home *** CD-ROM | disk | FTP | other *** search
- Xref: sparky comp.ai.neural-nets:4664 rec.games.chess:12148
- Path: sparky!uunet!zaphod.mps.ohio-state.edu!news.acns.nwu.edu!uicvm.uic.edu!u16244
- Organization: University of Illinois at Chicago
- Date: Wednesday, 30 Dec 1992 04:34:40 CST
- From: <U16244@uicvm.uic.edu>
- Message-ID: <92365.043440U16244@uicvm.uic.edu>
- Newsgroups: comp.ai.neural-nets,rec.games.chess
- Subject: Chess program info??
- Lines: 20
-
- I'm getting the hankering to start back on one of my old projects, my
- chess program. However, I plan to steer clear of the current trend of
- search-a-trillion-nodes-a-second, and instead use lots of knowledge and
- search a very sparse tree. To implement this, I plan to use several
- neural networks to select canidate moves at every branch of the tree.
- The search will be breadth-first. Instead of searching 1 ply, 2 ply,
- 3 ply, etc, etc.. The tree will be kept, reordered, and extended until
- time considerations set in. Moves that are discovered to be bad will be
- cut from the tree. I estimate that on a fast machine, 30 seconds will
- probably only produce a tree of about 250 nodes; however this tree would
- be deep and knowledgeable. I would be interested to see this play a
- current program.
-
- My point in this post is to see what work similar to this has been
- done. I don't want to reinvent the wheel, or do exactly the same thing
- someone else has done and failed.
-
- Much thanks.
-
- --djh
-