Author: Bob Durrett
Date: 05:32:54 11/17/02
Go up one level in this thread
On November 16, 2002 at 23:40:29, Zach Wegner wrote: >On November 16, 2002 at 22:00:27, Bob Durrett wrote: > >> >>I was thinking it might be *fun* to create a machine which does nothing more >>than create legal move sequences from some preset legal chess position. These >>sequences might be dumped into a large part of RAM for later copy to a hard disk >>or printout. >> >>The key idea I'm toying with is to represent a chess position by a listing of >>legal moves. Whenever a new move is made [by the person (or thing) playing >>against the machine, or by the machine if it's playing against itself,] then the >>machine would do nothing more than modify that listing (plus copy the move >>representation to a temporary storage place in RAM). The new listing of legal >>moves would then represent the new position. The key idea is to represent a >>position by a listing of legal moves. When a move is made, there is a "from" >>square and a "to" square. Only consequences of changes made on these two >>squares would have to be considered to modify the legal move list. >> >>Then, to make it more interesting, a really fast random number generator would >>be used to select one of the resulting legal moves. If the machine were playing >>against itself, the sequences of moves should be generated very quickly. How >>quickly? >> >>In the beginning, I am only interested in the time it would take to modify that >>listing. The machine could play both sides, removing the need for >>time-consuming input/output. After generating a legal move sequence ending in >>mate, it would then start working on the next legal move sequence. After a >>million or so moves were made, then the time required could be divided by the >>number of moves. That resulting time per move that I'm asking about. Rather >>than worry about the fact that some computers are faster than others, maybe the >>best bet would be to express it as number of clock cycles per move. A modern >>high-end processor should be assumed. >> >>Each sequence would be what two "really dumb" chessplayers would produce if they >>knew how to produce legal moves but knew NOTHING at all else about chess. >> >>P.S. Is there a better way? >> >>Bob D. > >This would be entirely inefficient, and it would take off most of the abilities >of the engine. Evaluating, for instance, would be practically impossible without >making a piece array. The hashtable would not be possible, for different >positions would be much longer move lists, even >100 moves. Then fen positions >would need to have a proof game constructed for it. Fantasy positions, e.g. ones >with more than 32 pieces, couldnt be represented. Also most modern engines are >bitboard-based, which means that more than one legal move is generated at once. >And sequences of moves made by the first legal move on the list would almost >always be repetition draws. Playing a random move could take thousands of moves >to mate or draw. My machine is playing a game against itself by playing random >moves, and it currently has played over 13000 moves. So finding a list from >every position could take a very, very long time. Thank you very much for the commentary. As I get more into the chess engine thesis I'm looking at, and the chess engines book I ordered from Amazon.com, I will become knowledgable enough, hopefully, to understand what you said. I will save this until I am ready to understand it. Right now, I don't know enough to appreciate the points you have made. Thanks again, Bob D.
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.