Author: Jeremiah Penery
Date: 07:34:15 05/23/00
Go up one level in this thread
On May 23, 2000 at 08:00:53, Steve Coladonato wrote: >On May 23, 2000 at 00:05:36, Jeremiah Penery wrote: > ><snip> >>http://www.rebel.nl/db2.htm > >The following in an excerpt from the above link and I don't fully understand the >evaluation. > > 13. Nc6-a5 > Kasparov,G :11.00 -0.03 Nc6-a5 Bb3-c2 c7-c5 d4-d5 g7-g6 b2-b4 > Rebel 8.0 :11.00 -0.03 Nc6-a5 Bb3-c2 c7-c5 d4-d5 g7-g6 b2-b4 > 14. Bb3-c2 > DEEP BLUE :11.00 0.13 Bb3-c2 c7-c5 b2-b3 Ra8-c8 Bc1-e3 g7-g6 > Rebel 8.0 :11.00 0.13 Bb3-c2 c7-c5 b2-b3 Ra8-c8 Bc1-e3 g7-g6 > 14. c7-c5 > Kasparov,G :10.00 -0.05 c7-c5 b2-b3 Ra8-c8 Bc1-e3 g7-g6 d4xe5 > Rebel 8.0 :10.00 -0.05 c7-c5 b2-b3 Ra8-c8 Bc1-e3 g7-g6 d4xe5 > >I am interpreting the evaluations to be relative and not absolute (for the 3 >moves above, the white position is evaluated better in all cases) So, at move >13, Rebel evaluated the position -0.03 at 11 ply. At move 14, the evaluation is >now 0.13 at ply 11 again. This I can understand because the line has changed >from d4-d5 to b2-b3 and move 14 has looked one ply deeper than move 13. >However, on black's move 14 the ply count is now 10 which indicates that the >depth reached in the analysis is the same as that for white's move 14 but the >evaluation has changed from +0.13 to +0.05 (absolute) the line being identical >except fot the inclusion of d4xe5 in the printout. My question is why would the >program change its evaluation of the position if it searched to exactly the same >depth? It looks just like a slight asymmetry in the evaluation function. When evaluating the same position from the white side and black side, I think a lot of the programs will get slightly different numbers, because of asymmetry. The asymmetry can be useful when you want to, say, value your own king safety more than the opponent's. This way, you can do more to prevent attacks on your own king, but you won't be making silly attacks on the opponent.
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.