Author: Robert Hyatt
Date: 08:47:03 08/23/01
Go up one level in this thread
On August 23, 2001 at 11:08:54, Sune Fischer wrote: >Hi Folkes ;) > >I just upgraded my little engine from using a negamx to an alpha-beta algorithm. Negamax _is_ alpha/beta. It is just re-formulated so that + scores are always good for the side on move in the tree. Normal alpha/beta has +=good for odd plies, -=good for even plies, which makes the code messier. > >It is much faster, but still.... >With the negamax I had a branching factor of about 35 to 40, now is around >18-23. It seems you are talking about negamax without alpha/beta? Which is pure minimax. Alpha/beta works with a normal negamax representation, or with a pure minimax representation. And the node counts should be identical as the two are equivalent in terms of functionality. Negamax might be a bit faster as it is somewhat smaller and will have fewer branches. But functionally, the two should be equivalent. > >I get the feeling that that is still too high, I know it depends heavily on the >evaluation, but what kind of pruning cut-off should I realisticly expect with >the AB? Normal alpha/beta should give you an effective branching factor of about sqrt(x) where x is the effective branching factor of pure minimax. That ought to be somewhere around 6-10 max. > >If may ask, what algorithm do you people use, is it alpha-beta or negascout >or...? >Is there a clear answer to what the best search algorithm is? > >Cheers, >Sune Most everyone uses negamax as the framework. Whether you use simple alpha/beta on top of that, or something more sophisticated (I use PVS for example) doesn't change the look of the basic search that much.
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.