Author: Robert Hyatt
Date: 15:13:22 08/28/98
Go up one level in this thread
On August 28, 1998 at 15:13:48, Mark Young wrote: >On August 28, 1998 at 10:12:22, Robert Hyatt wrote: > >>On August 27, 1998 at 21:51:33, Robert Henry Durrett wrote: >> >>>On August 27, 1998 at 20:39:45, Robert Hyatt wrote: >>> >>><snip> >>>>I think Junior's playing fine... It made two "odd" moves today.. one was >>>>Rh5, which put the rook in a bad place, the other was Bh8 which put the >>>>bishop on a bad square and also prevented the rook from being able to go to >>>>h8-b8 or wherever to get to the queen-side... But the moves were *not* >>>>gross errors, just simple positional errors... >>>> >>>>Yudasin also played some ugly moves along the way, and at one point turned >>>>a really significant advantage into nothing with just a couple of moves. He >>>>commented on this later. However, he took advantage of a black rook out of >>>>play and rushed into a nice tactical position that was winning. But he also >>>>*almost* let Junior equalize, although I think that Junior was more interested >>>>in kingside activity than queenside, and in this game the queenside was where >>>>all the action happened... >>>> >>>>Yudasin tried to attack on the kingside himself, but after black tangled >>>>his pieces (the rook and bishop) he switched to the queen-side quite >>>>nicely and broke through there. Well played. However, I disagree that >>>>Junior embarassed itself... it repeated the rebel/anand result, which is >>>>*not* easy... >>> >>>Yes, I got the same impression by watching the game. The thing, though, that >>>stands out in my mind is the way one of Junior's rooks got out of play in the >>>same manner (side of board locked in by pawns)in both games! Although this is >>>only a sample size of two, it still makes me sit up and take notice. Maybe >>>Junior should be "tweaked" to make piece coordination (especially keeping rooks >>>working together) have a higher priority. Also, this maybe should be considered >>>as also a piece activity issue. [A word of caution: consult a chessmaster >>>about this before modifying Junior.] I would suggest that Junior should have >>>become concerned about the extremely poor placement of the rooks and done >>>something about it long before it was too late. >>> >>>In the second game, unlike the first, there came a time when Junior could not >>>seem to figure out what was going on in the position and could not, therefore, >>>make anything like a plan. How do you fix that???? >> >>The following is pure speculation, and Amir/Shay can certainly correct what is >>wrong... but my opinion is that Junior relies heavily on "piece/square" tables >>for evaluation... Here's why I believe that: >> >>1. very fast NPS rate, approaching Fritz, which is an ASM program. >> >>2. a positional example that I have seen repeated often against both Ferret >>and Crafty on ICC: rook on the 7th rank. I have watched Junior repeatedly >>stick a rook on the 7th, after the king has vacated the 8th, and there are *no* >>pawns on the 7th either. IE the rook is attacking nothing, not constraining the >>king at all, yet it goes there and even more important *stays* there. >> >>If the piece/square speculation is true, that can lead to one type of problem, >>because you set the piece/square tables up at the root, using a pre-processor >>that tries to figure out the best/worst squares for each piece, based on static >>analysis of the board. But it is very difficult to statically figure outhow to >>untangle pieces, which might explain what was going on.. and when the action >>swung from the kingside to the queenside, this type of static analysis doesn't >>make a "smooth" transition from one side to the other, like an endpoing analysis >>could do; rather, it plays on the kingside until something becomes apparent to >>the pre-analyzer that causes it to adjust piece/square values to start >>attracting things to the other side... >> >>Again, all speculation, but it can explain this perhaps... > >If this is what is happening to junior is this problem fixable, or is this just >an inherent weakness in "piece/square" tables? Note that I am only speculating when I guess that junior uses piece/square tables. But I don't see any other way (except for a few really ugly ideas about using a static exchange evaluator in lieu of quiescence search here and there) to get the speed up to an assembly-engine level using plain C (or C++) programming. It has definite weaknesses. If something changes away from the root, you still evaluate as though the root circumstance is true, like a weak king at g8 that scampers to b8 but you continue to pile up on g8. It has advantages in that it provides a significant speedup (I spend way over 50% of my total time in the Evaluate() procedure. A piece/square program can get this way under 10% if wanted, which is worth almost another ply. From my perspective, having done this many years ago, I don't like the effect. But several programs are using it quite successfully, including Wchess (Dave Kittinger's program) and Fritz, to name but two. The problem is, however, as the search depth increases, the position you are evaluating looks more and more different from the position at the root of the tree. And it really has problems in situations like yesterday where the game suddenly switches from a kingside battle to a queenside battle. Because the "switch" (inside the program) doesn't occur very smoothly...
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.