Author: Thorsten Czub
Date: 04:55:14 12/31/97
Go up one level in this thread
>It's not a good example IMO. >When I set the G5 playing style to "Risky" Rxe6 is quickly found. When I set the style of different programs I do get any kinds of results. In fact, when I put genius into risky I would expect it to find sac-moves. So I don't see your point. Genius evaluates Re6 ONE PLY BEFORE much negative for black than it is NOT seeing the move one ply later ! We don't know what genius does when you switch to risky. It could be it changes the evaluation-weightings, it could be it computes the sac-branches first, it could be it increases the wide of the PRUNE-number (more moves were looked for in the 1,3,5,7 plies especially sac-moves). We don't know what genius does exactly when changing the style. I don't want to critic genius. My point is not: Oh - genius is a shit program, I don't like it, it oversees things. I know that any program has different weaknesses. And I also know genius is a strong program. My point is: There is a pattern with genius. Whenever it really fails, you can prove this yourself, than going 1 ply back or 1 ply forward finds normally the problematic move. This is what I often found when I analysed the LOST games of Genius (I am honest, i studied the WINNING games of CSTal against Genius and tried the problematical positions with going one ply back/forward). And very very often genius had NOOOOOOOOOO problem with the "shifted" (=key move is not computed with iteration 2,4,6,8..) search. Therefore I think it has NOTHING to do with the evaluations but with the search. You can now argue that latest efforts have combined this and lazy-evaluations DO only the functions that don't need much time if the position needs no better evaluation. But the asymmetrie thing is in genius from the very beginning of the AMSTERDAM module and has not changed (much). I think the search results in the style, more than the evaluation results in the playing style of a symmetric program. I use this example to show what I talk about ! I don't want to be 100 % precise or always right. If you have any evidence WHY my example is bad (despite the fact that risky style finds Rxe6 [what do you expect from a risky style ?? I HAS to find a sac-move !!!]) please comment on it so that I can learn from your programmers experience. Again: whatever Chris mentiones it is NOT my point to say: I don't like Genius. I do - in fact - believe that Genius has not made much programming progress from Version 2 to 5. But you NEED genius. At least to let it play against your favourite program. Genius is a kind of blind-test. You need genius as an enemy. If your program wins against genius, it cannot be weak !! So - believe whatever you believe: I have nothing against genius. I have a different opinion concerning the strength progress between the versions but I don't have anything against the program or the programmer or whoever. > >I think the G5 score differences between Bxc6 and Rxe6 are to >narrow to proof your point with this position. Really ?! I think you are dead when you oversee Rxe6. Although the evaluation is "narrow" .40 of a pawn, if you OVERSEE rxe6 you are dead as topalov was. So maybe this is narrow by evaluation but than the evaluation is wrong. > >Besides of that every selective program (null-move) or otherwise >has errors in pruning the tree. There will be no algorithm ever >which will replace the brute force approach. Just set any Rebel >version to brute force and watch the differences. Very shocking >sometimes... I have never said the opposite. I know this. But we discussed about Genius search. Or I did. > >The same applies for every program I have seen which offers the >user the "brute force" option. It shows you the weak points in the >selective search of a program. My idea is not to show the weak points, but to find out HOW THE MECHANISM WORKS. > >I am collecting these kind of failures since many years. In every >new Rebel version I try to fix one or two patterns. It's a >nightmare for every chess programmer. > Right. I still think my example is exactly a result of the search. But - only richard knows, for me the program is a pandorra-box and I can only guess. But - there are enough quiz-shows in the TV and it seems this is a nice job to do in your spare time. So lets ask us questions about HOW. Maybe we find out interesting things. And when we found out anything, we analyze Junior next !! :-)
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.