Author: Robert Hyatt
Date: 21:53:06 10/15/00
Go up one level in this thread
On October 15, 2000 at 23:51:09, Ratko V Tomic wrote: >> Speculation is one thing. Accuracy is another. There _is_ a >> middle ground... > >Are you saying Crafty isn't speculating when it attaches the value to a leaf >node? Anything but the table base (or a forced checkmate) is a speculation. And >nobody quite knows how much free space (due to the inherent speculative nature >of the truncated minimax) you have in any particular position. Depends on your definition of "speculation". IE +.3 is speculating, in a way. +1.3 is more dangerous, because if you have to choose between that +.3 (material is dead even) and the +1.3 (but you are really a pawn down) which is better? +.3 with a pawn up, or +.3 with material down but "positional advantage"??? That means you will sacrifice pawns at times. For something you "hope" is worth that much, without knowing. A common example here is the number of programs that will sacrifice a piece for two pawns... or two pieces for a rook and pawn, when they get some positional "edge". But later the two pieces the opponent has begins to sway things the other way. Take this to a choice between +.3 with a minor positional edge, or +3.3 in position, with a piece down (again a score of +.3). That is _really_ speculative. And you had better be right, because generally being a piece down is tantamount to losing if the attack fails. > >If the Gambit Tiger has different judgment of the amount of this freedom than >Crafty, they're both a priori equal judgments. There is no theory or model which >can quantify in advance how much freedom you have here. At present only the game >results can tell you which judgment is better. I don't disagree there. I simply said that I watched _many_ games with the older CSTal program. And saw the same eval "swings". And they were wrong more often than they were right. I have watched a few beta gambit tiger games where the same thing happened. Tiger may be more adept at not sacrificing too much material, I don't really know as I have not looked at the games in any sort of post-mortem. It might really be making good decisions. But in two games in a row, it failed. It appeared (to me) to have really big scoring bonuses for open files around the opponent's king, big scores for rooks on those files, and big scores for pieces close to the king. It didn't seem to notice cases where the opponent pieces were perfectly positioned so that the attack wouldn't work. I didn't intend to criticize it at all. I was simply saying that the so-called "hyatt paradigm" isn't dead. At least all current commercial programs except for (now) two seem to agree that a pawn is a pawn, and many may well gambit a pawn. But none want to play the Rc6 move. Whether for good reasons or not is the question. The question is, in how many ways can this position be permuted so that Rc6 fails, yet the program tries it? GMs have a nasty habit of noticing that kind of thing and exploiting it handily, given enough games to figure it out. Playing such moves is certainly nice, when it works out. This past week end the Auburn football team went for it on 4th down and 2 yards to go from their own 20, on the opening series of the football game. The punter faked the punt, threw a pass to a wide-open receiver that would have scored had the ball been thrown reasonably... but the pass was underthrown, Florida took over on downs and scored a touchdown a few plays later. "Impaled on one's own sword" is the metaphor that comes to mind, first. Had the pass worked, the coach would have been called 'brilliant'. He was roundly criticized and called an idiot by the newspapers the next day. :) > >Your comment makes it sound that just because Crafty sticks closer to the >inherently inaccurate truncated minimax, that this makes it closer to the >objectively better moves. I didn't say that, and certainly didn't mean to imply that either. If we both agree that _all_ programs have evaluations that are suspect at best, but then in adition to that suspect evaluation we add even more speculative terms so that we begin to sacrifice material in nearly every game, I wonder if that leads to better play. My take: let's wait until the thing is released and see how it does. Without beta testers that exert a bit of influence over the program's time allocation and book choices. We have had _several_ new programs show up over the years and appear to be very strong at first. Until others figure out what is going on and compensate. And then sanity is restored for a period of time. This is just the time for new engines to emerge. And a time for everybody to adjust to the new things being done... It happens _every_ year... >The middle ground you're talking is in reality not >between an "accurate" and a "speculative" method but merely between two >different speculations (or faiths). Otherwise, why would you need to move away >from the "accurate" toward the "middle ground" with the "speculative" at all? >Why not stick with the "accurate" if indeed it is what its label advertizes. >Clearly, you are well aware that it isn't quite what it sounds, otherwise you >wouldn't mention the existence of the proper middle ground. If I had a problem with my intestines, I would allow a surgeon to remove a piece if needed. But not the whole thing. That was my point. tossing pieces can be, and usually is, very dangerous. Unless the search sees a way to recoup the material. Otherwise you have to be _very_ sure that you are going to get it back.
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.