Author: Uri Blass
Date: 02:52:47 06/19/03
Go up one level in this thread
On June 19, 2003 at 03:53:12, martin fierz wrote: >On June 18, 2003 at 12:56:36, Russell Reagan wrote: > >>On June 18, 2003 at 04:24:00, martin fierz wrote: >> >>>better move ordering reduces the branching factor! you can easily see that from >>>best/worst case which is sqrt(N)(best) and N(worst) as branching factor for >>>alpha beta. that's not a constant speed improvement then... >> >>Hi Martin, >> >>Sorry if I wasn't very clear. >> >>I was wondering if pvs/mtd(f) with perfect move ordering led to a decrease in >>the branching factor over alpha-beta with perfect move ordering, or only a >>constant speedup. Better move ordering does lead to a reduced branching factor, >>but I was wondering about the situation where all 3 algorithms had perfect move >>ordering already. >> >>So, I am not talking about the improvement where you go from an N branching >>factor to a sqrt(N) branching factor, but when you go from alpha-beta with >>perfect move ordering, to pvs/mtd(f) with perfect move ordering. >> >>Does this make it more clear? >> >>Thanks, >>Russell > >as dieter points out, bob has answered this question for perfect move ordering. >this is an academic question though, as you never have perfect move ordering in >practice - you do have very good move ordering though. my experience with MTD is >that the speedup is not constant, i.e. the effective branching factor seems >smaller with MTD leading to an exponential speedup with increasing search depth. >i never made a very good experiment on this though! > >cheers > martin I read the claim that wrong pv is often thanks to mtd. If you have not pv that you can trust then it is not clear if MTD is better in practise. You may want to use the pv of the previous iteration for decisions about extensions in the next iteration. Uri
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.