Author: Robin Smith
Date: 11:29:19 05/17/05
Go up one level in this thread
On May 13, 2005 at 23:49:26, Komputer Korner wrote: "Komputer Korner", Your post is full of errors. See below for a point by point rebuttal. >On May 12, 2005 at 18:31:52, Dann Corbit wrote: > >>Bookups backsolving is basically a minimax algorithm + refutations. >> >>ChessAssistant does the same thing. >> >>It is a very good idea. >> >>IMO-YMMV. >The following explanation will prove that I am the real KK. Compare my answer >with logical opening theory articles and the 10 kommandments I did 5 years ago. >The problem is that minimax is only useful to a computer. Wrong. Minimax is useful to anyone doing tree searching. Chess masters use minimax all the time; they just don't call it that. Instead they say things like "16.Re1 initially looked appealing, but then I saw the killer reply Bxh7+". In essense this is a verbal description of either the masters thinking or the minimax algorithm. >We humans have brains. Some of us do. :-) >All one has to do is put a chess engine to work at any node to get an >evaluation. Usually the evaluation will be less than a pawn. If it is more than >a pawn, then 99.9% of the time the line is busted for one of the sides. I am sure the number is far lower. Just play through some GM games. Many times the evaluation will go above a pawn, even though the position is still drawn. This is especially true in the endgame, but can occur at any phase. Chess engines are not oracles. >If it is >less than a pawn, then the decision for the White side is how much of a pawn >advantage do you need before continuing with the line? The opening advantage is >about .13 of a pawn. The precision of your proclomation is absurd. Are you sure it isn't .14 pawns? Or how about .20 pawns? I have seen claims as high as .33 pawns. >If any subsequent node has an evaluation of more than that >then the line is worth persuing. It could be worth pursuing even if less than that. What if the line has 10 hard to find "only moves" in a row, but if, and only if, those moves are found it leads to dead equality. It might still be very well be worth pursuing. After all, with perfect knowledge every positoin is either win, = or lose. There is no position with .13 pawns advantage other than as a fuzzy and subjective measure of a positions pratical chances. <snip> >We humans don't need a backsolving algorithm to tell us what lines to play. We might, if the opening repertoire we are trying to learn contains 1,000,000 positions! And we enter a new novelty in a line full of transpositions that completely changes the tree. I have seen chess books that give +/- to a certain position, and then give -/+ to another position that white can force into the +/- position. Needless to say, the book author was not using Bookup and its backsolving function. Maybe he should have used it. >The >only time when backsolving would be useful is if by pressing a button one could > have a computer immediately tell us all the drawing lines if Chess is a draw >and all the winning lines if Chess is a win. The only time? Or the only time _YOU_ can think of. I can think of many other times. >Of course that will never happen. >so in the meantime we all keep adjusting our repertoires with results from games >and results from analysis (human or computer engine). So back to the first >sentence. once you have a computer analyze a line(or if you do it yourself) if >you simply put an evaluation against that node at move 12 why do you need >backsolving to fill in all the subsequent nodes by backtracking back up the tree >from move 30 back to move 12? You keep talking about a bushy, tangled tree (an opening database) as though it were a linear rope. Chess openings transpose. They have numerous side branches. When you make one change to the database, it does NOT affect only one line in the database (unless you have a very small, simple database). >Isn't it much better to have the computer analyze >the 12.Ng5 node and then put an evaluation against it than have the 30.Nxg6 node >analyzed and then backsolve back up the tree to 12.Ng5?? What purpose does it >serve to have all the individual nodes in one line attached with evaluations >when those nodes are not really part of the opening? It doesn't matter one way or the other. What does matter is all the side variations and transpositions. You can't just back up in only one line and get the same result. >Openings are >differentiated from the rest of the tree because they are lines that have been >proven over the years under practice to be worth repeating. Anybody who thinks >backsolving has merit is trying to treat chess as if it is one big opening >puzzle with perfect information. No we aren't. We are trying to treat it as though this is the best information we have _so far_. If new information comes along, then that gets added to the tree, backsolved, and voila, we have the latest state of the art theory for that opening. >Of course, openings cannot be solved with backsolving. No one says that they can. Only that it "solves" the opening to the best of our knowledge _today_. This is _not_ the same as Ken Thompson solving 5 piece endings, and no one is claiming that it is. >Even the name is wrong. Backsolving does not solve anything. Sure it does. It solves the minimax problem based on the data and assumptions contained within the database. Of course if you mean completely and forever solve chess, that is absurd. No one thinks that, no one claims that. >It >merely attaches an evaluation to each node based on an evaluation far down the >root of the tree. What use is that when there will be thousands of novelties >subsequently found as side lines which will put that line out of business >anyway. This is exactly what makes backsolving so powerful!! When new discoveries are made, you add them to the tree, backsolve, and voila; you have the latest opening theory completely up to date. >It is much better to analyze the 12.Ng5 node than to analyze the nodes >at move 30. Even GMs who do deep opening analysis have first exhaustively >analyzed the nodes at move 12,13,14,15,16,17,18... etc before they do the nodes >at move 30. Of course. With Bookup and backsolving, as with any computer tool, garbage in = garbage out. If you use the tool foolishly you will get foolish results. But that is not the fault of the tool. >We will never have perfect information and the chess puzzle will >never be completely solved. Of course. But so what. Backsolving was never intended to "completely solve" chess. This is silly. >I would be very very surprised if any GM ever >admitted to ever using the backsolving feature. I am a corespondence GM, and I have used it. >My expose of backsolving stands. You "expose" falls on its face. >It is simply a waste of bytes. For you I guess it would be a waste. I and others find it useful. -Robin P.S. I don't believe you are "Komputer Korner".
This page took 0.01 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.