Author: Dann Corbit
Date: 20:57:58 03/06/06
Go up one level in this thread
On March 06, 2006 at 23:51:29, Nathan Thom wrote: >On March 06, 2006 at 23:49:15, Dann Corbit wrote: > >>On March 06, 2006 at 23:45:06, Nathan Thom wrote: >> >>>On March 06, 2006 at 23:40:58, Stuart Cracraft wrote: >>> >>>>On March 06, 2006 at 23:36:22, Dann Corbit wrote: >>>> >>>>>On March 06, 2006 at 22:14:27, Nathan Thom wrote: >>>>> >>>>>>>>3. Search inefficiency (branching factor of a good program is definitely under >>>>>>>>4) >>>>>>> >>>>>>> * My branching factor is about 2-3 for these kinds of positions. >>>>>> >>>>>>How are branching factors calculated? I get wildly different values at each ply >>>>>>as each side usually has different numbers of moves available to them... and at >>>>>>the root node, its always the full number of moves isnt it? >>>>>> >>>>>>e.g, for 8/6k1/6Pp/3r1P2/6K1/n3BP2/1p6/4R3 w - - 3 51 >>>>>>I get branching factors at each ply of 26 2 20 4 16 3 13 3 10 >>>>> >>>>>The simplest and most accurate way to determine your branching factor is to >>>>>divide the time to complete iteration N+1 by the time to complete iteration N >>>>>(don't bother computing it if you had an interrupt halt calculations -- >>>>>calculate it only if it finished naturally). >>>> >>>>That's what I do, then I average them all together for the current >>>>iterative deepening 1-N set for the given search. >>>> >>>>After that I average all those averages together across a test suite >>>>to get the final branching factor. >>>> >>>>The former are br= in my listing and the latter are bf= which is an ongoing >>>>average of the averages. >>>> >>>>Stuart >>> >>>ahhh, that would be why mines so different. i actually keep track of the actual >>>number of moves followed at each ply which to me is what branching factor means. >> >>Look at your counts: >>Hi,low,hi,low... >> >>I think it is hash table that does that with your program, but I guess if you >>calculate the time you will not see the same crazy oscillations. >> >>You should not see branching factors near 30 unless you are using mini-max. Are >>you not using alpha-beta? > >Yes im using alpha-beta and such, but no pruning as yet, and my material only >evaluation doesnt really help :) If you use alpha-beta, it should be nearly impossible to see a branch factor over 20. Optimal choice of pv nodes would bring the branch factor to 6 or so, but even randomly selected pv nodes should bring the branch factor well under 20. I think something may be going wrong with the counts.
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.