Author: Ulysses Omycron
Date: 10:26:53 01/03/05
Go up one level in this thread
>>The problem with this methode is that you can miss a good move at the first move >>because of a deep fail high, this kind of thinks can happen especially when a >>king is in open air ... >> >>With your method it's very difficult to find a move like : Nxh6 ! >> >>r3qb1k/1b4p1/p2pr2p/3n4/Pnp1N1N1/6RP/1B3PP1/1B1QR1K1 w - - 0 1 > >Bad example, I should chose a deeper one ... The method I described above has been proved wrong by Uri. But he gave me lots of ideas about making a better and faster DPA, based on the fact that hash tables can save lots of time, and that you should only increase the Branches if you think a good move it's not being considered (But don't imput it yourself! That's cheating ;) I will report the best way of using DPA as I'm finding it, it seems flawlessish (Only flaws are it takes much more time... But you used 9 days... Other one is that it doesn't handle transpositions well, so you end with duplicated lines, but hash takes care of giving them different scores; And the last flaw it's when you search the same position again to the same move, it won't update its score even if you searched it more deeper [plywise], but this is solved by removing the Engine's commentary). I still have to find a testposition that fails to be analysed by DPA, all Uri gave me is positions in where the best move is there blinking and glowing, so it's very hard to ignore and Shredder finds it at ply 19 (Or much before) even with 24x4 branches (That takes around 10 minutes on my computer). So, I hope you get nice gifts next party :)
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.