Author: Vincent Diepeveen
Date: 08:40:14 05/17/05
Go up one level in this thread
On May 16, 2005 at 23:06:29, Darrel Briley wrote: The main problem by letting computers invent their own openings is that todays programs without opening book still are utmost beginners. It is like a race car with bicycle tires. The idea behind this would be that less resistance to the ground allows a higher speed of driving, as the car has to do it itself, instead of let the slicks work for it. In theory it would be able to do 400 kilometers an hour still, but you slip at every corner, most importantly the first one, and your starting position is real bad as the start goes very slow thanks to lacking grip. A big effort has been done by Dann Corbitt a few years ago with his Crappy Analysis Project. If i may recall the idea was that a book reinforced with scores from chess engines at all nodes, would create a good book. If i remember well i was one of the few directly writing down here it was not going to work. Several commercial authors also were a bit negative surprised that it worked *that* bad. The project died a slow death. Silent, out of sight, as the resulting book didn't kick butt. In short, there is big statistical evidence that every book, without human intervention is sucking more than a book with human intervention. That said i hope you realize one shouldn't confuse this with book learning. Book learning is very effective in that you can repeat your last win and can avoid losing/drawing in the same manner. But trusting the computer on his own from the first move, is a kind of suicidal thing to do. So to speak the computer is 1800 rated in opening, it's 2400 rated in middlegame, it's 2200 in endgame and it's tactical 4000 rated. Of course never making a tactical mistake soon lays the overall border to 2500-2600 when the weak chain called opening is removed by using a book. >On May 16, 2005 at 21:28:21, Komputer Korner wrote: > >>Now that chess engines are rated equal to top GMs, why can't some chess >>programmer write a script to enable a copy of his engine Ex: >>Fritz,Shredder,Crafty to analyze every position (where there is more than 1 move >>choice in the book) let us say for 2 minutes each starting at the beginning of >>the book. Then it would automatically put in the numerical annotation in the >>opening book. After a couple of months or so, X number of plies of opening moves >>would be completely annotated(depending on the size of the book. Or better yet, >>the openings could be split up and one computer each could be used on each >>opening and then at the end they could be combined. I am assuming that the books >>and combined master book would catch all transpositions like the ChessBase and >>Bookup opening books do. Then the programmer or company could sell or give away >>this annotated master book which the customer could change with new information. >> Or has this already been done. I know that there are lots of small books that >>have been done but what about God's opening book? > > > >I've been thinking for some time along similar lines. Why isn't the opening >book treating in a similar fashion to the hash tables (I.E. the final book >position annotated with the score, depth, result) and these annotations would >not be discarded from game to game, but would be maintained and built upon, in >this way it seems there could be continual improvement on the engine's book. At >first only the final book position would/should be annotated, but in subsequent >usage, additional moves/positions could be added to the book; one ply at a time. >If such "book" parameters could be adjusted by the user so that an already seen >position would only be analyzed if it did not meet a certain depth/result, I >think it would be invaluable in improving a poor book, and could make an already >strong book better. I realize this would entail a book becoming larger, but with >today's storage soloutions this doesn't seem to be such a major consideration. >Taken one step further... > >I'd like to build a PC that would be a dedicated chess computer, and ideally the >program's book handling would operate as already mentioned, but in addition to >this, idle time on the computer would be spent analyzing the book, and the games >the machine has played. This could be focused by the user, or could operate >fairly autonomously, at the discretion of said user through selectable menus. > >I'm not a programmer, so I'd appreciate any input/thoughts on the utility of >such an approach. I realize that book learning already does some of this >already, but the ability to add moves/positions to the book isn't currently a >feature that I'm aware of in any program. Also, the ability to have the >computer do analysis during idle hours, either automatically, or through a >user's guidance (but still largely automated, so as to alleviate some of the >tedious attention to detail required) would be a wonderful feature IMHO. > >It's not hard to envision other aspects of such a system. The engine would be >continually tailoring the book to maximize its own strengths, and to minimize >its weaknesses. Also, if the program had the capability to recognize unique >opponents, and to keep a record of the opponent's openings and results, the book >learning would be so much more effective. Any thoughts?
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.