: "just select the best move of both". > >I assume you mean 'automatic' decision. Suppose after >d4 d5 c4 e5 dxe5 that both programs are out of book. >Tiger of course till the end of days say: "please play >dxc4 otherwise i'm a pawn down", whereas rebel might >say "let's go for d5-d4". > >What move to select? > >If you say: "just select the best move of both", >then you would not need the both programs. We only >need your module: "select the best move" to play perfect >chess. > >Basically your whole module is on a few rules going to >decide which move is best? > >So basically then these few rules limit the playing >strength from your program! Also i bet it's a preprocessor >module too. > >It'll play for sure worse as a ply1 decision in root goes >of course always for the biggest patzer move in the >position or a misevaluated score! > >Also note that you have no info from either program how >well the move the other program selects is! > >Anyway with a few metarules you are planning to overwrite >the choices a program makes, so basically the quality >of your module is going to decide how well you play. > >In other words you could run 3 programs. > >Program A (rebel) >Program B (tiger) >Program c ( 1 ply search program) > >Now if both A and B prefer the same move, then decision is >obvious. If they don't then program C's closest match >is going to decide. > >Let's do a small theoretic experiment on paper. > >Let's not run only program A and B, but another 20 programs >and let's decide based upon that. > >Basically what you get then is the average move of 22 programs. > >So if your decision criteria is taking the average move, then >obviously you play according to the average STRENGTH. > >Now suppose we would do that with 22 programs. For sure >the average rating of 22 programs is *not* better as the >program that's ranked highest from those 22. > >So any non-human decision criterium is complete bullshit IMHO. I disagree The decision can be based on the main lines and the scores of the programs. Examples: 1)Suppose Tiger failed high after 20 seconds with score of +1 and later failed low after 2 minute with a score of 0.02 and changed its mind to another move. Rebel failed high after 2 minutes on the same move that tiger failed high Rebel had the same main line of Tiger and did not fail low. I suspect that in this case Tiger outsearched Rebel so it is better to trust Tiger's move. 2)It is also possible to use the main line to decide if the program has a plan. If you see that the main line of Rebel has a lot of wasting tempo moves when the main line of Tiger has not wasting tempo moves then you can decide to prefer Rebel's move because Rebel seems to have a plan. 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.