Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Fail-soft or Fail-hard ?

Author: Robert Hyatt

Date: 09:13:03 11/04/00

Go up one level in this thread


On November 04, 2000 at 11:33:22, Ulrich Tuerke wrote:

>>I'm not aware of anything that says that when using fail-soft, you _must_ start
>>off at +/- infinity.
>
>Isn't it the definition of "fail soft", to initialize best = -infinity in the
>full search region ? May be I'm confusing things again.
>
>> I don't, for example, yet I still let the search return
>>values outside alpha/beta.
>
>I don't see how to get an upper bound < alpha, when the best value had been
>initialized to alpha. Your search will just return the initial value in case of
>a "fail low", won't it ?
>
>> I found it slightly better because you get a bit
>>better info out of the hash table.  But it was only a "bit" better.  Not a huge
>>win.
>
>Why do you get better info out of the hash table in this case ?
>
>>
>>More important if you use mtd(f) of course.
>
>Thanks, Uli


No...  IE suppose you search at ply=10, and get a score of +.20...  Then
at the next iteration you start with a window (aspiration window) of something
like -.10, +.50...  with fail-soft you can get a score of +.90 or -.90 back.

While that won't be an accurate score, it supposedly gives you a better idea
of how far you need to relax the bound for the re-search...

normally you do something like this:

  v=-Search(-beta,-alpha,depth-1,etc...)
  if (v >= beta) return (beta)

the change is just to say

  if (v >- beta) return (v)

and do that everywhere...




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.