Cho Cho
User
Newbie
| Posts: 2 |   |
Karma: 0
|
re:Fail Soft Alpha Beta & Transpositions - 2006/12/16 18:47
Yes. The important thing is what we know about the children. If we searched them *and* got exact scores, then we know an exact return. But if we got only bounds, because of beta cutoffs [ie, some of the grandchildren weren't looked at], then we know only a bound.
When searching children, if [eg] your alpha and beta bounds are close to zero, then winning a pawn will cause a beta cutoff [other things being equal], but you may have missed the win of a queen. That is the beta bound that you understand.
If your move *loses* a pawn "at first glance" [so to speak], eg in the principal variation, then your child will similarly have a beta cutoff but may have missed winning a queen, and so *you* may have missed *losing* a queen.
[In human terms, this is actually the more intuitive part of alpha-beta pruning. If you see that a move is bad, eg that it loses a pawn, then you don't consider it further, and in particular you don't waste time determining exactly how bad it really is. But the computer only knows you lose a pawn because it has analysed further, and the other player is seen to be able to win a pawn.].
---------
Ask others about themselves, at the same time, be on guard not to talk too much about yourself. - Mortimer Adler (b. 1902), American philosopher, educator, How to Read a Book
Popular posts by Cho Cho
Rebel & Inclusive/Exclusive Analysi...
|
post new topic
