Author: Brian Richardson
Date: 14:37:40 05/05/00
Go up one level in this thread
On May 05, 2000 at 09:58:45, Robert Hyatt wrote: >On May 05, 2000 at 07:49:03, Brian Richardson wrote: > >>On May 04, 2000 at 22:03:12, Robert Hyatt wrote: >> >>>On May 04, 2000 at 19:05:20, Lance Nine wrote: >>> >>>>how do i calculate the nodes per second my program checks when i use the >>>>alfa-beta algorithm? do i count the nodes it does not look at due to cutoffs? >>>> >>>>if i don't, the speed in nodes per second will be the same for negamax and >>>>alfa-beta, right? >>>> >>>>the speed of my program is rather disappointing otherwise... >>>> >>>>thanks, >>>>--lance >>> >>> >>>The most common way to count nodes is a "nodes++" at the very top of >>>Search(). divide that by the time used. No way to count the nodes you >>>don't search. How many are there and how would you know that? >> >> >>How about nodes in the quiscence search? I was double counting (first q-search >>level nodes) by doing nodes++ in search(), and again in q-search() along with >>qnodes++, but changed to put the nodes++ in search after the if(depth<=0) >>return(q-search... > > >I am not sure how you would double-count. I don't call search unless there >is depth remaining, otherwise I call quiesce() instead. IE for one node, I >don't get into search and then hand that node off to quiesce, I go to quiesce >directly. > >you are correct that you should inc in both, of course. I was not testing in search like if (depth>0) search() else quiesce() at all recursive calls to search. As I reacall, Crafty used to have a macro for this at one point. I test at the start of search as well (and will likely leave it for peace of mind), but testing before each call is even better to avoid a wasted call to search that in turn simply calls quiesce anyway. Thus my less efficient implementation was double counting the first level call to quiesce since nodes++ is done for each search and in quiesce nodes++ and qnodes++ is done. I should have read the other post mentioning qnodes and not complicated things, but I just found the double counting this week.
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.