Author: Don Dailey
Date: 21:21:18 02/03/99
Go up one level in this thread
On February 02, 1999 at 19:32:40, Larry Griffiths wrote:
>Thanks for the help a while back with Alpha-Beta and MinMax.
>I did indeed get the same results with or without Alpha-Beta code.
>I want to put Hash Tables back into my program.
>(The last hashtable code made my program play like a drunken sailor).
>
>My new question is:
>
> Should a hashtable produce the same score as MinMax provided that
> the scoring only occurs at the leaves?
> (I will disable my incremental scoring code from branches).
>
>Thanks in advance for your replies!
>
>Larry.
Very complicated question. The answer is probably NO. This is because
your program probably has move order dependencies in it. Here is a
way to fix this:
1. Turn off repetition detection and 50 move rule. Pretend these rules
do not exist.
2. Allow hash table matches only at EQUAL depths.
3. Take out ANYTHING that is based on move ordering dependancies.
a.) Many kinds of extensions (checks probably ok)
b.) Most kinds of selectivty.
c.) Preprocessed evaluation.
4. Make sure hash tables know about castling and en-passant status.
If you do these things, probably now you are guaranteed equal scores
if your program is otherwise completely correct and you evaluation is
not dependant in any way on move history.
This is a useful mode to have in your chess program for testing and
debugging, assuming you have the right kind of evaluation to support
it. Otherwise it would be difficult to implement.
- Don
P.S. You don't have to restrict to end nodes, just compare only
equal depth positions.
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.