Author: Andrew Williams
Date: 01:48:32 05/20/99
Go up one level in this thread
On May 19, 1999 at 19:23:43, Will Singleton wrote: >On May 19, 1999 at 17:53:49, Andrew Williams wrote: > >>On May 19, 1999 at 15:01:20, KarinsDad wrote: >> >>>On May 18, 1999 at 21:20:10, Will Singleton wrote: >>> >>>>On May 18, 1999 at 09:29:52, Andrew Williams wrote: >>>> >>>>>On May 18, 1999 at 09:08:45, Steffen Jakob wrote: >>>>> >>>>>[snip] >>>>> >>>>>> >>>>>>A different thing: currently I experiment with using the number of searched >>>>>>nodes for each move in the root search to sort the move list. I know that Bob >>>>>>and Bruce are doing this too. Anybody else? Any experiences? >>>>>> >>>>>>Greetings, >>>>>>Steffen. >>>>> >>>>>I tried this and it didn't help me much. My approach is to retain the list >>>>>of moves at the root and keep track of the score for each one. This is a bit >>>>>more complex, but it works well for my program. >>>>> >>>>>Andrew >>>> >>>>I use the method given by Steffen. Seems to result in better move ordering for >>>>me. btw, how do you get the score for root moves that aren't the best move? >>>> >>>>Will >>> >>>I do not know how Andrew does it, but you could fake your program out into >>>thinking the root node was each position one ply down (temporarily) and getting >>>an "actual" score for each move that way (but this may mess up a lot of your >>>other search code). >>> >>>KarinsDad :) >> >> >>Hi Will, KD. >> >>What I did was just to use the score that came back. And of course, this >>shouldn't work very well. But it's just one of those things in my program >>that seemed to work even if I knew intellectually that it ought not to. >>BUT I've just reimplemented the nodes trick. I tried a few test sets and >>got no real difference. A 120-game match against my original program, >>however, resulted in a big win for the nodes version (which I now suspect >>I implemented incorrectly before). I'll try a longer game with longer times >>later on to see if I can confirm this result. So once again, a discussion >>in CCC results in a potential improvement for PostModernist. Thanks, Will. >> >>Andrew >> > >I've had similar results, that is, improvements don't necessarily show up on the >test suites. I'm currently running a test version of Amateur against the >current one, testing the effect of reducing or eliminating extensions on null >move branches. Too early to tell yet. > >Also, do most folks call eval() at interior nodes, for the purpose of obtaining >a score for various selective cuts? > I don't in PostModernist. My evaluation is very inefficient, so anything that requires this would have to be a very big improvement to get over the cost of the evaluation. >Will > > >> >>PS allow me to go off-topic for a moment >> >>MANCHESTER CITY 1 WIGAN ATHLETIC 0 (CITY WIN 2-1 ON AGGREGATE) >> >>WEMBERLY, WEMBERLY, WE'RE THE FAMOUS MAN CITY AND WE'RE GOING TO WEMBERLEY! >>(repeat ad nauseum at the top of your voice) >> >>Moderators, you can moderate this message into oblivion, if you like. >>I don't care because City are going to WEMBLEY! >> > > Hmmmm... so is it Wembley or Wemberly or Wemberley? ;) I remember I used to >get excited about baseball, but when so-so pitchers get $100,000,000 multi-year >contracts, then make excuses for not performing, well... anyway, you're right, >it's off-topic. I think we ought to have an off-topic section, as someone has >suggested. > >Will It's Wembley, but it's pronounced (and spelled) differently if it's being shouted very loudly (that's my excuse anyway). I have a very sore throat this morning - I wonder why. Andrew
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.