Computer Chess Club Archives




Subject: Re: Performance of MTD(f) versus eval granularity?

Author: Andrew Williams

Date: 08:59:33 11/02/01

Go up one level in this thread

On November 01, 2001 at 12:47:43, Werner Mühlpfordt wrote:

>Hi all,
>I've got a question on how the performance of the MTD(f)
>search algorithm will depend on the eval function's granularity.
>At first glance, it seems that with increasing number of
>decimals delivered by the evaluator, the search time will go
>up because the binary search will need more cycles to reach
>convergence. On the other hand, a search of a given depth
>reaches only a limited amount of positions with an eval near
>the best one. Thus, any more accuracy would be for free when
>the "tick" size is below the difference to the position that
>is closest to the best one, wouldn't it?
>And if the number of decimals chosen is sound, then the difference
>between this "next-to-best" position to the real best one would
>really matter, and the time would be well spent.
>Does this allow to conclude that the factors of choice for eval granularity
>are no way different for MTD(f) than for other algorithms?
>Thanks in advance;

I've used MTD(f) in my program for a very long time, but I've never
tried an eval with other than centipawn granularity. If you test this,
I (and I'm sure other people) would be most interested in your results.



This page took 0.02 seconds to execute

Last modified: Thu, 07 Jul 11 08:48:38 -0700

Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.