Author: Robert Hyatt
Date: 17:46:21 05/17/00
Go up one level in this thread
On May 17, 2000 at 19:20:40, Ernst A. Heinz wrote: >Hi Bob, > >>I think playing Bxa7 very quickly is a bug. Because you are right, trapping >>the bishop is easy, and breaking the trap is deep. For every game you win by >>playing Bxa7 you will lose 10. > >Calling this a "bug" is a bold conjecture IMO -- your observations >above might be true for blitz games but they certainly do not hold >for longer time controls according to my experience. I based that on the fact you said "DT likes this from the get-go". That shows that at shallow plies it only sees winning a pawn, but not getting the bishop out of the trap. That is _very_ dangerous. There are plenty of positions where no search can see the bishop escaping, nor being lost. But it is loss nevertheless. If you would play on ICC a few months, you would probably decide that is a bad idea after you do it vs an IM or GM, and he notices. He will start leaving the pawn undefended, you will continue to take it, and he will continue to rack up rating points... Several of us have been there, done that. :) > >I prefer the approach of adding as little evaluation code for >special situations as possible. The "trapped Bishop" should >certainly get punished for losing most of its mobility. And >the potential necessity to defend the piece will punish its >side further. So why do we need special code for this? Because in a blitz game you will play that as a blunder. You don't play much blitz? OK. But what about near the end of a sudden-death time control? > >Moreover, the alleged "trap" often turns out to be a "fata >morgana" of human players who fear it without any good reason. >Hence, you may rob your program of many potential chances >if you severly punish "trapped Bishops" statically. > >=Ernst= As I said, for every pawn you win, you lose the bishop 10 times. If I can't find a way out after 12-14 plies, it is very likely it won't get out ever. I value the bishop over the pawn. :) I can point to many games vs commercial programs on ICC, where before they started to evaluate this term, they were losing the bishop right and left. Because my code tends to leave the rook pawns undefended since it knows that a pawn push traps the bishop. And if it can't see how to break the trap quickly, it assumes it won't break it ever...
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.