Author: Peter Kappler
Date: 14:44:46 10/06/99
Go up one level in this thread
On October 06, 1999 at 13:55:45, Robert Hyatt wrote: >On October 06, 1999 at 13:34:59, Steve Maughan wrote: > >>Bas >> >>> >>>First: are you sure you don't doublecount nodes? >> >>I'm not double counting. >> >>>If you put Nodes++ on top of ab and Qnodes++ on top of qsearch, then you >>>doublecount the depth = 0 nodes. Because you count em in the normal search *and* >>>the qsearch. Therefore, before you go into the qsearch do a qnodes--. >>> >>>Counted that way qnodes should be a smal fraction of total nodes. It depends >>>also on the position. And according to my experience the use of nullmove worsens >>>that rate, because normal nodes get pruned at the cost of extra qsearches. >>>Without null I see qrates of 10-20% and *with* nullmove it is more like 50%. >> >>I can't see how the ratio of QNodes:NormalNodes can go below 1:1 (eg 10-20%) as >>each NormalNode must give rise to at least one QNode. Where am I going wrong? >> > >nowhere... you are exactly right. but then those particular q-nodes are not >optional, while any below those qnodes are. There is where you can save... but >you won't get below 1:1 counting like you are. An alternative is to count >internal nodes (depth > 0) as nodes, leaf nodes (depth == 0) as leaves, and >the rest as qsearch nodes. the latter is the only one you can control... > I'm getting confused as a I follow this thread. In my search, only about 15% of the total nodes searched are q-nodes. My counting scheme is very simple - I do either nodes++ or qnodes++ depending on where I am when I call make_move(). I don't do anything fancy to trim q-nodes - just some simple futility pruning. I don't even have a SEE to toss out bad captures, so I'm always surprised when I see that my qnode/total_node percentage is low compared to everybody else. I suspect the difference is that I'm counting depth==1 nodes as regular nodes, and others are counting them as q-nodes. I can't think of any other explanation for this disparity. Comments? --Peter
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.