Author: Tony Werten
Date: 00:36:31 08/08/04
Go up one level in this thread
On August 08, 2004 at 01:47:12, Stuart Cracraft wrote: >On August 08, 2004 at 00:31:58, Jeremiah Penery wrote: > >>Perhaps you should try searches longer than one second in order to see if your >>SEE really performs worse. I don't think the performance of a pruning method at >>very low search depths is very indicitave of its performance in more real-world >>situations (like in games). > >Jeremiah, here's the answer to your question. I ran the >same 300 problem test at 5 seconds a move and this showed >SEE with an improvement: > >202i **** 77% 231/300 1249.79 272263232 907544/4/217848 0/0/8711442/0/0/0 > >202h **** 75% 225/300 1255.87 303000768 1010003/4/241269 0/0/17889688/0/0/0 > >It is a smaller improvement, only 6 problems, so about 2% better result. >Speed of searh is slower which confuses me somewhat. As move ordering >improves, speed of search goes down as does node count. More cutoffs >I suppose. That is the idea. You spend extra time (linear loss) to get a smaller tree (exponential win ) Tony > >Anyway, thanks to all who helped with SEE. > >Now my new question is what features do you think would help in solving >problems rapidly. Extensions?
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.