Author: Robert Hyatt
Date: 11:39:56 08/12/99
Go up one level in this thread
On August 12, 1999 at 12:57:40, blass uri wrote: >On August 12, 1999 at 09:47:51, Robert Hyatt wrote: > >>On August 12, 1999 at 08:32:02, Vincent Diepeveen wrote: >> >>>On August 12, 1999 at 01:33:35, Dave Gomboc wrote: >>> >>>>On August 12, 1999 at 01:10:14, blass uri wrote: >>>> >>>>>This example shows that the null move is not a very good idea. >>>>>If you need more 3*2=6 plies to see the right move and there is no >>>>>zunzwnag(playing no move cannot help black) then you are not close to see >>>>>everything to depth n-r with null move. >>>>> >>>>>I thought depth n when you use null move with R=2 means that except for >>>>>zunzwangs you analyse everything to depth n-2 and I see that it is not the case. >>>>> >>>>>Uri >>>> >>>>Well, null-move is as you described, but what is shown above is "recursive >>>>null-move", which seems to be pretty popular, and some people abbreviate >>>>this as simply "null-move". It is the sort of search algorithm that gives >>>>people a happy feeling inside about how deep they are searching, but leaves >>>>holes for programs like DB to drive a truck through. Of course, Bob has pointed >>>>this out more than once before. Consider how much is being chopped out of some >>>>13 ply search, and you might agree that Bob isn't just being stubborn: there's >>>>actually quite a big difference in coverage. >>>> >>>>Of course, those who use recursive null-move are making the reasonable gamble >>>>that the extra coverage isn't beneficial at their search speeds. >>>> >>>>Dave >>> >>>Now this is pure BS. Even at 5 0 Deep blue gets kicked silly by a laptop. >> >> >>total bullshit. >> >>no counter-argument needed. When was the last time you beat my program? My >>program can't beat Cray Blitz. Cray Blitz couldn't beat Deep Thought. Blitz >>or tournament. > >I am not sure if your program cannot beat Cray Blitz. > >Did you try to play with your latest crafty against the same cray blitz that >could not beat Deep thought? We did (not myself, but a member of the CB team with Crays in every room where he works) play lots of games. CB was winning about 9 of every 10 games. Note that this was a 'reliable' version, and not one of the versions we wanted to try in ACM events but which often blew up in unexpected ways due to lack of testing.) For the record, our most reliable version (prior to singular extensions) was very 46e, which is what we used for playing about 50 games. version 47f was actually stronger, but the singular extensions code broke in the last ACM event, and even though I had fixed it, I was not sure enough that it was 'safe'. > >there are 2 reasons for my doubts if cray blitz is better than crafty. > >1)Cray blitz has bugs and lost a game against wchess because of it and I did not >see impressive results of cray blitz against computer programs at tournament >time control(I expect parallel crafty to do better results against the same >opponents). > Check out the following tournaments: 1983 wccc event. 4 wins 1 draw. Including beating the then world champion Belle and drawing Nuchess. 1984 ACM tournament, where we beat the #2, #3 and #4 finishers in that event. In fact, we had won before the final round started our pairings were so tough. 1986 WCCC rounds 3, 4, and 5. Those were all highly regarded games. And most of those events included 'micro smashes' here and there. >2)Crafty may have better positional understanding relative to cray blitz because >you learned to change crafty from experience of many games in ICC when you had >not the same experience with cray blitz. They are _very_ similar. Most of crafty's eval came from Cray Blitz, except for things that I can't do due to them being too expensive on a non-vector machine. Cray Blitz actually has a _bigger_ eval than Crafty, not smaller. > >Uri
This page took 0.01 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.