Author: Odd Gunnar Malin
Date: 00:40:10 05/02/05
Go up one level in this thread
On May 01, 2005 at 08:00:28, Uri Blass wrote: >On May 01, 2005 at 05:52:27, Odd Gunnar Malin wrote: > >>On May 01, 2005 at 01:22:07, Uri Blass wrote: >> >>>On April 30, 2005 at 19:52:25, Odd Gunnar Malin wrote: >>> >>>>On April 30, 2005 at 15:07:16, Derek Paquette wrote: >>>> >>>>>The term 2500 fide elo positional and 2900 fide elo tactical still comes to mind >>>>>when i think of people's opinions regarding programs, how does shredder 9 fair >>>>>against this criticism, >>>>> >>>>>does it show more understanding of the game than past engines? or is it still >>>>>lost in many positions and still getting hopelessly outmanouvered on ICC by >>>>>grandmasters. >>>> >>>> >>>>Hi. >>>> >>>>I do some lecture from ImproveYourChess.com to get in shape for the summer >>>>tournaments and one of this lecture is to solve 'a guess the move' type of test. >>>> >>>>To see what an engine manage in these test I have run my favourite analyse >>>>engines through the test I have done myself and here is the score. >>>> >>>> Shredder 9 Tiger 2004 Gandalf 6 Me Max >>>>Game 1 39 (75%) 28 (54%) 34 (65%) 32 (62%) 52 >>>>Game 2 42 (67%) 38 (60%) 41 (65%) 28 (44%) 63 >>>>Game 3 42 (63%) 31 (46%) 30 (45%) 31 (46%) 67 >>>>Game 4 46 (53%) 46 (53%) 34 (39%) 42 (48%) 87 >>>>Game 5 45 (78%) 40 (69%) 34 (59%) 32 (55%) 58 >>>> >>>>The computer was a 2GH Centrino laptop with Keep hash/learning etc. on and 512MB >>>>hashtable size for each. The computer got 3 minutes for each move. >>>> >>>>As you see on this table Shredder 9 scored highest in all games. Thoug, talking >>>>about 2500 elo in positional strength seems a bit away from the thruth. I added >>>>my score (Me) to the table to compare, and my rating is 1500. >>>> >>>>A rough estimate of Daniel Kings suggestion for the score is: >>>> >>>>>90% GM >>>>>75% IM >>>>>60% FM >>>>>50% NM >>>>>35% Strong clubplayer >>>>>20% Average clubplayer >>>> >>>>But since I'm a Average clubplayer these scores are most to please the pupil. >>>> >>>>Odd Gunnar >>> >>>I think that Daniel king suggestions may be wrong in part of the cases and the >>>moves that he suggests are not the best. >>> >> >>It would be nice if you back up your accusation with some samples or is this >>pure speculations. > >I do not have the positions so I cannot give more than speculation >but I know based on analysis of other notes by GM's that it is possible to find >mistakes by a computer analysis. GM King have made these tests some kind of his main learning method an have published at least two book on the subject. I would guess that if he didn't check the moves somehow he would get a lot of negative feedback from users. > >> >>>I doubt if he used shredder to check his analysis in order to see if there are >>>cases that there are alternative moves that he missed or there are moves that he >>>did not evaluate correctly. >> >>Doesn't they all use Fritz nowadays, so I doupt too that he checked with >>Shredder. > >Shredder is better than Fritz. I only have older copies of Fritz (ver. 6) so I can't compare, but I think you have too much faith in the engines. While a positional move isn't too much different from a tactical move, only the score is a bit lower, it seems much easier for an engine to push such position over the horizont. My selection of analysis engines includes Shredder, Gandalf and Tiger. There are pluses and minus with them all. One minus for Shredder is its habit to repeat moves, the search probably only give drawscore on the 3.-rep. This is probably good (but not polite) in games, but when analyzing it is a bit annoying, and put the horizont even farther away. To thrust that an engine give you the best move in a non-tactical position is a bit optimistic. If let say they give you the best move in 75% of the positions, how could you know that in current position the move are nonsense? You always need to work with the position and suggest moves/variation for the engine to try out. When working with the three engine above you even manage to 'convince' them about a move or variation, this is the cause of my selection of engines (plus that they fit smoothly into Chess Assistant). > >> >>I have done this type of tests in the past and what they have shown is that if >>you double the thinkingtime the score goes up (you will probably find these with >>a search). > >It may be that 10% of the moves are wrong and they will never find them. > >Of course if 90% are right then more time can help the computer to improve from >70% to 80% so if you find that programs score better with more time it does not >prove no mistakes in the test. > > For now, I don't have time to do this on these tests. I only run them >>to give me a competitor so I think a little harder to find the correct move. >> >>To quote Bent Larsen in his book of this type if a GM took the test: >>"They could of course have complaint about the score, and had to agree with them >>in a few times." > >This is another possibility and it is possible that the score for GM result is >simply wrong. > I think it was GM Yermolinsky in his 'Road to Chess Improvement' who said something like: "Apart from the opening choice, 95% of the moves in a GM game is tecnique. This leaves only 5% to the style/mood etc." Odd Gunnar >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.