Author: Dan Newman
Date: 15:42:18 01/19/00
Go up one level in this thread
On January 19, 2000 at 17:55:50, Robert Hyatt wrote: >On January 19, 2000 at 14:26:25, Ricardo Gibert wrote: > >>On January 19, 2000 at 09:58:32, Robert Hyatt wrote: >> >>>On January 19, 2000 at 01:25:19, Ricardo Gibert wrote: >>> >>>>On January 18, 2000 at 23:20:29, Robert Hyatt wrote: >>>> >>>>>On January 18, 2000 at 18:54:46, Ricardo Gibert wrote: >>>>> >>>>>>On January 18, 2000 at 18:35:52, Robert Hyatt wrote: >>>>>> >>>>>>>On January 18, 2000 at 13:57:54, Dann Corbit wrote: >>>>>>> >>>>>>>>On January 18, 2000 at 12:49:38, Bruce Moreland wrote: >>>>>>>>[snip] >>>>>>>>>>Opinions? Am I all wet? >>>>>>>>> >>>>>>>>>Yes, you are all wet. I will resist the temptation to use a drug metaphor since >>>>>>>>>people seem to be a little cranky about that today. >>>>>>>>> >>>>>>>>>I don't see any reason to suppose that you can't use induction to predict the >>>>>>>>>characteristics of a 25-ply search by examining the characteristics of a 15-ply >>>>>>>>>search. >>>>>>>> >>>>>>>>I know you know a lot more about it than I do, and everyone is in agreement that >>>>>>>>I am wrong. But I still don't understand why. From the plethora of posts I >>>>>>>>have seen here where a program fails to find a move in a test position and it is >>>>>>>>found that it is zugzwang, I presume that it is not terribly rare. Now, >>>>>>>>ignoring NULL moves makes you run so much faster that it almost always a good >>>>>>>>idea. You get a full ply more -- sometimes two (if I understand correctly). >>>>>>>>But it seems to me that NULL move is dodging bullets in the sense that you >>>>>>>>almost never get bitten. But if you ignore thousands of them, maybe one of them >>>>>>>>was dangerous. And if you ignore one million of them, it could be even worse. >>>>>>>> >>>>>>>>On the other hand, I also recognize that there are more than one good pathway >>>>>>>>from most board positions. So perhaps even when it does go wrong, NULL move >>>>>>>>pruning may still pick out a good path most of the time. >>>>>>>> >>>>>>>>I am sure that my supposition is wrong, since so many others think that it is. >>>>>>>>But I still don't understand why. >>>>>>> >>>>>>> >>>>>>>Here is a "hint"> :) >>>>>>> >>>>>>>what makes you think that in a 10 ply search, where there are N zug positions, >>>>>>>that in a search space 10 times bigger there are more than 10*N zug positions? >>>>>>> >>>>>>>That is point 1. Point 2... there _are_ more zug positions overall. But there >>>>>>>are also more non-zug positions. And for a zug position to screw up and then >>>>>>>cause a key score to change is no more probable in a tree with M positions and N >>>>>>>zug positions than it is in a tree with 100M positions and 100N zug positions... >>>>>>> >>>>>>>Everything grows at the same exponential rate... and stay exactly proportional >>>>>>>to each other... >>>>>> >>>>>>I don't necessarily agree with Dan, but there's a fly in your ointment. >>>>>>Everything does not stay proportional. The deeper you search, the more >>>>>>simplified the position is. The more simplified the position gets, the more >>>>>>likely it may be zugzwang. >>>>> >>>>> >>>>> >>>>>That isn't necessarily true. I have seen 100 move games with queens and rooks >>>>>still on the board. And (at least in my case) we can take evasive action to >>>>>recognize some zug positions and not let them become a problem... >>>>> >>>>> >>>>> >>>>>> The character of the search and the topology of the >>>>>>tree does change the deeper you go. The branching factor of the tree changes as >>>>>>the position gets more simplified. The relative value of the pieces changes as >>>>>>the position becomes more open. The King becomes more of an asset than a >>>>>>liability, etc. >>>>> >>>>> >>>>>However, I read his question as from position N, do a 10 ply search and then a >>>>>20 ply search, and the 20 ply search should have more serious null-move >>>>>problems. I don't agree. 10 more plies does not appreciably simplify the >>>>>position in the majority of the pathways.. >>>> >>>>Perhaps, but it breaks your argument all the same. >>> >>> >>>ANd how would that be? From the opening position, I have _no_ chance to reach >>>a pieceless ending in 10 or 20 plies. Until at _least_ 1/2 of the total >>>material on the board is gone, I don't reach 5 piece endings and do EGTB probes. >>>It is also unlikely that I reach a significant number of zug positions either, >>>at least a number large enough to affect the root score, which was the original >>>premise of this... >> >>This is the first mention of an "opening position" in this thread. I agree you >>you can save your argument by reinventing the premises until it works. > > >OK... change that to "middlegame position". Same result. I don't reach >endgame positions very often from middlegame positions. Usually not until >around move 40 or so in a real game. That leaves 40 moves to search with >no regard to null-move failures at all. And if the program is smart enough to >switch null-move off when it is not appropriate, rather than just turning it off >at the root, this is a total non-issue... for _any_ position you care to name... > >So my original statement remains accurate... A bigger tree is _not_ more >prone to null-move failures than a smaller one... In fact it's really the other way 'round. Null-move doesn't work very well if you only do a shallow search. It really needs a good deep search to avoid another sort of failure... -Dan.
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.