Author: Ratko V Tomic
Date: 10:43:26 11/10/99
Go up one level in this thread
> I did not think to pick a uniform odds for counts for pieces. > > It is possible to calculate the probability of every material > configuration and choose 1000 random material configuration... Well, that's point was making -- to get these probabilities you will have to calculate how many placement permutations each MC state has, which amounts to exactly the calculations you and I had done. These approximate probabilities will then carry the sampling bias of exactly the same magnitude as the level of approximation used to obtain them. Any further sampling experiments, biased this way (since our methods were approximate), will then carry that bias in any figures they produce, in addition to the regular statistical uncertainties of finite samples. In any case, from the experimenting with this code, I think that the figure I got is at most 20% above the actual number of legal positions (some minor improvements could be obtained via bishop optimization and some simple types of the wrong side check estimates). But even the a global constraint strenghtening done to get from your to my figure, which removes entire classes of MC states, including all the placements for such MC states, (as opposed to checks which only remove a small fraction of placements within a given MC state), produced only 33 percent reduction in the upper bound. So, all the other stuff which could practically be estimated may likely amount to merely another few percent reduction.
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.