Author: Uri Blass
Date: 10:33:37 05/30/01
Go up one level in this thread
On May 30, 2001 at 09:10:54, Vincent Diepeveen wrote: >On May 29, 2001 at 14:48:06, Uri Blass wrote: > >I guess what you basically want to know Uri are the moves >that first failed high for black on a depth n , but now at >a depth n+1 they fail low for black! > >Kxg7 is such a move! > >So basically moves flipping from >= beta to <= alfa is what you >want to know. > >However in a search that big we talk here about quite >some moves. Suppose 2 million moves that flip in this huge >search from >= beta to <= alfa. Is that what you want to >see on the screen? > >How big must your screen be? > >The main problem is that search works perhaps a bit different as you >think it does. > >Bxg7 Kxg7 now white has say 40 legal moves. That means that black after >all these moves has 40 moves that give a cutoff for it! > >In short in a search you need more or less branching factor to the >power depth, that's the number of moves that gave a cutoff! > >So you definitely don't want the cutoffs only. You want the flips >from >= beta to <= alfa. > >But here lies the problem. the programs main variation is based upon >loads of moves. You want to see a line which an intelligent person like >Uri wants to consider. > >Already a year or 55 ago it was estimated that strong players are >considered to have a branching factor of 1.76 > >Now you want as a human to see usually 1 move and sometimes 2 moves. > >How is the computer going to figure out what moves out of the 40 >legal moves you want to see? I want to see the line that the computer considered for more time. Bxg7 Kxg7 Ne5 is clearly considered for a lot of time even if the computer rejected Kxg7. Bxg7 Qe5 Qxe5 wins a queen for white and I am sure that the program did not consider it for a long time(I assume that Qxe5 is the killer move for Qe5). If you suppose that you find the most common 1000 lines that were considered for more time then this line may be something like line number 501. The first lines are the interesting lines that I want to see. I want to see something like this in the following position: [D]2r2rk1/p1q1bpp1/1p6/n2R4/8/P4N2/1B2QPPP/5RK1 w - - 0 1 1)1.Bxg7 10 hours 2)1.Bxg7 Kxg7 2.Ne5 7 hours 3)1.Bxg7 Kxg7 2.Ne5 Rfd8 Qg4 5 hours 4)1.Bxg7 Rfd8 Rxd8+ 4 hours ... 501)1.Bxg7 Qe5 2.Qxe5 1 minute I know that the computer may consider some stupid lines for a long time because of not having the right killer move but there are still a lot of interesting lines and it is easy to see if a line is not interesting in a few seconds. I am afraid that the main line of Ferret may be something like 1.Bxg7 Rfd8 2.Rxd8+ and I will not get the main important information for the reason that Ferret failed high on 1.Bxg7(I remember that something similiar happened in one of the nolot positions). It is possible to ask bruce to do a long search in some interesting position 2 plies or 4 plies after the root in order to get better information about the best move and not the killer move but the information about the killer move is good enough for me. 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.