Author: Don Dailey
Date: 19:48:26 05/23/98
Go up one level in this thread
On May 23, 1998 at 21:01:53, Komputer Korner wrote: >On May 23, 1998 at 20:34:59, Komputer Korner wrote: > >>On May 23, 1998 at 12:06:28, Don Dailey wrote: >> >>>On May 19, 1998 at 12:45:59, Komputer Korner wrote: >>> >>>>On May 19, 1998 at 12:42:14, Komputer Korner wrote: >>>> >>> >>>>There was an article on refining piece values in the ICCA journal last >>>>year, but there was a major flaw in the research becuase they used >>>>lookahead search of only 4 plies and that is not enough for knight >>>>manoeuvers. >>>>-- >>>>Komputer Korner >>> >>> >>>Dear Komputer Korner, >>> >>>I just read your statement. I suggest that you haven't thought this >>>through very well, but perhaps you have. Your statement seems to >>>indicate that you consider yourself an authority on this kind of >>>scientific research. I consider the author more of an authority >>>than you must be, but that does not mean we cannot form our opinions >>>and in fact this is a natural process. >>> >>>However, I do take exception with your style of presenting it. You >>>presented this statement as a matter of fact. Considering that >>>you call yourself "Komputer Korner", this magnifies any statement >>>you make like this, I suggest you try to be a little more objective >>>in light of this. >>> >>>In my own opinion, your statement is probably not correct anyway, but >>>this is not really the point. Even if it was, your statement lacks any >>>documentation whatsoever as to how you established this as a fact. >>> >>> >>>- Don >> >> >>Hi Don, > >Damn IE 4.01 browser. I have responded to Don Beal in another post >about this. The fact that they used only 4 plies to search and then set >weights based on this, seems to be self evident that the methodology is >flawed. I will quote from the article about this. " ..... Throughout the >course of a single game, a record is kept of the value returned by the >search after each move, and the corresponding principal position. These >values are converted into prediction probabilities by the squashing >function and adjustments to the weights are made according to the >differences between successive predictions, and the differences in piece >counts of the corresponding principal positions." My concern is that >because knights are not sliding pieces, they are much easier trapped >than sliding ones. Searching only 4 plies will not see that knights will >be lost more often. Indeed the final values for knights are much >undervalued. Bacause the values are constantly updated each move >(temporal differences), the final value for a knight at the end of the >game will be less because of the above non sliding piece horizon >phenomenon. After the knights are lost their values can't be updated in >the rest of the game. Sure there was a captures -only quiescence search >at the horizon but this doesn't cover the cases where the knight is >trapped after intervening moves. Thank you for the explanation, now at least we have something to talk about. I'm sorry but I don't really see your point about the knights, it seems like you singled them out as a special case and I don't see them as being particularly trappable. Even if they were, shouldn't this be a function of their value? In my opinion, it's probably the rooks that would benefit the most from increased search depth, not the knights. But all of this is only subjective guesswork on either of our parts and I really don't believe it affects the validity of the study in any significant way. I'm not trying to be disagreeable with you, but after your post I felt obligated to defend this study, to counteract your dismissal of it. Please consider that the things you post may carry some weight in the minds of the less knowledgable readers. - Don
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.