Author: Terry McCracken
Date: 22:44:57 05/18/05
Go up one level in this thread
On May 18, 2005 at 22:22:50, Komputer Korner wrote: >I have made my point but so few people seem to get it. People are confusing >minimaxing and backsolving. Minimaxing simply lets one decide what path to >follow. However backsolving would be very useful if you had the whole tree >evaluated(that is every node). Then you could press a button and voila the whole >tree would have annotations against every node and the winnng or drawing path >would become clear. However we will never obtain this position, so why be >concerned with having evaluations attached to only part of your ultimate opening >repertoire? Having an opening repertoire involves a constant process of fixing >holes and experimenting with new lines. It shouldn't be a process of evaluating > endgame positions and proprogating the result back down the tree. If you are >doing that then you are not using backsolving correctly. The study of openings >is not the study of endgame positions. Who cares if a 40 move line has >evaluations at every node if 99.9999% of the time, you will never meet that line >in your play. The opponent will have veered off far sooner. So what has >backsolving actually done for you? It simply has minimaxed it's way to attach >evaluations to nodes in your opening repertoire. Backsolvers like to be content >with the knowledge that the computer is helping them attach openining >evaluations to all the nodes in their opening tree. They think the computer is >doing useful work for them. However if they don't spend time analyzing the moves >at the start of a game all they will have is a tree that has a precious few >refuted lines that they should never play for one side and always play if >allowed for the other side. Meanwhile non backsolvers do not care about the >results of games. They care about the analysis of opening positions. They put >the engines to work on opening problems and opening strategies. A number of >posters who back backsolving think that I am against Bookup. I think Bookup is a >great program. I simply claim that backsolving is an overhyped feature in the >program. A number of posters have claimed that backsolving helps them with >transpositions. Backsolving has nothing to do with transpositions. Let us take >an example. Let us say that you have the moves 1.d4d5 2.c4c6 3.Nf3Nf6 4.Nc3 in >your opening repertoire tree. Let us say that at the final position you have it >analyzed as + .16 pawns for white. You backsolve it and then every node gets a > +.16 evaluation attached to it. You then play a game as black against someone >who opens up 1.d3. you probably should play 1...d5 or 1...e5 or 1...c5 but you >reason 1...d6 is an acceptable move against most white openings therefore why >not play it against 1.d3 Maybe your opponent will play 2.d4 so you play 1...d6. >Now your opponent does indeed play 2.d4 in effect wasting the white half move >opening advantage. However since you don't have these moves in your opening >repertoire, if you play anything but 2..d5 you are on your own. So being a >committed backsolver that you are, you play 2...d5 because now your opponent and >you will be back in your bullet proof opening repertoire that you have built up >by backsolving results of games. Since you have previously backsolved this >position, indeed because of bookup's position based format, the transposition >immediately brings you to the same node position as after 1.d4d5. So you are >happy now, because from here on until the opponent veers out of your book, you >will have an evaluation against any position you and your opponent reach. >however if you had manually put the +.16 result of 1.d4d5 against 2.c4 yourself >you wouldn't need to have backsolving do it. The point is that even in this >simple example of backsolving, backsolving didn't help you discover the >transposition. The transposition was already there just waiting to be played >and waiting for Bookup to discover it. Automatic transposition paths in Bookup >are a wonderful thing but backsolving doesn't discover the transpositions for >you. Transpositions are only useful by going forwards not backwards. You don't >play chess backwards. You play chess forward one move at a time. Of course you >need an evaluation wherever you have a move choice. However the moves have to >be already in the opening tree for backsolving to attach evaluations to them. >Because the moves 1.d3d6 were not in your opening tree, backsolving couldn't >attach an evaluation against these moves. It had only the position after the >starting moves 1.d4d5 to attach moves to , not the 1.d3d6 position. of course, >after 2.d4d5 Bookup automatically brings you to the node transpostion , but >backsolving is not at work here. only after you put the moves 1.d3d6 2.d4d5 and >then backsolve again will you have evaluations against those 1.d3 and 1.d3d6 >nodes. Of course the .16 pawns evaluation automatically gets added to each node >but that is not because of the magic of transpositions. Backsolving did that. . >It backed up the .16 pawns right back to the begining of the game. however if >you analyze the position with an engine at 2.c4 and the engine comes up with >.16 pawns you can put the .16 pawns in yourself at that position. You don't need >the .16 pawns evaluation at 3.NF3 or 4.Nc3 unless there are move choices at >those position. if there are move choices you must have added them yourself. >Either the evaluations at the other move choices were put in by backsolving or >you manually added them. Either way they have to get there. in order for you to >know which path to take. If the evaluations were put there by backsolving, what >end position is it depending on? the 10th move, 20th, 30th 40th? The farther >out you go the more worthless the evaluation. So even if you never backsolve >any farther than the 10th move out, you still have to analyze any node where >there is a move choice. The backsolvers will argue that with backsolving you >already have an evaluation at every node and thus when you add a new move choice >at a particular node, then you have saved the work of manually adding that >evaluation to the 1st move choice that was at the node that you are adding a new >move to. So this then is the ONLY ADVANTAGE OF BACKSOLVING. It will save you >the work of adding the evaluation to a node that in the future will have another > move alternative added to that position. THEREFORE I APOLOGIZE. Backsolving >does have 1 advantage. It is interesting that no proponent of backsolving has >pointed this out. However backsolving has a dangerous side to it that I have >pointed out. This dangerous side is the fact that backsolvers will base their >repertoire paths based on backed up evaluations that were made near the end of >the game. These evaluations are very dubious. You should restrict your >backsolving to no later than the 20th move in most lines (some book lines go >out as far as 30 moves or later , so an exception can be made for those). Then >you can backsolve in confidence knowing that you will save time in the long run >by not needing to manually add those annotations. The important thing to >remember is to only backsolve while you are in the opening . NO LATER. SO I >REPEAT AGAIN, I APOLOGIZE. Well written...I hope the forum recieves your well thought out apology and explanation. Regards, Terry
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.