Author: Dave Gomboc
Date: 15:50:11 04/05/99
Go up one level in this thread
On April 05, 1999 at 17:40:50, Bruce Moreland wrote: >On April 05, 1999 at 16:29:16, Dave Gomboc wrote: > >>On April 01, 1999 at 13:11:29, Bruce Moreland wrote: > >>>But what I do is only include *game* position that have occurred *twice*. A >>>position that occurs *once* is not included in this table. >> >>And you can clear out this table after every non-reversible move, yes? > >It is a hash table, not a list. If I cleared it I would have to reconstruct >it, or copy ir around all over the place. Since it's a hash table it doesn't >cost a lot to look things up. The impossible entries don't cause any trouble. > >bruce Hmm... this got me thinking. I have been keeping track of repetitions in my awari program by linking back through every second position (but checking for non-reversible moves on every position). This method is discussed on the Cilk Pousse page: http://supertech.lcs.mit.edu/~pousse/#HASH (thanks, Don! :). But I only do the checking when there are less than eight stones on the board, because it hurts my search performance quite a bit. In Awari, the first time a state is re-reached, the game is decided (by the current material count), so I don't have the same difficulties that chess programmers have. So if my normal hash table is doing the job well (which it isn't really yet, because I haven't made it two-tier) then I wouldn't have a need for this special hash table. Well, that's not true either, it will still be faster if I have a small hash table for actual game positions encountered, assuming that the probe cost doesn't make it not worthwhile. (In chess, the probe cost is negligable, but the overhead of a make-move is *really* low in Awari, and while my program uses a hash table, some of my classmates found themselves running faster without one!). Thanks for the insight. Dave Gomboc
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.