Author: Dave Gomboc
Date: 22:01:40 11/28/99
Go up one level in this thread
On November 28, 1999 at 19:40:18, leonid wrote: >On November 28, 1999 at 18:36:46, Dave Gomboc wrote: > >>On November 28, 1999 at 16:30:46, leonid wrote: >> >>>On November 28, 1999 at 15:00:19, Dave Gomboc wrote: >>> >>>>On November 28, 1999 at 14:05:01, leonid wrote: >>>> >>>>>Hi, >>>>> >>>>>For those who spoke about the difference between "legal", "illegal" moves based >>>>>logic. Question was: What logic will be more rapid? After my trial, legal moves >>>>>based logic do the best. Difference insignificant and is around 4%. >>>>> >>>>>What exactly was changed and where? >>>>> >>>>>1) Before only ply zero had illegal moves based logic. All other plys used legal >>>>>moves logics. >>>>> >>>>>2) Were changed three plys over ply zero to see the difference. >>>>> >>>>>3) Tryed both logics for speed 8 plys deep, brute force, same positions. No >>>>>extensions. >>>>> >>>>>4) In illegal moves based logic all old alignment procedures are the same. For >>>>>instance, procedure that put "checking moves" at the head of moves line. This >>>>>last procedure is as much big and time consuming, as the procedure that before >>>>>watched for legality of all moves in a ply. >>>>> >>>>>Leonid. >>>> >>>>I'm not sure why you would force the checking moves to be searched first. Use >>>>your standard sorting order for the moves. Searching checks first is going to >>>>be bad, compared with e.g. searching promising captures first. >>>> >>>>Dave >>> >>>It is not the fantasy setting that I had in my moves alignment. In my usual >>>logic (only ply zero use illegal moves, all other plys legal) this give big >>>speed improvement. Don't forget one obvious fact, you must put at the head of >>>your moves line those that lead to the biggest material advantage ever. And >>>checking moves do right this. Not all the time, and this is where everything can >>>be lost in confusion. Only statistics can clean it up to the "final truth" and >>>offer us its final picture. For me, this was already done. >>> >>>Leonid. >> >>Lines that lead to the biggest material advantage ever usually contain captures >>like "knight takes queen", not "queen to h7, check". If you're solving for >>checkmates, it might be quite reasonable to search checks first. But if you're >>just searching to find the best move, checks are not typically first on the >>list. >> >>Dave > >As strange as it can sound to you, this is the truth. This move ordering is the >best when searching for mate or material advantage. Even if the mate don't >existe for sure in given position, mate will be calculated repeatedly by your >logic. When your moves ordering will be right, all the responses will come at >more expidient pace. > >Leonid. I understand what you are saying. What I am saying is that the move ordering will be better if you search good-looking captures before checks. There are certain things that vary from program to program, but it is my impression that this is not one of them. Dave
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.