Author: Robert Hyatt
Date: 21:59:21 12/30/00
Go up one level in this thread
On December 30, 2000 at 19:58:32, William Bryant wrote: >On December 30, 2000 at 16:26:30, Scott Gasch wrote: > >My numbers are similar, averaging about 85% ranging from 79% to 95% depending >upon the position. Bob quotes Crafty as approx. 92%. I could easily believe >that my move ordering could be tweeked a bit. > >William >wbryant@ix.netcom.com > > >>Hi, >> >>I remember a while back Bob said a good statistic to measure move ordering by >>was the percentage of beta cutoffs that occur on the first move tried in a >>position. I'm looking for bugs / ways to improve my code by instrumenting it >>with a lot of counters and dumping statictics at the end of a move... I recently >>added this metric and ended up with a rate lower than I had expected. >> >>In typical positions the rate is about 80%..85%. Only in positions where there >>is a winning tactical move or a piece threatoned does the rate exceed 90%. >>Despite this I think my tree size is fairly good and move ordering is fairly >>standard. Does anyone else measure this statistic and, if so, what are your >>results? >> >>Thanks in advance, >>Scott I would suggest you _both_ try crafty on the same position you are testing with, so that you can compare apples with apples. I would definitely be concerned if the fail high on first move percentage gets below 90... But note that I do not count quiescence searching at all since move ordering is a lot simpler there. And note that I only count this on nodes that fail high on some move. If it fails high anywhere, in increment a counter. If it fails high on the _first_ move searched, I increment another counter (I don't count the null-move as a move here, for obvious reasons). But make sure you test crafty on your position, so that you can see whether you do better or worse on the same position. This can vary pretty wildly in some positions...
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.