Author: Gerd Isenberg
Date: 13:13:15 04/05/04
Go up one level in this thread
On April 05, 2004 at 12:54:30, rasjid chan wrote: > >I think most chess programmers implement hash tables my way, ie >by thinking it thru themselves. All the internet articles I found >about hashing only mention the the bare idea of hashing. I am not sure >whether any book on chess programming devote a full chapter on hashing >and if they do, what details do they provide. > >I remember a certain professor (Marsland) who wrote something like... >the bulk of chess programming bugs come from hash-tables.... >alpha-beta interacts with hash-tables in strange ways... > >At least they interact strangely for me. Every a little while, it seems >something new turn up and I just have to post one example, a recent >observation,to this forum to be safe that I am not way wrong ! > >I do fail soft and fails / hash outside window bounds. >If on a hash probe and the hash is EXACT and fails high, >I return hash value ( > beta). The ply below fails low with >this "exact score". If your stored depth (draft) is greater or equal to your current search depth, exact is exact, in fail soft independent of your current search window. e.g. see Bruce Moreland's excellent site: http://www.seanet.com/~brucemo/topics/hashing.htm > >If after seaching all moves of a node and it fails low with a best-score >< alpha, the normal hash is hash as best-score as upper bound. >But if this best-score happen to be "exact", I fail low but hash as >EXACT. > >Hope I am not missing something big ? Replacement scheme is a big issue here. Replace always, Replace smaller depth only, or probably better, some slots (4..8) per entry with a more sophisticated replacement scheme. There are even two table approaches with different replacement schemes. Considering latency or even tlb-misses with huge tables, it may be worth to exclude qnodes from hashing a huge transposition table. Cheers, Gerd > >Rasjid
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.