Author: JW de Kort
Date: 13:58:22 05/19/03
Go up one level in this thread
On May 18, 2003 at 23:17:14, Ross Boyd wrote: >On May 18, 2003 at 09:52:43, JW de Kort wrote: > >>Dear friends, >> >>For a long time i have been planning to use attack information in my 0x88 based >>program. Before evaluation starts, my program calculates a table with attack >>information like the one discribed on the internet page of Ed Schroder. The >>result is disaterous: nps count has dropped by 50%! I have taken a look at the >>Sjeng code but as i see it, this must also be a serious reduction of the NPS in >>this program. >> >>Can anybody give me some tips to do this efficiently? >> >>Thanks in advance, >> >>Jan Willem > >Hi Jan, > >I'm in the same process of implementing an 0x88 SEE routine (for move ordering >purposes) and then when that's working, adding attack tables. > >The Sjeng SEE is quite costly, and beware that it does not correctly evaluate >xray attacks.... if you look closely, it doesn't look at bishop/queen attacks >that xray through a pawn/queen/bishop of the same side. Same goes for multiple >rook/queen attacks from the same direction... but its easily fixed though... at >the cost of some more CPU cycles... > >Anyway, it still compares poorly with my old mvvlva based ordering. It searches >much slower but the tree is only slightly smaller... >So, I guess we're in the same boat. There has to be a good way of doing this, >perhaps incrementally... > >Ross Dear Ross, i have a working SEE and i think i can correctly evaluate xrays but i'am not 100% sure. It sure is a costly routine but in my program the result is less nodes but at a lower speed. in the end the reduced number of nodes gained about the same time as the extra time use by SEE costs. regards Jan Willem
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.