Author: Robert Hyatt
Date: 08:11:12 10/30/01
Go up one level in this thread
On October 30, 2001 at 06:35:57, Simon Finn wrote: >On October 30, 2001 at 06:34:27, Simon Finn wrote: > >>On October 29, 2001 at 10:46:55, Robert Hyatt wrote: >> >>>EvaluateWinner() is designed to catch positions like KRB vs KRPP. The KRB >>>has no chance of winning. KRPP _can_ win. EvaluateWinner() returns a >>>result that indicates that which tells the Search to avoid such positions >>>with the KRB and accept them (if the pawns are advanced enough) for the KRPP >>>side. >>> >>>In KRB vs KRPP the score will _never_ be positive. It will either be zero, >>>if the pawns are not advanced very far, or less than zero if the black pawns >>>are getting advanced enough to be a problem. >>> >>>But _never_ greater than zero since white has no chances. >> >>That's not literally true. It's true in normal positions, >>but how can you decide if a position is "normal" without search? >> >>KRB is mating material, and the mates can be >>quite deep. For example: >> >>[D]5rk1/5p1p/7K/6B1/8/7R/8/8 w - - 0 1 bm Bf6 >> >> >>After 1. Bf6 Rc8 2. Rg3+ Kf8 3. Kxh7 Ke8 4. Re3 >> >>White wins by bringing his king to b7. >> >>I'm sure there are plenty of other examples. >> >>Simon However, I will bet that for every example you can produce that causes my test to fail, I can produce one billion that work. _that_ is the point here. 100% accuracy is impossible. So go with the probability of which is better, (1) don't evaluate it at all and reach KRB vs KRP endings that the KRB side thinks it can win when it can't or (b) evaluate it as I do and avoid a great many embarassing positions. Note that I have been doing this evaluation for a couple of years so far, with no losses caused by it (so far)... The odds are in my favor here. And note that _every_ evaluation term has exceptions. _none_ are perfect. By definition, in fact...
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.