Author: KarinsDad
Date: 09:00:29 06/09/99
Go up one level in this thread
On June 09, 1999 at 04:48:18, Ricardo Gibert wrote: >Deriving statistics from endgame databases can be very misleading. This would >not settle anything. Giving every position in the ETB equal weight is an >arbitrary decision that is unwarranted. Garbage in, garbage out. All the >positions are not equally likely. It is quite conceivable that some positions >in a database never occur with correct or reasonably correct play. > >I have a vague recollection of someone (Nunn?) citing a statistic about RP+R vs >R endings. Roughly the same proportion of positions were winning as in CP+R vs >R endings (C=center). Quite contrary to accepted experience that the CP is much >stronger than a RP in such endings. One case where experience is more reliable. Hence, the reason I used the words "might be". To me, it seems totally reasonable that a perfect tablebase would kick butt on every game. It would be like God playing checkers against a monkey. No contest. However, the point I was trying to make was that there are more drawn 3 piece endings then wins. This is easy to show since you are talking about 2 kings and a piece. If the piece is a bishop or knight, the ending is always drawn. If the piece is a rook or a queen, the ending is won UNLESS the rook or queen is next to the enemy king and it is not protected, in which case the king takes the piece and it is a draw. Since there are approximately the same number of positions of kbk, knk, krk, and kqk (the number of each would be exactly the same if not for side to move being the same as the side with the piece and the opposing king is in check; this are illegal positions) and since krk and kqk can have draws and kbk and knk are nothing but draws, it appears that there are more draws than wins for 3 piece positions. Having said all that, the point I was trying to make is that as you add more pieces, the chances of winning may increase (i.e. there may be a higher percentage of wins in 4 piece endings and an even higher percentage of wins in 5 piece endings). The only data we have to go on is whether this is true or not (and yes, I realize that there are 4 piece endings that are almost always draws such as kbkn). However, if it is true, a postulate can be formed that: the more pieces you add, the higher the percentage of wins until you get to the point (with 32 pieces) that with "perfect" play, there are nothing BUT wins unless the opponent also plays perfect. Another way of looking at it is that if you write a chess program that just looks at a material evaluation, just tries to push pawns, and just searches 6 ply down, this program will play in the ballpark of 1300-1400 elo chess. When an 800 elo player plays against it, he will very rarely win since the program does not make a tactical mistake within 6 ply whereas the player does. When you create a perfect tablebase program, it does not make a tactical mistake down 150 ply (or so). The best players in the world would not be making a tactical mistake down 10 ply or so. But, the best players in the world would be making strategic moves (i.e. moves which appear to give a tactical advantage later in the game). If these strategic moves are not perfect, then they could result in a tactical error 12 ply down, 20 ply down, or even 130 ply down. It is the same as the 800 elo player playing the 1300-1400 elo program (or even a better example is the 800 elo player playing Deep Blue). He doesn't have a prayer of a chance since he will always eventually make a mistake in the game. Statistically speaking, if there are an average of 3 good moves (i.e. GM level) on the board at any time and only 1 of the moves is perfect, in a 60 move game, a GM would have about 1 chance in 4.24e28 of playing a perfect game. Kasparov being the best GM in the world may drop it to 2 good moves or 1.15e18. These numbers are so astronomically large that the human mind cannot even grasp them(except in abstract terms). This is similar to playing the lottery and winning it 4 times in a row, 3 times in a row for Garry. If this does not sound impossible to you (for all practical purposes), flip a coin 60 times in a row and get tails everytime and let me know when you manage it. KarinsDad :)
This page took 0.02 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.