Author: Uri Blass
Date: 14:03:59 12/13/05
Go up one level in this thread
On December 13, 2005 at 16:40:27, Ryan B. wrote: >On December 13, 2005 at 15:51:41, Vincent Diepeveen wrote: > >>On December 13, 2005 at 14:35:27, Ryan B. wrote: >> >>>On December 13, 2005 at 08:35:07, Chrilly Donninger wrote: >>> >>>>On December 13, 2005 at 06:08:49, Ryan B. wrote: >>>> >>>> >>>>> >>>>>Does the standard version of Hydra use any foward purning other than Null move? >>>>> >>>>>Ryan >>>>Everybody uses nowadays additional pruning. Otherwise the big search-depths >>>>would not be possible. Hydra uses them too. But so far not History-Pruning. Did >>>>not work with the other pruning techniques. >>>>The kind of pruning is a secret. One big error - publishing NullMove - is >>>>enough. But NullMove was anyway known in the community. >>>> >>>>Chrilly >>> >>>I suspected that most strong programs use additional pruning. I asked because I >>>read a claim by Vincent that Deip and Zappa only use Nullmove for pruning. >>> >>>Ryan >> >>Diep world champs 2005 version used history pruning. Normally it runs without >>pruning. dutch champs it used no pruning other than nullmove. >> >>Zappa uses only nullmove. The world champs version of zappa was just nullmove. >>I'm not sure whether he's gonna use pruning in future, you'll have to ask >>himself. >> >>Diep is not going to use history pruning *ever again*. Last plies pruning or >>something else i might retry however. >> >>Did do a lot of tries there. Point is, when i play a 100 games or so and the >>statistics indicate it scores me less points, then i kick out something. >> >>That happened to history pruning too. The tournaments we play, we play at quads >>or something big, that means that testing a 100 games is not easy, as >>in my case i simulate it with 180 0 games at a dual k7, each program its own >>dual k7. >> >>Vincent > > >Interesting, in my experience other than null move most pruning methods are only >good at blitz time control or on slow hardware so I am not surprised that Diep >only uses null move as I doubt you are interested in blitz results. history pruning is not good for fruit only at blitz. I believe that lower values than 70 are better at long time control but I believe that it is also productive at long time control. One that I >think is only a small improvement but helps at all time controls for me is >extended razoring with researches when value is over alpha. A more complicated >algorithm for doing researches may help but I have not found it yet. > >Ryan In my experience history based pruning is one of the most logical pruning because it is based on statistics. If a move never failed high and failed law a lot of times then it is logical to prune it. Note that to be more correct it is the only reduction that I use. I have pruning based on evaluation and other factors that I use in the last plies(I do not give exact details). History based pruning usually only save me 1/2-1 plies Uri
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.