Author: Clive Munro
Date: 11:36:58 11/12/04
Go up one level in this thread
On November 12, 2004 at 12:36:04, Uri Blass wrote: >On November 12, 2004 at 12:27:09, Clive Munro wrote: > >>On November 12, 2004 at 10:49:17, Uri Blass wrote: >> >>>On November 12, 2004 at 09:14:26, Clive Munro wrote: >>> >>>>On November 11, 2004 at 11:57:01, Uri Blass wrote: >>>> >>>>>On November 11, 2004 at 11:34:35, Derek Paquette wrote: >>>>> >>>>>>On November 11, 2004 at 08:17:36, Uri Blass wrote: >>>>>> >>>>>>>On November 11, 2004 at 08:11:03, Uri Blass wrote: >>>>>>> >>>>>>>>On November 11, 2004 at 07:44:01, Andrew Platt wrote: >>>>>>>> >>>>>>>>>On November 11, 2004 at 02:15:20, Derek Paquette wrote: >>>>>>>>> >>>>>>>>>>I play email chess all the time, aswell as blitz games on the net. I also play >>>>>>>>>>engine matches online and this came across. >>>>>>>>>> >>>>>>>>>>If someone were to play an email game, say a Grandmaster vs Me, and I was to use >>>>>>>>>>a computer, and the return move time was say 5 days a piece. If I were to keep >>>>>>>>>>Shredder 8 on Infinite analysis for those 5 days per move (say i keep the cpu >>>>>>>>>>very cool and i give it a few hours of break) >>>>>>>>>> >>>>>>>>>>would shredder 8 play a perfect game? or near perfect where the gm would have no >>>>>>>>>>chance without the assistance of another computer? >>>>>>>>>> >>>>>>>>>>I know that the longer the time control the better for the computer but to what >>>>>>>>>>end? >>>>>>>>>> >>>>>>>>>>with 5 days per move, that is 120 hours,(7200min) >>>>>>>>>>using the starting position as a bench mark for depth vs time, (where shredder 8 >>>>>>>>>>goes 1 more ply after it doubles its time searching) >>>>>>>>>> >>>>>>>>>>I calculate on my athlon 1700xp that after 5 days shredder 8 would reach >>>>>>>>>>.... >>>>>>>>>> >>>>>>>>>>depth 21 in 27 minutes...knowing that... >>>>>>>>>> >>>>>>>>>>depth 22 in 54 min >>>>>>>>>>depth 23 in 108 min >>>>>>>>>>depth 24 in 216 min >>>>>>>>>>depth 25 in 432 min >>>>>>>>>>depth 26 in 864 min >>>>>>>>>>depth 27 in 1728 min >>>>>>>>>>depth 28 in 3456 min >>>>>>>>>>depth 29 in 6912 min >>>>>>>>>> >>>>>>>>>>So basically shredder 8 would reach a ply of 29 in just under 5 days... >>>>>>>>>>could ANY human EVER beat it EVER? without computer assistance? >>>>>>>>> >>>>>>>>>Yes of course a human could beat it EVER. If it's a quiet position, the human >>>>>>>>>could use those 15 moves to significantly improve their position while Shredder >>>>>>>>>flounders around a bit. >>>>>>>> >>>>>>>>Shredder can also use the 15 moves to improve it's position. >>>>>>>> >>>>>>>>> >>>>>>>>>What would work much better, and what I do for post-game analysis of my games, >>>>>>>>>is to let Shredder think for a while and then move, let it think, move and see >>>>>>>>>what happens to the position. Or let it play against itself. I often do that, >>>>>>>>>then find improvements for Shredder, get it to try that out. >>>>>>>> >>>>>>>>Shredder could find a lot of these improvement if you give it more time and it >>>>>>>>can also find improvement that you did not find in that way. >>>>>>>> >>>>>>>>Uri >>>>>>> >>>>>>>I can add that 29 plies of shredder does not mean that it can see everything in >>>>>>>the next 14 or 15 moves of both sides. >>>>>>> >>>>>>>It is obvious that shredder does a lot of pruning to get that depth. >>>>>>> >>>>>>>Uri >>>>>> >>>>>>Yes it definately does pruning, but what is your thoughts on a human beating it >>>>>>at those time controls? 5 days per move. (no computer assistance for the >>>>>>human) >>>>> >>>>>I guess that inspite of the pruning No human will beat it with no computer >>>>>assistence at 5 days per move. >>>>> >>>>>I do not claim that no human can do it but the humans who can do it have no >>>>>reason to spend time on it because they are smart enough to use their time >>>>>better(for example toearn money in over the board game). >>>>> >>>>>Top correspondence players have hard problems against computers even when they >>>>>use computer help and using computer help is usually allowed in correspondence >>>>>games. >>>>> >>>>> >>>>>Uri >>>> >>>>I agree with Uri, I think that any chess computer/program will only play to a >>>>certain level. Even given that amount of time if the program has pruned a good >>>>positional move early on, it can only see the consequences if it were on a brute >>>>force search. And then only if the program were clever enough to recognise it. I >>>>bet Shredder would not play much better over 5 days a move than 1 hour a move. >>> >>>I am sure Shredder will play much better with 5 days a move than 1 hour a move >>>unless it has some bug. >>>Based on experience computers find better moves with more time and there is no >>>limit. >>> >>>Uri >> >>The computer is only as good as its program. Take for example an older computer >>program say Frita 3. I doubt whether it will play much better than its rating of >>about 2400elo given a week per move! The program would ignore better moves not >>because it can’t see them but because its program would think that they are not >>such good moves. > >If it searches deep enough it may see that they are good moves. > > I very much doubt that Fritz 3 would beat Shredder 8 if >>Shredder had 1 hour a move and Fritz 3 5 days a move. > >I also doubt it but the reason is that the gap between Shredder8 and Fritz3 is >huge. > >If you take Shredder8 and something that is 100 elo weaker than I expect the >weaker thing to win at 5 days against 1 hour. > >Note I do not claim that there is no diminishing returns but I believe that the >returns from doubling the speed is at least 30 elo even at correspondence time >control and 5 days is 120 times 1 hour so I expect difference of 200 elo between >shredder8 at 5 days per move and shredder8 at 1 hour per move. > >The difference between shredder8 and Fritz3 is bigger than 200 elo. > >Uri Yes I agree maybe that was an unfair example. And I would expect Shredder to play better given more time. However there is a limit even for Shredder at which doubling its time would not improve its chess. This is much more evident with slower dedicated m/cs. Clive
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.