Author: Dann Corbit
Date: 22:48:18 03/25/04
Go up one level in this thread
On March 26, 2004 at 01:41:07, Johan de Koning wrote: >On March 25, 2004 at 19:46:25, Robert Hyatt wrote: > >>On March 25, 2004 at 18:10:13, Anthony Cozzie wrote: >> >>>On March 25, 2004 at 16:32:09, Robert Hyatt wrote: >>> >>>>On March 25, 2004 at 14:28:09, Anthony Cozzie wrote: >>>> >>>>>On March 25, 2004 at 13:35:03, Dann Corbit wrote: >>>>> >>>>>>On March 25, 2004 at 10:02:57, Anthony Cozzie wrote: >>>>>> >>>>>>>On March 23, 2004 at 18:18:51, Uri Blass wrote: >>>>>>> >>>>>>>>On March 23, 2004 at 17:28:17, martin fierz wrote: >>>>>>>> >>>>>>>>>On March 23, 2004 at 17:13:46, Aivaras Juzvikas wrote: >>>>>>>>> >>>>>>>>>>On March 23, 2004 at 16:40:46, Renze Steenhuisen wrote: >>>>>>>>>> >>>>>>>>>>>On March 23, 2004 at 16:38:28, Aivaras Juzvikas wrote: >>>>>>>>>>> >>>>>>>>>>>>forgot to mention, i dont try null move on 0 ply >>>>>>>>>>> >>>>>>>>>>>Than what's your test set? >>>>>>>>>> >>>>>>>>>>test set?i just let two versions of my engine play each other a couple of 15 0 >>>>>>>>>>games, the result is either a draw or a win for the one w/o null move, even tho >>>>>>>>>>it searches deeper as i already mentioned >>>>>>>>> >>>>>>>>>"a couple" meaning...? >>>>>>>>> >>>>>>>>>if it's two games, forget it. if it's 10 games, forget it too. start believing >>>>>>>>>it when it's 100 games... >>>>>>>> >>>>>>>>I think that if you do not get improvement with null move based on 10 games then >>>>>>>>there is good chance that you have a bug in the implementation. >>>>>>>> >>>>>>>>Uri >>>>>>> >>>>>>>I have to agree with Uri here. If your program plays weaker with null move >>>>>>>after 10 games, you screwed something up. >>>>>>> >>>>>>>Null move is simply _that big_. >>>>>>> >>>>>>>Getting 2 extra plys should show up long before 100 games . . . >>>>>> >>>>>>I have to disagree with you. You can implement null move incorrectly and still >>>>>>score better in ten games. The reason I say that is because I have seen it. >>>>> >>>>>Please do not put words in my mouth. I said "If A then B", which you corrupted >>>>>to "If !A then !B". >>>>> >>>>>I stand by my statement: If you implement null move correctly, it _will_ win a >>>>>10 game match. 2 ply -> 100 elo -> dominance. Someone can do the math here on >>>>>confidence regions, but I'm very sure the version with null move has a 95% >>>>>chance or better to win. >>>>> >>>>>anthony >>>> >>> >>>>Null-move is better, but it is _not_ 200 elo better. Try it. Both Bruce and I >>>>played some of these matches (null on vs null off). It is more like 50-60 Elo >>>>improvement. >>> >>>I said 100. 200 would indeed be a little much :) But even 60 elo is a pretty >>>clear difference; you would definitely notice that in a 10 game match. >>> >>>>And it _definitely_ isn't "2 plies". There is a great difference >>>>in accuracy between 12 plies no null and 12 plies with null... You go 2 plies >>>>deeper, but you don't outplay the no-null opponent like it would outplay itself >>>>with a 2 ply handicap... >>> >>>I do checks in q-search which mitigates this problem somewhat. >>> >>>anthony >> >> >>If you do checks in the q-search, your non-null 12 ply search will _still_ kill >>your null 12 ply search... > >Yes, it will win a fair share of the endgames. :-) > >But besides that, I think it's wishful thinking. >Like wishing DB's 11 ply is going to kill Fritz' 14 ply. >It's not going to happen. Even if DB would be alive and would accept >the challenge, it still would not be going to happen. > >But back to the issue: we're not talking about 12 versus 10 ply here. >At roughly 100 M nodes / search we're talking about 9 ply full-width >versus 14 ply full-nulled. Assuming both are reasonable tuned, my money >is on a 300 Elo difference. That is, in a computer pool. Less against >humans, more against itself. > >Consider it an invitation to you and to Bruce to run more experiments. >Or at least as an invitation to wonder why that 50-60 Elo improvement >was quite below expectation. Suppose that the thing that was broken with null move was the depth reduction (e.g. it was twice what you wanted with R=4/6 instead of 3/2). Or some other strangeness. Then what sort of result would you expect to see?
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.