Author: Bert van den Akker
Date: 15:48:04 11/21/99
Go up one level in this thread
On November 21, 1999 at 17:15:49, Robert Hyatt wrote: >On November 21, 1999 at 14:59:53, Bert van den Akker wrote: > >>From the start position I get a branche factor of about 14% >> >>This is in a 10 ply search from the start position. >> >>The branche factor is defined as: >> >>(total_number_of_moves_looked_at_in_a_node_before_cutoff / >>total_number_of_generates_moves) * 100 >> >>Only in the normal search no (quiesence search not counted)) >> >>In my case fo a 10 ply search >>branche factor = 14.2408293344493% (955856/6712081) >> >>Is this branch factor good or bad? >> > > >That isn't the proper definition for 'branching factor'. What you are >describing is "effective branching factor". You want to do this the other >way also... Divide the number of nodes for the N ply search by the number >of nodes for the N-1 ply search. That should give you a number that means >something to everyone that sees it. 3.0 is good for the middlegame if you >use null-move or some forward-pruning algorithm. 5-6 is normal for regular >alpha/beta... > > > >> >> >>BvdA The main reason why I think my definition of branching factor is important because in general only 14% of all generated moves are used in a position. So generating not all moves in one go in a position in the tree or generating only the new moves in a position can speed up a chess program if the overhead < 86% of generating all moves. Here is my branching factor for ply 10/9 a) number of calls of the evaluation function ply = calls evaluation function 8 = 169054 9 = 291827 10 = 710146 ratio ply 10/9 = 2.43 b) number of nodes searched ply = (normal nodes) (quiesence nodes) 8 = (nodes= 47599) (q nodes= 142901) 9 = (nodes= 99770)(q nodes= 248053) 10 = (nodes= 236843)(q nodes= 599652) ratio ply 10/9 = (236843 + 599652) / (99770 + 248053) = 836495 / 347823 = 2.4 BvdA
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.