Author: Robert Hyatt
Date: 11:23:22 08/19/05
Go up one level in this thread
On August 19, 2005 at 12:38:56, Drexel,Michael wrote: >On August 18, 2005 at 10:51:55, Robert Hyatt wrote: > >>On August 18, 2005 at 10:42:33, Drexel,Michael wrote: >> >>>On August 18, 2005 at 09:48:19, Peter Berger wrote: >>> >>>>On August 18, 2005 at 09:01:51, Majd Al-Ansari wrote: >>>> >>>>>I was watching the game on playchess and looks like Fruit won the game. Very >>>>>impressive game considering Crafty was using 8 processors and Fruit only 1 >>>>>processor. But I have to admit I think that Crafty was caught by poor book >>>>>opening. >>>> >>>>I am not sure I completely agree with this assessment, at least not when put >>>>this way. >>>> >>>>The potentially controversial move is 11. ... g5 I suppose. I don't think this >>>>move is that bad at all, objectively. >>> >>>I hope you know that objectivity doesn´t count in computer chess at all. >>>Crafty 19.19 thinks the position is about +1.2 for White after this move and >>>Fruit 2.1 thinks it is +1. >>> >>>Before that move both think it was equal. >>>So the game was practically decided after this move. >>> >>>Let's move a little further: 12. Qd2 h6 >>>>13. Qc3 Qf6 14. Kh1 . Here Crafty played 14. ...Bb7?, that is clearly a bad >>>>idea. The king has to stay in the centre or castle queenside if necessary, the >>>>bishop belongs to d7 or e6 and then the rooks both belong to the kingside to >>>>attack white's king. >>>> >>>>After 14. ...Bb7 white has an edge, but black is not lost. After 15. f4 gxf4 16. >>>>Bxf4 the next critical point is reached. Here a possible move is 16...Rg8 e.g. - >>>>nothing to brag about, but black is still well alive. Instead 16. ..O-O ?? is >>>>just suicide. >>>> >>>>So yes - this line should never have been in Crafty's book, because it could not >>>>deal with it. I am to blame for that, so maybe 0.6 points were lost because of >>>>that - bad enough !! The rest is Crafty's fault, that just castled into it. >>>> >>>>Congrats to Fruit - nice game! >>>> >>>>Peter >>> >>>Your book did a good job so far. Nobody is able to prevent such things >>>completely. >>>Bob should develop some tool which analyses all the relevant final positions of >>>a Crafty book automatically and in case Crafty does evaluate this positions >>>worse than -0.7 (for example) writes the associated line and score into a >>>textfile. >>>That should make life much easier for his book cooker :) >>> >>>Michael >> >> >>It isn't quite so simple. This wasn't necessarily the "final position". It was >>a "side position" where white could play a move we didn't have in our book. >>Primarily because the move 12. ... g5 is not in any book I have, which took >>Fruit out at a point where the "refutation" was obvious. >> >>This just happens. In Cray Blitz I had a utility that would "walk through" the >>book, and search at _every_ position. And it caused just as many problems by >>suggesting that a "pawn grab" was safe when it was not, or vice versa, so it >>ends up either being a "learning" thing, or a human analysis problem. >> >>This is what makes this kind of book preparation so very difficult to pull off. >>Not only do you have to look at the position at the _end_ of the book line, but >>you also have to study the positions all along the line from the root to the >>end, to make sure that there is not some tactical shot that was overlooked. > >Certainly it is much better to look at all moves but it is also much more time >consuming. Simply not feasible for a non-professional book cooker since you have >to check _every_ position manually. There might be always some positional >compensation in a position that engines can´t see yet. > >Still, looking only at the leaf positions can prevent you from getting some dead >lost positions. >Just remember the Shredder-Hydra game in which Hydra was completely lost >directly after leaving the book. > >Michael > What I finally ended up doing with Cray Blitz was to have it search every position, sort of like the "crafty annotate command" of today. It would print out a note for any position (interior position or leaf position) where the score was significantly skewed away from zero. Then we had to manually go through all of that and look at the flakey positions to decide if they were ok, or needed to be flagged so we would not go down that line at all... very time-consuming... > > >>I've seen that happen _many_ times. Follow common GM opening moves and suddenly >>_wham_. A move nobody had noticed. In the Cray Blitz days, Bert Gower used to >>spend weeks playing through book lines. When he was preparing white openings, >>he would step through every position, making CB do a deep search for black >>(without book) at every move in the path, to see if there was some deep tactical >>thing that had gone unnoticed previously... It was immensely time-consuming. >>Probably what i should do today is to run on our 128 node cluster, and play >>crafty with book vs crafty without book, to let that pick up the bad openings >>and cull them with learning. That at least would get through a lot of games in >>a hurry, playing 128 at a time...
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.