Author: Robert Hyatt
Date: 16:18:52 08/09/98
Go up one level in this thread
On August 09, 1998 at 10:22:31, Christophe Theron wrote: >On August 08, 1998 at 00:10:39, Shaun Graham wrote: > >>On August 07, 1998 at 18:17:45, Christophe Theron wrote: >> >>>On August 07, 1998 at 15:48:43, Shaun Graham wrote: >>> >>>> Is there a trade off in strength that "can or perhaps does" occurs for >>>>programs that play blitz exceptionally well, when it comes to playing standard >>>>games? For instance Genius is held by many to be the best blitz player, my >>>>feeling is, that this is simply due to it's excessive tendency to play >>>>defensively(defense comes quick to genius). This makes it considerably more >>>>difficult to form a constructive plan of activity, which in the end results in >>>>opponents falling too short of time and losing. However this strength which >>>>makes it so great at blitz also is one of the main causes for it falling behind >>>>the other top programs(IMHO). >>> >>> >>>I think differently. Genius strength in blitz comes from its search algorithm. >>>Genius and its predecessors were designed to play the best on slow hardware >>>(Mephisto computers), and Lang's search was designed at that time. And was the >>>best. >>> >>>The result is that Genius sees many long combinations instantly, but it has a >>>bigger branching factor. As time controls become longer, the branching factor >>>becomes a problem and other programs are able to go as deep, or even deeper than >>>Genius. >>> >>>It is also possible that the defensive behaviour you describe comes from the way >>>the last plies of the search are treated. Computers moves are not processed in >>>the same way as opponent's in the last plies of the selective search. However I >>>don't think it is the reason why Genius is so good in blitz games. >>> >>> >>> >>Someone not too long ago posted a quite detailed description of how Genius does >>it's analysis. Indeed according to my memory of the post Genius does analyse >>it's opponents move differently than it analyses it's own(i'm not sure if this >>is common practice or not among other programs). > > >Thorsten talked about this, but maybe it applied to older versions of Lang's >programs. Nobody has given evidence that Genius selects differently its moves >than the opponent's, except in the very end of the lines. > > > actually, several posted a few positions where Genius appeared to show an asymmetric search. It couldn't find the key move with a very deep search, yet if you play the key move and let it play the other side, it would find that it was lost very quickly. Which lends credibility to the idea that it looks at everything for the opponent, but prunes (forward prunes) its own moves quite a bit in the right circumstances... > >>I'm certainly not as expert on >>programming as you Christophe, but i do believe that the search strategy >>employed by genius does often result in a defensive behavior were less weakneses >>are created, thus making it much more difficult to find a constructive plan, and >>thus slowing down the opponent(at least when human). >> Really though the general question i was asking is there a trade off that can >>or perhaps does occur, in current programs when it comes to a program being good >>at blitz and Standard? For example i was attempting to describe, how perhaps >>the features that cause genius to be so good at blitz are perhaps the same exact >>features that cause it to perhaps be less good than some other top programs at >>standard time controls(40/2). >> >>Shaun > > >In the case of Genius, I don't think the defensive behaviour you are describing >is the reason why it is less good at slow time controls. > >I repeat that I think it comes from a bigger branching factor, that makes >-maybe- Genius search less effective when compared to other programs at slow >time controls. > >This is not criticism against Genius. I think it is one of the best program ever >written, and its search was very well adapted to slower computers. > >Now that computers are faster, Richard has to adapt its search. No doubt that he >will do it and strike back. > > > Christophe
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.