Author: Bruce Moreland
Date: 13:04:45 11/29/00
Go up one level in this thread
On November 29, 2000 at 05:38:35, Andrew Williams wrote: >On November 28, 2000 at 19:35:17, Scott Gasch wrote: > >>On November 28, 2000 at 14:06:27, Andrew Williams wrote: >>> >>>PostModernist Monsoon >>>1 145 45 >>>2 214 621 >>>3 864 3725 >>>4 3330 14146 >>>5 14057 38694 >>>6 38838 183449 >>>7 120729 598020 >>>8 294806 1286875 >>> >>>This is with all my normal extensions on. >> >>This idea of 2 sets of killers -- one from this ply and one from last time side >>to move moved is an interesting one. Also I thought that a seperate history >>table for each color was a cool idea. >> >>Are you doing internal iterative deepening in PM? Monsoon does not do that yet > >Nope. Not tried this yet. > >>nor does it use a SEE to decide which captures are really "winning". I think >>the next thing I implement is going to be a SEE so that I can improve capture >>sorting in the search and qsearch. It seems like you are still an order of >>magnitude better than monsoon in tree size, though. I wonder if I am doing >>something wrong or your are doing something right or both. If your nullmove R=2 >>or 3? >> >>Scott > >I think I wasn't clear. I use either 3 or 2 based on depth remaining and pieces >on the board. I originally borrowed this from Ernst Heinz's book, but I've >fiddled about with it a bit since. > >Andrew You can get variation depending upon whether you switch back and forth between different moves at the root. You have to search a lot more nodes if one of the pawn captures fails high after the first one has been searched. So you guys should know that aren't really comparing anything other than rough estimates. I think internal iterative deepening helps a little, but it's not an order of magnitude thing or people would be screaming about it. Adding a SEE to quiescent search will help dramatically. bruce
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.