Author: Don Dailey
Date: 10:26:09 12/26/97
Go up one level in this thread
On December 26, 1997 at 11:30:40, Amir Ban wrote: >On December 26, 1997 at 10:59:45, Thorsten Czub wrote: > > >>My mood while watching Mchess7 reduces more and more. >>I have seen bishop-game where Mchess was 27 moves deep in book meanwhile >>hiarcs computed a while. >>Now I see c28 vienna game happening and we are in the 32.move and mchess >>still in book. One game after the other is cooked out somewhere else >>(sandro necci, or in massive autoplayer-games-merged into many >>booklines), and I don't see much sense in doing this. >>WHO can trust that these games, if played under the same conditions as >>in the original "citchen" , will not result in the same LOSS for hiarcs >>? >>I play 100 games hiarcs6 vs. mcp7 and make an opening book out of the 30 >>losses of hiarcs6. >>Now I put these 30 games into a book. >>When stupid customer or ssdf-guy plays mchess7 vs. hiarcs6 he will get >>openings beginning in exactly these 30 losses and also some other >>openings caused by whatever circumstances. >> >>I think this helps mchess7 to get a better score at all. And hiarcs gets >>a weaker one. But hiarcs was released before mcp7, so it cannot defend >>much. >>Ok - they all have learning algorithms, but I doubt that these mechanism >>will always help...not from my experience, what I have seen. > >This sounds incredible. Incredibly stupid, IMO. > >Doesn't all this mean that the professional programs don't have enough >variation in their tournament book ? Looks to me like they painted >themselves into a corner and are playing an obscure book-war. What good >would these obscene practices have against a program that plays a >different line every time ? > >Amir I also find this quite distasteful. I think it's becoming a war that they feel they have to fight just to stay on even ground. I'm wondering if providing a lot of variety would really help much. I want to think about this a little before deciding and except to see some responses. But here are some ideas: I'm assuming the other program has complete access to your book and wants to cook it. But now you will tend to be putting 2nd best moves in your book simply to provide variety. There are many positions in the book where only 1 move makes sense but still there are opportunities for variety so maybe this will be ok. I also assume the other program will have some automated procedure to play thousands of games, and remember which lines work. But it sounds like a key point to Marty's technique is to identify when a program is out of book, and then make a note of whether the game is won or not. When Marty's program wins, you place the actual moves of the game into the book. Pehaps in these tests Mchess is given a great deal more time for book cooking. It seems to me that the large variety style of book has some advantages and also (maybe) some drawbacks. First of all, it doesn't prevent book cooking, but I believe it will require significantly more effort on the part of the book "cooker." But eventually the cooker will traverse every book end node and will have decided where to vary. It's also possible that the variety will work against you. The more variety you have, the less tuned your own book will be. You may be forced (for variety's sake) to include moves that are not quite as good or at least not quite as good stylistically for your program and it may give the cooker more opportunities to hurt you. But all in all, my current sense of this (subject to change) is that this will probably be a fairly effective way to combat the cookers. The cookers must prepare against multiple opponents and not just one and this will greatly increase their efforts, especially if they all increase their variety. Also you will have to expend much more effort preparing a variety book that does not weakening your program significantly. This may be a tall order but perhaps not impossible. But we do not really know exactly what Marty is doing either. I made some assumptions about how he might be preparing the books. A point I didn't mention was how to control the actual move your program (you as the book cooker) plays. You can do this with a min-max procedure, where you include only the book moves, and try to identify the quality of the end nodes (win lose draw.) You will probably need a scoring function that takes into account mixed results against different opponents. Then you simply min-max the tree keeping the returned "score" from each search attached to the book move itself for selection purposes. The actual implemenation will contain lots of details not included here! But if there is a move guaranteed to win against every opponent, this type of procedure will find it. -- Don
This page took 0.02 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.