Author: Uri Blass
Date: 06:57:18 11/16/03
Go up one level in this thread
On November 16, 2003 at 09:29:07, Derek Paquette wrote: >On November 16, 2003 at 02:21:32, Uri Blass wrote: > >>On November 16, 2003 at 01:29:52, Timothy J. Frohlick wrote: >> >>>To get to 14 ply with 11 to 12 good moves per ply would require a million >>>billion choices to be searched. >> >>11 to 12 good moves per ply is only your imagination. >> >>alphabeta means that the program usually searches only 1 option in the even or >>the odd plies. >> >>Even the branching factor of tscp is smaller than 11 and I think that it is 7 >>and having better order of moves including using hash tables should help >>significantly. >> >> >> There are only 86,400 seconds in a day. Full >>>exhaustive searches to 14 ply could not be done on todays PCs. >> >>Wrong. >>I believe that branching factor of 5 is possible to achieve even without >>pruning(suppose even that only 6 is possible) >>6^14<10^11 >> >>if you search 10^9 nodes per second you can get 14 plies brute force in 100 >>seconds(not that I think that it is good strategy). >> >> >> Only 31 moves >>>could be made in one year if a machine were searching at billion positions/sec. >> >>I do not understand it. >> >>> >>>Selective search which involves massive pruning of the search tree chooses only >>>three to five best first moves and examines the best responses again >>>selectively. Much less computer power is needed. >>> >>>Most GMs select the one best move depending on their analysis of the board >>>position and their memory of similar/same positions. The one best move approach >>>also depends on the attacking plans of the GM. >> >>No >> >>GM's look at more than one move and at more than one line when they analyze. >> >>> >>>The program that the article dreams about is not similar to today's PC programs. >>>It does not filter most of the choices in the search tree. >>> >>>TJF >> >>I do not know about which article you are talking. >>The poster said that there is an article and did not give a link and I am too >>lazy to search for it. >> >>> >>> >>>On November 15, 2003 at 23:47:04, Derek Paquette wrote: >>> >>>>This is a very ignorant question coming from me, >>>>but I'd love to hear the answers, it is bugging me. >>>> >>>>Ok, hypothetical question, Deep Junior 8 is playing against kasparov... >>>>it is a difficult board position, around 7 ply the computer should be coming >>>>across the correct move, there is only 1 correct move to play without a lose >>>>along the road... >>>>now if DJ8 is filtering at 99.99999% of the moves, >>>>why would it matter if it had quad 2.8ghz chips, or even 8 chips... >>>>if its not seeing the move, why would it at 22ply suddenly see it? >> >>Why not? >>There are moves that you need many plies to see them. >> >> >>>> >>>>on the x3d site there is an excellent article, and it says, a definate way to >>>>beat a super grandmaster is to build a machine running at 1 billion positions a >>>>second, and have it search to only 14ply, making thoroughness over filtering and >>>>deep looking a priority... >>>>so can someone explain to me why faster hardware makes a difference, if even my >>>>home pc can look at ply 18 with deep junior... >> >>When Junior says depth 18 it does not mean 18 plies. >> >>9-36 plies dependent on the line is more accurate. >> >>Uri > > >If the machine is filtering so much, what does extra hardware do for it neway >if its just not going to see the position? Why after 5 minutes would it see the >right position that it should see after say 7 ply, when in 30 seconds it >couldn't...it already left that move possibility in the dust... No pruning that is done today is often not what moves not to consider but what moves to search to reduced depth. If the program at depth x searches part of the moves to depth x/2 then it may see something that is 7 plies only after 14 plies. Another point is that bigger depth is always important and 15 plies even with brute force is more than 14 plies(there are tricks that you even need 30 or 40 plies to see and even with extensions 14 plies is not enough(for example the draw that kasparov missed against deeper blue)). >that is my question, >I see it all the time, after 13 minutes a program finds such and such a >move...but the actual move was only 7ply deep..but it took 45 minutes to find >it. >etc. > >I don't understand...does a computer keep looking back and going through >different moves from the start if it doesn't find anything positive?? The computer always looking at the same moves to bigger depth when it starts a new iteration. 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.