Author: Robert Hyatt
Date: 19:38:17 11/23/98
Go up one level in this thread
On November 23, 1998 at 17:49:30, Amir Ban wrote: >On November 23, 1998 at 17:14:28, Robert Hyatt wrote: > >>On November 23, 1998 at 11:50:01, Amir Ban wrote: >> >>>On November 23, 1998 at 09:37:25, Robert Hyatt wrote: >>> >>>>On November 22, 1998 at 11:49:54, blass uri wrote: >>>> >>>>> >>>>>I did not ask for all the tree but only the tree up to the point that my >>>>>programs can see by search of 3 minutes that black has at least 1 pawn >>>>>advantage. >>>>> >>>>>This is clearly less positions >>>>>because if in the leaves it is -2.xx then Junior can see some moves before the >>>>>leaves that it is -1.xx >>>> >>>> >>>>ok... rather than 10 million pages, it might only be 1 million pages. How >>>>would we get those to you? :) >>>> >>> >>>I wonder how many people reading the last few posts of this thread have been >>>reminded of the story of the King's New Clothes. >>> >>> >>>>what you are overlooking is the point that junior (and all the other programs) >>>>look at a fat, shallow tree. >>> >>>I am quite sure that the opposite is true. All PC programs have a much smaller >>>effective branching factor than DT/DB. This is because they all do forward >>>pruning, many of them aggressively, while DT/DB did none, and they do >>>extensions, most at least as much as DT/DB, and at least in Junior, much more >>>aggressively than DT/DB. >>> >>> >> >>I'm going to try to keep this simple. Here is a point-blank question: if you >>really believe that nonsensical statement you wrote above, then how can you >>reconcile that with a program that is searching at least 1,000 times faster >>than you, yet only gets to depth 10-11 in the game? If they are not extending >>far more than you could ever hope to then exactly *what* are they doing with >>that factor of 1,000? And remember that they have a pretty simple quiescence >>search and they toss out bummer captures as well, so the work is *not* in >>looking at zillions of captures. >> >>now, in light of that, if you believe that "you extend much more aggressively >>than they do" then *where* are those nodes of theirs *going*??? You have a >>printout to look at. Ought to be able to answer that somehow... >> > >I answered this in the first paragraph you snipped. > >Instructions to the reader: To get the simple answer to Bob's simple question, >go up two posts, and look at the question and answer that followed the one >above. > > >> >>And then we will return to the definition of "hyperbole"... > >Amir I'll play that game. we are talking about a factor of 1,000. You implied this could easily be explained by their not doing null-move or other forward- pruning tricks? That is your explanation? I'd like to suggest you break out the calculator. Null move does *not* reduce the search by a factor of 1,000. Not by a factor of 100. Generally not by a factor of 10. So, I re-ask... if they are only searching to 10 plies, *why* does it take them so many nodes to get to ply=10. Want some math? perfect tree ought to be 2*38^5 moves. They search that many nodes in under 1 second (that is about 160M nodes). Most agree that current programs search within a factor of two of the optimal tree size (references available if needed). so lets say they can fully search this tree in 1 second, even assuming imperfect ordering... Now, again, I'd like to ask the *same* question again, and this time get a *reasonable* answer: If they take (say) 5 minutes to do a 10 ply search, at 250M+ nodes per second, that is over 300X the number of nodes a full-width search to depth=10 should search. If you factor in a q-search that is the same size as the full-width part, we have a missing factor of 150 to account for. I say that is *all* search extensions. And I say that is *far* more than any of the rest of us do in terms of extensions. How *else* would you characterize this? Hint to the reader: previous answer was based on "air". and wasn't acceptable. Question was *not* answered as implied...
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.