Author: Komputer Korner
Date: 11:28:01 05/14/05
Go up one level in this thread
On May 14, 2005 at 13:00:34, David Rudel wrote: >KK, >I think you have both the situation and Stephen's response wrong. > >Possibly it is a language barrier problem [I do not know if English is your >native language], but you say > >>You admitted that you don't use it. > >What Stephen actually said is > >>>I have a Bookup database with 1.3MM moves in it, all input manually by me over >>>the past 15-years. Given that size, I never use the automatic backsolve >>>function. > >The only thing Stephen is saying he doesn't use is the AUTOMATIC backsolve. I am >sure he still backsolves after inputting a new set of moves, he just doesn't do >it after each one is put in separately]. > >Since Stephen goes on to say how helpful and useful the backsolving feature is, >it seems that there either must be a language barrier here or a willful interest >in selective reading. > >To answer some of your other questions: > >Bookup backsolves both the numeric assessments [that is the assessment an engine >would place if it had all the assessments from all the moves later in the tree] >and the Human evaluation [informator symbols] assessments that you enter based >on literature or your own evaluation of the material. > >It also backsolves the total number of variations past a position. > > >It handles all your transposition concerns because Bookup is a positional >database, so all it cares about is whether a certain position with a certain >assessment is reached with best play from both sides [exactly the min-max >method]. However, it will show all move orders and do other transposition >gymnastics if you wish. > > >I do not know what your beef is with Backsolving, but the claim is that >backsolving is a much better method of finding the correct evaluation of a >position than using statistical inference based on a selection of human games. >In this contention, backsolving has a lot to be said for it. > >As for how a move done at move 30 can be a refutation to a choice chosen at move >12, that is the nature of chess. You make a choice at move 12 and that choice, >with what is known to be good play from both sides, leads to a certain position >well down the road. If someone finds a new move in that position, an >improvement has to be found to save the line. That improvement may be hard to >find going back until the choice made on move 12. > >Now, the numbers 12 and 30 in the above are a bit of hyperbole, but the general >idea is still valid. > >Here is an example in a somewhat popular variation of the slav: > >1.d4 d5 2.c4 c6 3.Nf3 Nf6 4.e3 Bf5 5.Bd3 Bxd3 > >If one were just looking at databases, Bg5 would be the overwhelming choice >here, statistically Black has a huge advantage with Bg5, so why does a >world-class GM play Bxd3 against Karpov? > >6. Qxd3 e6 7.O-O Nbd7 8.Nc3 Bb4 9. Bd2 > >And here is where we get to the answer. From this position, WHite does very >well [statistically] agaisnt everything except... > >9...a5 > > >So, since a5 is not an obvious move, you will have many games where Black choses >something else and White does well. Since this is a very large branch of this >opening, it will skew 5...Bxd3 to look unfavorable [and 4.e3 to look >favorable]. However, Black does very well in the one variation 9...a5, so if >you had entered all the main lines into bookup and your engine correctly >evaluated the end-nodes correctly, backsolving would show this whole line to be >rather lackluster for White, whereas statistics will show it to be reasonably >good. > > >Another example is the Winawer Counter-gambit, 1.d4 d5 2.c4 c6 3.Nc3 e5!?. This >variation scored very well for Black for a few years until Kasparov found a >REFUTATION on 7 plies later. Whether you do automatic backsolving or manual backsolving, you are still not really doing anything. Of course one cannot rely on move statistics because one strong move can refute a whole line. However backsolving does not solve the problem of novelties refuting your repertoire. If a novelty refutes a line, you simply manually change the evaluation at move Ex: 12.Ng5. You don't need to have evaluations at the subsequent nodes if you aren't going to play that line. If you are going to play that line, then at each subsequent branch as white you need only one node evaluated(the node move that you will play against any sucker unfortunate to follow you down that line). You don't need the black evaluations because you aren't going to play that line as black. Think of it this way. Whether you add one line at a time or else add a bunch of lines before backsolving all the nodes for white and black get evaluations. I suppose backsolving would be beneficial if you could specify whether only white nodes or black nodes get added evaluations and only those nodes where there is more than 1 choice. Your opening repertoire is for you only. It is good for no one else. Therefore you don't need evaluations at every node.
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.