Computer Chess Club Archives




Subject: Re: evaluation tuning tricks

Author: Renze Steenhuisen

Date: 00:42:31 03/18/04

Go up one level in this thread

On March 17, 2004 at 14:17:45, milix wrote:

>On March 17, 2004 at 11:02:13, Renze Steenhuisen wrote:
>>On March 17, 2004 at 05:26:50, milix wrote:
>>>On March 17, 2004 at 04:58:06, Peter Alloysius wrote:
>>>>what are tricks for evaluation tuning so that it could
>>>>search less nodes?
>>>>i noted that my engine use exactly same move ordering scheme as
>>>>crafty but it still search much more nodes.
>>>>my engine also use adaptive null move, and futility pruning.
>>>>so i think the problem is on evaluation tuning.
>>>>i heard that tuning evaluation function can reduces node searched,
>>>>so what's the trick?
>>>I think that there is no special trick. A bad evaluation will missguide the
>>>search. When I turn off the positional evaluation (and have only material+piece
>>>square scores) my engine is doing too many researches and the move ordering is
>>>also very bad. Same effect if I alter a positinal characteristic too much, like
>>>giving a very big bonus in advanced pawns or penalize bad king safety too much.
>>If you turn positional evaluation off, what are typical numbers for
>>move-ordering? Because this is exactly what I do for now, but the History
>>Heuristic nor Killer Heuristic don't give very nice move-ordering %'s!
>I think both heuristics are suffer for bad evaluation especially when going
>deeper in the search tree. To test the bad behaviour of a weak evaluation you
>can set the rook value to 2 pawns and see what happens in your search especially
>in middle-game to end-game positions.

At the moment I only have the PST and Material Evaluation, and I hve bad FH-%,
so I wondered if you could give me the numbers? I would really appreciate it (an
approximation cq recollection from your memory is also just fine!)


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.