Author: Bob Durrett
Date: 16:16:25 01/03/04
Go up one level in this thread
On January 03, 2004 at 17:32:42, Odd Gunnar Malin wrote: >On January 02, 2004 at 11:32:28, Bob Durrett wrote: > >>On January 02, 2004 at 02:13:51, Odd Gunnar Malin wrote: >> >>>On January 01, 2004 at 20:55:49, Bob Durrett wrote: >>> >>>> >>>>Suppose someone wanted to produce a chess-playing program which emulated the >>>>play of a human rated 1200 to 2200. >>>> >>> >>>Hi. >>> >>>I'm a bit focused on how to help a 1200 player etc. instead how simulating one. >>>Eg. the program should setup position/combination of a type hver a certain level >>>player have problem with. Maybe configurable (or automatic adjusted) for which >>>types to set up. >>> >>>A sample. >>>For a beginner in the childgroup there are three overwhelming errors. >>> >>>1. Don't see own pieces in price. >>>2. Don't see opponent pieces in price. >>>3. Leave a piece in price because of the move made (most common). >>> >>>Of course you have counting errors and so on, but if you look at one game these >>>take the cake. >>> >>>Here a playingprogram allways have to take a piece left in price no matter how >>>low it is adjusted. Only by this you can help a player with the error 1 and 3. >>>It could (should) also leave it's own piece in price occasionaly, here maybe >>>some types of misses more than others. >>> >>>Odd Gunnar >> >>I, too, am interested in helping chess beginners get started. As it happens, I >>am an also an avid reader of printed [hardcopy] chess books and have read a >>number of them devoted to the teaching of beginners. My most recent acquisition >>of this type was "Everyone's 2nd Chess Book" by Dan Heisman, copyright 2000, >>ISBN 0938650556, Thinker's Press, Inc. >> >>Dan lists "the eleven most common mistakes of players rated 800-1400" as >>follows: >>(1) Missing a simple tactic >>(2) Not looking at what your opponent is threatening >>(3) Not getting all pieces into play >>(4) Not knowing basic opening traps >>(5) Phantom fears >>(6) Overly worried about the value of the pieces >>(7) Overly worried about positional liabilities >>(8) Playing too fast >>(9) Not looking for a better move >>(10) Being afraid of the opponent >>(11) Worrying about your rating >> >>The problem with the above list is that it is not easy to see how to program a >>chess engine to have these faults. In fact, that is probably impossible. >> > >I don't have his book but I have read his column at ChessCafe. >It is a good place to get input for a coach, but for (young) beginners the >language barrier for none nativ english speaking is too high. A year back I >offered him to translate his column to Norwegian but got a negative respons. There must be good coaches in countries which speak Norwegian. Dan is good but not unique. You may be one of them! > >Your (his) list above is better for an adult beginner or at least one who have >passed the first stage of 'ludo' playing, I think he called it awareness in one >of his column. He also spent some time describing the problem you describe, but said it was found more in players below the 800 level. > >If you get the beginner to think through the 3 points I mention above before >each move he allready have taken a big step. >1. Any of my pieces in price. >2. Can I take an opponent piece for free. >3. When I do the move I thinking of, can my opponent take any free pieces. If one wished to create a "black box" which made move choices which were consistent, statistically, with the above three situations, then that "black box" might be useful as a sparring partner for the purpose of providing the player with the opportunity to spot and punish such errors. For example, the "black box" could be programmed to occasionally overlook the fact that one of it's pieces is enprise. The probability of occurrence of this event might be adjustable so that improvement of the human learner could be matched by reduction of the probability of the occurence of this event in the "black box" output. Similarly, the "black box" could be programmed to occasionally fail to capture pieces which were enprise. The probability of occurrence of this event might also be adjustable so that improvement of the human learner could be matched by reduction of the probability of the occurence of this event in the "black box" output. Finally, the "black box" could be programmed to make moves which allowed the human to capture an enprise piece. The probability of occurrence of this event might be adjustable so that improvement of the human learner could be matched by reduction of the probability of the occurence of this event in the "black box" output. In other words, it should be possible to design a chess-playing program which satisfied the above three requirements. The heart of this chess-playing program might be a very strong engine, but programmed to play this way. > >When looking at games, this could get one player from the bottomhalf of the end >table to top the list. > >One way to get some knoweledge for what kind of errors is to look at games >played by the beginner. >At our club we have got the beginners to write down their games. They most play >rapid or blitz games, but for rapid ( >=30 min) there is enough time for they to >write the moves down. I believe that it is essential that the game scores of weaker players not be mixed up with game scores of stronger players because the CHARACTER and rates of occurrences of the errors will be heavily dependent on the ratings of the players. >I have put up some games at the WEB: >http://norbase.sjakk.biz/index.php?page=base&lang=en&inter=14&subpage=search&fdate=2003&siteid=22&eventid=50&action=view >and here: >http://vadsosjakklubb.org/arkiv/Default.asp?id=138 >The last link is in Norwegian but the games