tag:blogger.com,1999:blog-336308386934546555.post3488586796642816981..comments2009-12-28T00:00:16.436+08:00William Shieldshttp://www.blogger.com/profile/18356811199950883367noreply@blogger.comBlogger4125tag:blogger.com,1999:blog-336308386934546555.post-88478010402240169442009-12-28T00:00:16.436+08:002009-12-28T00:00:16.436+08:00Very interesting post which I have to investigate a little bit further. As it seems you are using CPT I just wanted to leave a short note that CPT 4 is using the huffman coding instead of the simple FEN (for example) which was used in previous versions to save database space.<br /><br />StefanStefanhttp://www.blogger.com/profile/06193709795748789633noreply@blogger.comtag:blogger.com,1999:blog-336308386934546555.post-85227079915302874192009-12-09T23:34:06.441+08:002009-12-09T23:34:06.441+08:00The piece position encoding can be further optimizied. For example, for eight pawns, the pawns are interchangeable, yielding only (48!/(40!*8!)) different pawn configurations. This kind of optimization is used in endgame table bases. Even if you use the simpler encoding, like 48^8/8!, you save some 16 bits per side. The same goes for the pieces (to a lesser degree), which brings a reduction by a factor of 2! for each piece pair which sums up to 3 bits per side on a full board. And there is of course the special case of the bishop, which has only 32 squares to visit, but two different base-square-colors. This is usually handled by treating the white-squared bishop and and black-squared bishop as different pieces for the purpose of counting.drhirschhttp://www.blogger.com/profile/10919861063391740606noreply@blogger.comtag:blogger.com,1999:blog-336308386934546555.post-60359753655014651272009-12-08T11:21:10.400+08:002009-12-08T11:21:10.400+08:00The solution you hint at in passing, where all possible legal moves are calculated, and the encoded value simply selects which one was played, is very efficient - if (at a guess) there are on average around 50 possible legal moves from any position, then you only need on average 6 bits per move.<br /><br />This could then possibly be improved upon by some form of Huffman encoding combined with an algorithm for determining which moves were most likely, however that is adding a lot of complexity.Anonymousnoreply@blogger.comtag:blogger.com,1999:blog-336308386934546555.post-16733163065280457202009-12-08T02:08:15.479+08:002009-12-08T02:08:15.479+08:00Outstanding post, but the link to the question on SO is broken. Also, I think you mean &quot;specifies&quot; not &quot;species&quot;.Dour High Archhttp://dourhigharch.pip.verisignlabs.com/noreply@blogger.com