Author: Andrew Williams
Date: 05:19:12 09/09/99
Go up one level in this thread
On September 08, 1999 at 18:07:13, Peter McKenzie wrote: >On September 08, 1999 at 06:40:39, Andrew Williams wrote: > >>On September 07, 1999 at 19:10:37, Peter McKenzie wrote: >> >>>On September 07, 1999 at 13:54:40, Bruce Moreland wrote: >>> >>>>On September 07, 1999 at 10:27:06, James Robertson wrote: >>>> >>>>>On September 07, 1999 at 09:01:34, Andrew Williams wrote: >>>> >>>>>>Using this scheme, what would you count if you enter a search node, check for >>>>>>extensions, find there are none and decide to go into the qsearch? One node >>>>>>or two? >>>>>> >>>>>>Andrew >>>>> >>>>>Er.... I think that case is impossible in my program. :) >>>> >>>>This kind of stuff is not impossible using the generic Thompson search/qsearch >>>>model. You can get a case where you call qsearch, notice that you are in check, >>>>and do "return search(...)", which would count a node for "qsearch", and another >>>>node for the recursive call into "search", even though only one node was really >>>>dealt with. This of course depends upon where you put the nodes++, which isn't >>>>present in the Thompson pseudo-code. >>> >>>Seems to me the simplest place to put nodes++ is in the MakeMove function. >> >>I can't really do that as I use my hash table for generating my PV and need >>to use make_move to walk up and down the PV at the end of iterations. I guess >>I could write a stripped down make_move, but I haven't really looked at that >>yet. > >or: > >nodesCopy = nodes; >MakePV(); >nodes = nodesCopy; > >or maybe the solution is worse than the problem :-) > Yes :-). I'm actually quite happy with the way I count nodes, I just wanted to be sure that the number I'm getting is generally comparable with others' figures. BTW Congratulations on the latest version of lambchop. Seems very strong. Andrew >> >>Cheers >> >>Andrew >> >>> >>>> >>>>I think that one node should be counted in cases like these. If you end up >>>>counting a lot of nodes twice, you can be comparing with someone else who has >>>>written essentially the same program, and it looks like you are going faster. >>>> >>>>bruce
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.