Page 1 of 1

Simple Alpha-Beta Negamax search question

Posted: Wed Jun 30, 2010 3:07 pm
by zullil
While searching possible moves from a position, we find a move whose score S is greater than or equal to the value of beta. Thus we can disregard the remaining moves from that position, and return a score. It seems that whether we return beta or S (or any number greater than beta) makes no difference, since the position in question will not occur along the principal variation.

Is this correct?

Thanks.

Re: Simple Alpha-Beta Negamax search question

Posted: Wed Jun 30, 2010 4:05 pm
by Edmund
Not quite right ..
its the difference between fail-soft and fail-hard (http://chessprogramming.wikispaces.com/Fail-Soft)

As a result of returning a higher bound you will get tighter bounds on the parent's transposition entries which might improve future searches. Furthermore the null-move mate threat extension requires the fail-soft framework .. should you be interested in trying that.

regards,
Edmund

Re: Simple Alpha-Beta Negamax search question

Posted: Thu Jul 01, 2010 8:33 am
by Mincho Georgiev
Edmund wrote:Not quite right ..
its the difference between fail-soft and fail-hard (http://chessprogramming.wikispaces.com/Fail-Soft)

As a result of returning a higher bound you will get tighter bounds on the parent's transposition entries which might improve future searches. Furthermore the null-move mate threat extension requires the fail-soft framework .. should you be interested in trying that.

regards,
Edmund
Not quite right either. Fail-soft framework suggests that you can return score lower than alpha, if a node's best score do not exceeds alpha (fail-low). I know some will disagree, but regardless of the nomenclature, I prefer to consider 3 distinct approaches, in practice, looks like that:
1. Fail-hard - returned result is the upper bound of a node.
2. Fail-soft - returned result is the best score achieved (whether in or outside the window).
3. "? - the one I use" - returned result is at least alpha or the result of the further search (evaluation) of a node if greater than alpha.
Some papers indeed consider the 3th one (which also is the topic's subject) as fail-hard since no value, lower that alpha could be returned, but they are still different.

Re: Simple Alpha-Beta Negamax search question

Posted: Thu Jul 01, 2010 9:47 am
by Mincho Georgiev
Mincho Georgiev wrote:lower that alpha could be returned, but they are still different
correction: lower that alpha could NOT be returned, but they are still different.
typo, i apologize.