Author: Ed Schröder
Date: 00:42:34 11/30/99
Go up one level in this thread
>Posted by Bruce Moreland on November 29, 1999 at 19:30:02: > >In Reply to: Re: Positional/Real Sacrifice posted by Ed Schröder on >November 29, 1999 at 19:00:25: > >On November 29, 1999 at 19:00:25, Ed Schröder wrote: > >>I can not believe you saying this. After 1.Bxh6 gxh6 2.Qxh6 there is a queen, >>2 rooks and a bishop all pointed very dangerous at a naked king. Rebel giving >>+1 is fully justified. >> >>Every 1800 rated chess player will play 1.Bxh6 immediately, no need to >>calculate. > >He is just being conservative. Every 1800 rated player will play this, and >they'll beat every other 1800 rated player, but they'll lose a lot of the time >against a strong computer. > >I'm not trying to say you are wrong in this case. In this case the sac works, >no doubt. And I'd bet that in a too-high percentage of cases it either >works or I have to watch the silly human opponent miss a +2 move before >going astray and losing. I agree with you that blitz is a whole different topic as even stronger players like an IM may let black escape at that time control certainly if they are playing a computer. But any 1800 rated player is supposed to win after 1.Bxh6 gxh6? at 40/120. >Are you saying that you don't try to calculate these, that you'd give that +1 >based upon static eval? Do you do this based upon real evidence that it >works? Yes, and suppose there is an escape for black after all then SEARCH will do its job being a filter for possible exceptions. To me this is a classic case of the very basic of king safety: QRRB all pointed at a naked king, the king having hardly any escapes, the black pieces not able to defend its own king. >I feel better about this if you are doing this statically, what worries me is >that everyone might be seeing material coming back in the eval, or seeing in >eval or some other way that the other white pieces really can infiltrate. I >don't understand how to do the former, although the latter sounds like a >challenge worth exploring. Same here. Note that its precursor Rebel10 needs 58 seconds and 7 plies to see 1.Bxh6 score going up from 0.40 to +1.00 not expecting 1..gxh6? also, meaning 1.Bxh6 probably was found pure on tactical grounds (not sure if this is entirely true maybe gxh6 is rejected on positional grounds as well). Now in Rebel Century 1.Bxh6 is played on ply 4 (default settings) and I am pleased to see that. In Rebel Century I rewrote king safety after the loss against GM Rohde (first game in the GM challenge) as Rebel had to see a losing move f4??? in its search and since it was too deep the bad move f4??? was played and Rebel lost because of that move. After a close look at the position in question I came to the conclusion Rebel's king safety was lacking important information and f4??? could be solved by just making king safety more intelligent. As a result f4??? is now evaluated 0.75 lower and the good move Bg2 (that perhaps had saved the game) is now played 2-3 plies sooner. In general I believe that as long as positional sacs in eval covers 90% of the cases (thus being not accurate 1 of 10 times) SEARCH is able to filter these cases most of the time. Meaning to say, in very rare cases you will be faced by an incorrect sacrifice (I haven't seen them yet) of your program but so what? Just fix it in case it happens. And yet if such a rare case occurs (an incorrect sacrifice) the human will have a very hard time to escape from the attack as there definitely is a lot of compensation. Humans (even GM's) may become under time pressure and the chance they make a mistake is quite high. This all is (most of the time) not true in comp-comp as an incorrect sacrifice most of the time will be proven by them. You need a good king safety desperately against these GM giants playing at 40/120, if your program can't fathom a position like 1.Bxh6 or similar then better not play them. That's what I learned from this first game in the GM challenge. This is quite different in comp-comp. If both are dumb they perfectly keep the balance both not seeing simple cases of a king attack. Ed >bruce
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.