Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: mtdf

Author: Andrew Williams

Date: 02:32:48 05/15/99

Go up one level in this thread


On May 14, 1999 at 12:50:29, Dave Gomboc wrote:


[snip]

>>>very well.  Another observation is that it is usually less work to prove that
>>>you are too low than you are too high.  This is due to the nature of minimaxing.
>>>
>>
>>Yes. For this reason, for successive fail-highs, I add on 1 then 2 then 3 then
>>4 then 5 etc etc. For successive fail-lows, I subtract 4, then 9, then 16, then
>>25, then 36 etc. I have messed about with this a *lot*, and at the moment I'm
>>happy with this. I use fail-soft, of course as you say.
>
>So if you search the window for +50, and the fail-soft returns 49, the next
>null-window you try is 46?  If the fail-soft returns 40, are you still using 46,
>or are you using 40?
>

I'm glad you asked this. I was going to say that I start with the score I get
back from the search, but at some point I must have changed it. I use the
guess and then add (or subtract) the appropriate amount. If the result is
outside the appropriate bound I just set the guess to be the bound +/- 1. If
this isn't clear, I'll post the code, but it's pretty ugly - a result of
changing it over and over, I guess. If I find time, I'll change it to see why
I don't use the returned score as the basis for the next guess - I have the
feeling that it didn't mix well with my approach to increasing the steps, but
now I think about it, maybe I "fixed" the wrong thing. It wouldn't be the
first time.


Regards

Andrew



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.