Author: blass uri
Date: 04:03:08 04/27/98
Go up one level in this thread
On April 27, 1998 at 01:11:22, Komputer Korner wrote: >Oh No, here we go again, another challenge to the mathematical solidity >of Alpha Beta. I would suggest that any poster that thinks there is a >better way to search, please read up on alpha beta as very few non >programmers really understand Alpha Beta with null window searches. >There is a good article on this in the latest ICCA journal March 1998 >titled " Are there practical alternatives to Alpha Beta?" > >-- >Komputer Korner I do not say not to use the Alpha Beta algoritam but to do also other things I am sure that building a sensible tree and using the Alpha Beta algoritam for every leaf is better than the Alpha Beta alone. you can use the alpha beta algoritam to generate the relavent tree I do not need to understand the Alpha Beta algoritam to know that programs use it to find the next best move if I ask for it and to know that you can calculate for a short time the next best move of the opponent after every legal move and to find the best move if the opponent go wrong next move. so I can decide this move will be in the sensible tree I generate. Uri > >On April 25, 1998 at 15:24:51, blass uri wrote: > >> >>On April 25, 1998 at 14:36:49, Robert Hyatt wrote: >> >>>On April 25, 1998 at 10:48:18, blass uri wrote: >>> >>>> >>>>On April 25, 1998 at 09:02:46, Robert Hyatt wrote: >>>> >>>>>On April 25, 1998 at 05:39:58, blass uri wrote: >>>>> >>>>>>1)I think that if a computer program "thinks" it stand better then it >>>>>>should "think" more to accept a move and to check more the >>>>>>possibilities of the opponent. >>>>>>did some programmer try this idea? >>>>> >>>>>what does taking more time accomplish??? other than to get you into >>>>>time trouble later in the game.. >>>>> >>>>my idea was not to waste more time about a move but to waste more time >>>>about accepting moves and less time about rejecting moves. >>>>It is important when the computer evaluate it has adventage because >>>>in this way the computer can see more quickly if its evaluation is >>>>wrong. >>>>If its evaluation is right it is not very important if it miss something >>>>better >>>>because the result may be that it win more slowly >>>> >>>>> >>>>>> >>>>>>2)did a computer programmer try to find the probability his(her) >>>>>>program changes its mind if he(she) double the time of it >>>>>>as a function of time? >>>>> >>>>> >>>>>find the ICCA Journal article "Crafty goes deep" written by Monty >>>>>Newborn. He analyzes this on a large set of random positions, searched >>>>>all the way out to 15 plies.. >>>>> >>>>> >>>>> >>>>> >>>>>> >>>>>>3)did any computer programmer try to give the computer to calculate >>>>>>what is the best move if the opponent will choose the second best >>>>>>move and decide about this move that it should be analysed? >>>>> >>>>> >>>>>again, what would be the point? If your "best" move prediction isn't >>>>>very accurate, that needs to be fixed, rather than trying to >>>>>second-guess >>>>>yourself and waste time searching things you think are second best.. >>>> >>>>the point is if the "best" move evaluation is correct but there is >>>>something better for example I see that programs do not devote >>>>enough time to Chaos move in the 1974 computers championship >>>>against Chess4.0 Nxe6. >>>>I checked that after the forced line Nxe6 fxe6 Qxe6+ Be7 Re1 >>>>my programs evaluate the positon after some minutes as adventage >>>>for white while before Nxe6 the position is evaluated >>>>as advantage for black. >>>>it is clear that if white can tell black not to play some move Nxe6 is >>>>best >>>>so Nxe6 should be checked >>> >>> >>> >>>Note that for every position where Nxe6 is correct, there are 1,000 >>>positions where Nxe6 loses outright... so trying to pick up such movs >>>is a speed issue, *not* a time spent issue... I would not want to waste >>>time on such moves at the expense of other things in the search... >> >>if you give the computer 3 minutes per move you are probably right >>but if you give the computer 3 hours per move I believe that it is >>better that the computer will use one hour for analysing such moves >>at the expence of other things in the search. >>today 3 hours can be in some years 3 minutes. >>and >>I use the computer for hours in my correspondence games.
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.