Author: Ernst A. Heinz
Date: 13:13:36 10/05/00
Go up one level in this thread
On October 05, 2000 at 11:48:46, Severi Salminen wrote:
>Hi!
>
Hi Severi,
>I ordered a copy of Heinz's book in the library of my home town.
>A few question about his AB explanation:
>
>1.He says: "While fail-highs with result>=beta place lower bounds on the real
>scores of failed searches, fail-lows with result<=alpha place upper bounds on
>them."
>
>So, what bound is he meaning (maybe my english is not so good): real scores of
>failed searches? I thought that Alpha and Beta are the only bounds involved -
>they are the lower and upper bounds of real score, aren't they.
No, how could searches ever fail then? Alpha and beta only
restrict the range of scores that you are _interested_ in
at the current node.
If a search fails, its result is not an exact score but just
a bound on the real score of the search.
* In case of a fail-low, the result returned is an
_upper bound_ on the real score, i.e.,
"result >= real score".
* In case of a fail-high, the result returned is a
lower bound on the real score, i.e.,
"result >= real score".
It is probably a good idea to explicitly add the inequalities
above to the bulleted list about the topic on page 14 of my
book "Scalable Search in Computer Chess" which you refer to.
>2. I didn't understand the fail-hard concept? How would a fail-hard
>implementation of AB work compared to fail-soft?
Instead of always returning the best score so far, fail-hard
searches always return the value of alpha in case of a fail-low
and the value of beta in case of a fail-high (see bottom of
page 15 in my book).
Thanks for your feedback and please enjoy the book!
=Ernst=
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.