Author: Scott Gasch
Date: 16:19:14 08/03/99
Go up one level in this thread
> >The move is actually 1. e6, but mine needs to get to depth 9 to find it and >see the point of it. Before it sees e6, it fiddles about with 1. f3 or 1.h3 >or 1. Rae1 etc etc > My god, I must be doing something terribly wrong. Do you have a static exchange evaluator in use? How do you keep the qeval nodes so low? >> internal nodes..................................7367 >> leaf nodes......................................0 >> qeval nodes.....................................4095107 (2661756 cuts) >> depth nodes.....................................0 >> ------- > >This seems a very odd balance between internal nodes and qeval nodes. >What sort of pruning are you doing in your qsearch? (FYI to get to depth >4, I use a total of about 4500 nodes, ~750 in the main search and ~3750 >in the qsearch. The "internal" nodes are ones in the ab search routine that have children to consider. Some of the qeval nodes are "internal" too -- I count all nodes considered by the qsearch routine as qeval nodes whether they are "leaf" nodes (i.e. I call the static evaluator and get their score) or they lead to recursion. My main search is an ab "negamax" search. It begins calling qeval once it gets to the desired depth. At which point qeval seems to balloon out of control in a busy position like this. I am seeing millions of qeval positions. I added some logic to limit this and still I am seeing millions of positions. I think I need to understand and write a static exchange evaluator? Thanks for the info... I am going to overhaul the qeval routine tonight and research / write a SEE. Scott
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.