Author: Robert Hyatt
Date: 13:32:32 10/21/98
Go up one level in this thread
On October 21, 1998 at 04:39:59, Ernst A. Heinz wrote: >On October 20, 1998 at 20:33:27, Robert Hyatt wrote: >> >>On October 20, 1998 at 08:20:53, Ernst A. Heinz wrote: >> >>>On October 20, 1998 at 08:08:23, Robert Hyatt wrote: >>>> >>>>Quite easy. With alpha/beta the only "order" we have is (1) best move and >>>>(2) rest of moves. There is no "order" to the moves other than one move is >>>>better than all the rest. No idea how much better, no idea where the other >>>>moves rank with respect to each other... >>> >>>In order to avoid wide-spread confusion, I would like to add that Bob's above >>>statement implicitly assumes "alpha-beta" to mean "PVS/NegaScout" as employed >>>by most chess programs. >>> >>>Pure alpha-beta without null-window searches and eager resolution of any >>>fail-high/fail-low situations can of course calculate accurate search scores >>>of all moves. But it would be horribly inefficient as compared to PVS (yet >>>still much better than naive minimax). >>> >>>=Ernst= > >>No "pure" alpha/beta can do this. But if, at the root, you simply reset >>alpha/beta after each root move, you get a score whether it be from a PVS >>search a negascout search or a traditional alpha beta > >Right, that is exactly what I said -- so we agree on this. :-) > >It is not alpha-beta that hinders the calculation of accurate search scores. >Instead, the null-window searches which represent the heart of PVS are the >real culprits. If you remove the null-window searches and resolve any >fail-high/fail-low at the root you can easily calculate the accurate search >scores of all moves. Exactly as I said. however, traditional alpha/beta also won't return true scores, even without PVS or other null-window ideas... because if you search the first move best at the root, that's all you know when the search is over... that the best move is best, but no idea how much better than #2, or even what number 2 is. Unless you do the same trick I suggested... Alpha/beta only proves that non-best moves are non-best, no matter which variant is used... and in the case here, any attempt to extract more information (which is best, second-best, etc...) comes at enormous cost... > >>(I don't see what null- >>move has to do with this issue at all). > >Where did you read "null move"? I wrote "null-window search" ... :-( > mis-read... sorry... >=Ernst=
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.