Author: Scott Gasch
Date: 10:50:48 10/06/99
Go up one level in this thread
On October 06, 1999 at 09:41:00, Steve Maughan wrote: >What is the 'best' or lowest ratio of quiescent nodes to normal nodes. I am >currently analysing about 5:1 ie the vast majority of my time is being spent in >QSearch and I was wondering what is 'normal'. > >For the aviodance of doubt, I have two search procedures AlphaBeta and QSearch. >The ratio that I'm interested in is the ratio of nodes processed by QSearch >against those processed by AlphaBeta. Hi Steve, I too have this problem -- I see about 4x more qnodes than tree nodes. I am doing winning captures only (with a SEE) and a futility check (delta pruning ala crafty) in the qeval routine. I am not sure what I could be doing wrong unless my move ordering is messed up somehow. My move generator returns: 1) PV nodes 2) winning captures 3) losing captures 4) killers 5) the rest ordered based on history table I appreciate any advice... here are some sample search stats after 1. a2a3 searching to 6 ply: Thinking. 1 39 0 55 g8f6 2 4 0 752 g7g6 g1f3 g8f6 3 29 0 976 g8f6 g1f3 b8c6 4 18 0 3905 d7d5 b1c3 d5d4 c3e4 c8f5 5 4 100 22049 g8f6 g1f3 b8a6 6 21 500 111840 g8f6 g1f3 b7b6 g2g3 c8b7 f1g2 b7f3 move g8f6 ------------------------------------------------------------------ search stats: 111842 search nodes (111840 internal, 2 leaf) and 453383 qeval nodes. 1221 qeval futility cuts, 1263 search futility cuts, and 220881 lazy evals. processed 82076 line extensions. 79357 null move successes, 29216 failures, 73.09% null move efficiency. searched for 5.24 sec at a rate of 107951 nodes/sec. hash stats: main hash: 24217 adds, 1637 (14) hits, 110203 misses, 1.46% efficiency. pawn hash: 153812 hits, 78691 misses, 66.15% efficiency. misc. stats: aspiration: 6 successes, 0 failures, 100.00% efficiency. ------------------------------------------------------------------
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.