Author: Angrim
Date: 13:40:29 10/22/01
Go up one level in this thread
On October 22, 2001 at 01:00:01, Ratko V Tomic wrote: >>>encoding and a good quality generator, one may encode the best >>>move in a fraction of a single bit, at least for those clasess >>>of endgame positions where the generator can offer the best >>>move with a probability greater than 50%). >> >>Optimist :) To compress to less than 1 bit per position you >>would have to guess right a lot more than 50% of the time. >>When you pass 50% success rate, the cost to store a position which >>was guessed correctly does drop below one bit, but the wrong >>guesses take several bits each to store and pull the average back up. > >It is not that bad. Consider all positions for which you can guess >the best move in the first try 51% of the time and in the second >try 49% of the time. The average bits per position for this set >would be -0.51*log2(0.51)-0.49*log2(0.49)=0.999711. Hmm, math looks right, although the assumptions required to get this results are rather extreme. namely that you never miss with guess #2 I was thinking of first guess 50%, second guess 25%, third guess 15%, fourth guess 8%, fifth or sixth guess %1 and I was thinking that this was the optimistic case. I guess that the case you suggest could happen, ie. in KB vs KNN where almost all positions are draws, and almost all moves would be tied for the best.. but the current compression method does very good with these also since a whole bunch of draw scores can compress down to a few bits. Angrim
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.