Author: leonid
Date: 11:47:49 09/20/99
Go up one level in this thread
On September 20, 1999 at 09:19:33, Robert Hyatt wrote: >On September 20, 1999 at 07:19:53, leonid wrote: > >>On September 19, 1999 at 22:10:39, Robert Hyatt wrote: >> >>>On September 19, 1999 at 21:51:12, leonid wrote: >>> >>>>Please, say briefly what can reduce the "branching factor". I hope >>>>that something very simple I am missing in this aspect. My branching factor >>>>is only reason to feel me humiliated about my game. Maybe different logic >>>>lead to different "factor"? Do somebody came through the similer experience? >>>>And if it was so, what was the reason that you found? >>>> >>>>Thanks, >>>>Leonid. >>> >>> >>>There is only one way to reduce the 'branching factor'... that is to use a >>>purely selective search algorithm, where you throw out moves at every ply >>>without searching them at all. IE alpha/beta reduces the branching factor >>>from about 38 to sqrt(38). >>> >>>You can reduce the 'effective branching factor' by using null-move, or any other >>>idea that selectively reduces the search depth for selected branches... >> >>I use alpha-beta already (found it some 8 month ago) and do my move ordering >>after the material advantage that each move give. I use also the best moves from >>the previous searches. But how about the null move? I never used it for one >>simple reason, my idea is that null move give you the speed but make you >>loose the precision from your search. In other words, null move is good way >>for speeding the game but lead sometime to wrong moves. True or not true? >> >>Leonid. > > >It will give you two plies of deeper search. It will make some errors happen. >But overall, it is stronger (for me) than non-null-move searches. IE when I >play crafty vs crafty with null-move disabled, crafty wins more games will >null-move on than with it off... But how about the "brute force" search. It is just in this level that my engin loose. The speed that "nulmove" give is not that important to me. Two additional plys, while looking though fast logic, my game can do easely and I even can explain how, if somebody interested. But once again to bad "brute force" "branching factor". To give the idea what I am been talking about, here is one typical example. I tried it today. Solving the position in 6 plys my engin lead but loose as usually up to the 10 ply. AMD 400Mhz. Hash tables off for both games. My engin: 6 plys 0.22 sec. 8 plys 18 sec. 10 plys 14min. and 54 sec. Rebel 10: 2.6 sec 40 sec 12min. and 8 sec. If my "branching factor" between 6 plys and 8 is 82, for Rebel it is only 15.3. This is how Rebel take over on the 10 ply. Number of positions per second are almost indentical but my are higher. Average number of nodes per ply is 36. Do you have some idea why this "branching factor" could be so high? Thanks for every idea that can lead to some solution, Leonid.
This page took 0.01 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.