Author: David Rasmussen
Date: 16:37:11 12/16/02
Go up one level in this thread
On December 16, 2002 at 07:18:11, David Rasmussen wrote: >On December 16, 2002 at 06:22:26, Uri Blass wrote: > >>>00:00:37.73 3967kn 14/26/27 +0.510 1. b4 Re4+ 2. Kd5 Rd4+ >>> 3. Ke5 Re4+ 4. Kd6 Rd4+ >>> 5. Ke6 Re4+ 6. Kf5 Rf4+ >>> 7. Kg6 Rg4+ 8. Kf7 Rg7+ >>> 9. Ke6 Rxc7 10. Kd5 Rb7 >>> 11. Kc5 Re7 >> >>Something seems wrong with the evaluation of chezzz here. >>It gives white a positive score even after white sacrifice the rook. >> > > >Mmm. Thanks for pointing it out. If I give Chezzz the position after >8... Rg7+, it chooses Kxg7 instantly with a 0.000 score (since it's stalemate). >If I give it the position after 9. Ke6, it instantly chooses >9... Rb7 with a Mate In 43 Ply score (Mate in 21, I guess). I will have to look >into that. Come to think of it... This might not be a bug at all. What happens is that b4 is searched until it suddenly fails low (you can't see this in the output), then it researches with a ]-INF;alpha] window, and b4 is searched first, as it was in the previous PV. The position at the end of the new PV starting with b4 gets the score 0.510, which is propagated back. The rest of the moves are searched, and Kc3 is found to be better. This is not so weird is it? What do others do in this case? As long as the fail low extends the time, to find Kc3 (in this case), it's ok. Still, there is something weird about the "sub pv" from 9. Ke6 being what it is, since when that positions is searched as the root position, the mate is found. Why isn't it found in the original search? Also, why is 9. Ke6 prefered in the search, when Kxg7 is prefered when that position is "on the board"? I will have to look into that, but it might not be as obvious a bug as you (and I initially) seemed to think. It is normal to fail low, on this move that is to be avoided, and then to find a better move. After white sacrifices the rook, why should white's position be negative? I know it's a mate, but if it takes 10 ply to see that, it is natural that the 4 ply search below the position after Rxc7 doesn't find it. The position at the end of the PV gets a +0.510 score. That is not totally unreasonable for a static evalutation, is it? /David
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.