are clickable in the first table. >On both pages you can get the pgn-file with the link under the list. >These are not selected games so it would give an average of errors in those two >tournaments. > >A little sidenote is that when I put it on the web I got response that more of >the players looked through (analyzed) their games. Maybe clicking on the >Internet is more fun than draging out a chessboard to play through the game. > >>Instead, one might determine how each of the above "most common" mistakes would >>affect the selection of moves. From that, it might be possible to model >>[statistically or otherwise] the move outputs of the "players rated 800-1400." >>In other words, what is desired is to create a "black box" who's chess games >>could not be distinguished from the chess games of "players rated 800-1400." >> > >With the exception that the focus should be to help, eg. allways punish an >error. Please elaborate. > >>Perhaps attempting to use Heisman's list would not be the best way to model the >>play of a chessplayer. Instead, just obtaining and analyzing a collection of >>games played by people of the same rating might be better. The types of >>mistakes could be determined, their rate of occurrences noted, and then a >>statistical model could be produced to represent the player. Two parameters, at >>least, would be important: (a) rating of the player and (b) time limits on the >>games. Generally, mistakes are much more prevalent for fast games than for slow >>games and the nature of the mistakes is also dependent on the speed of the game. >> > >The timefactor are more a problem for older players (still talking about >beginners) above 20. When observing youngsters the problem is to get they to use >their time. In a 45 min. game you seldom sees anybody get less the 15 minutes on >their clock. I have observed the same phenomena. Attention span is a problem for the very young. This is discussed by Dan Heisman in his book too. If a player plays a complete game in less than 15 minutes, then that would be Blitz. > >I saw in Jesper Hall's book (I think) a method to learn that is adaptable to a >playing program. The method is simple, the board should be 1 move behind the >actual game. After some practice this could be 2 moves and so on.. >- White move 1.e4 on the scoresheet eg. tell the program. >- Black move 1.e5 on the scoresheet. >- White moves 2.Nf3 on the scoresheet and 1.e4 are updated on the board. >By this the player is allways in 'analyze' mode. An interesting and novel idea. > >>I feel that human chessplayers, regardless of rating, can benefit greatly from >>play against other human chessplayers who are at about the same strength or >>maybe slightly stronger. The reason is the same as that which I mentioned >>earlier. It is beneficial to learn how to see and punish the opponent's >>blunders and other errors. A game cannot be won if the opponent plays >>perfectly. If one wishes to gain skill in spotting and exploiting errors, then >>it is necessary to play against an opponent who makes such errors. Beginners >>have no chance of seeing and exploiting the types of errors made by Kasparov, >>but they can see and exploit errors of the type made by other beginners. The >>top GMs need to play against opponents who make "GM errors." > >Yes, playing against human opponents is better and of course even more 'better' >if the opponent is stronger than the player because he would punish your >error. But if too strong, he will not make punishable errors simple enough for the student to recognize. The student would get no practice at spotting and punishing errors. >One problem is that playing only one day in the week at the club isn't enough to >get progress. He have to play some more games and a little more attention from >the programmer that their program are a training tool could maybe help. >Chessmaster *) have done some job here with their courses and created low-rated >personalities. Though I would have liked it if they took away the lowest >personalities. They don't punish and they create another problem that is hard >(and frustrating) for a beginner being the stalemate problem. > >*) I haven't seen the 'Fritz and Chester' program. > >> >>A strong chess engine [Fritz, Hiarcs, Shredder, Tiger, Junior, Crafty, >>Chessmaster 9000, etc.] can help the beginner to see his/her mistakes and >>oversights during post-mortem analysis. If a beginner does not realize that >>he/she has made such errors, he/she will keep on playing such errors and be >>"dumb, fat, and happy" and never get much better at the game. My opinion is >>that revisiting positions which the player previously failed to solve, and then >>making a second attempt to solve those positions would be educational. On the >>other hand, just looking at the solution as presented by a chess engine [or a >>human] would have little benefit. > >I don't like to use any computer for analyzing, it would spoil you 'best' >chessbook you can get that is your own games. These games should be analyzed >over and over and eventually they have found their errors. That may not be a realistic expectation for the very young who have limited attention spans and have not yet developed the necessary self-discipline. >When you get a bit past the absolute beginner you can start looking at your >opening in the game with help from books (or computer) to widen your >openingreportuare one move at a time. > >> >>A human trainer is clearly better than a chess engine, at least today. Maybe >>tomorrow [in the future] than may not be true but that seems very unlikely. We >>must wait for HAL! : ) >> > >A human trainer would probably allways be best, so much fait don't I have in >computers. > >Odd Gunnar
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.