Author: Vincent Diepeveen
Date: 11:52:13 10/18/02
Go up one level in this thread
On October 17, 2002 at 22:22:29, Jeremiah Penery wrote: >On October 17, 2002 at 20:58:39, Uri Blass wrote: > >>On October 17, 2002 at 18:14:56, Jeremiah Penery wrote: >> >>>On October 17, 2002 at 13:09:58, Vincent Diepeveen wrote: >>> >>>>On October 17, 2002 at 06:46:50, Jeremiah Penery wrote: >>>> >>>>You don't put on your 20GB harddrive box that it is >>>>a 14(6) GB harddrive. >>>> >>>>Instead you sell it as 20GB. >>> >>>If you have 2 partitions on it, you can split it that way. Or do you think you >>>should write that one partition is 20GB and the other is 6? >>> >>>>You are a complete idiot if you believe that 12(6) is 18 ply. >>> >>>I don't think I said anywhere that I believe 12(6) _is_ 18 ply. But I'm pretty >>>sure it's more than 12. >>> >>>>There is so many statements now which if you litterary interpret them >>>>means that 12(6) = 12 ply. >>>> >>>>The clearest statement is the advances of artificial intelligence. >>>> >>>>It seems to me that the only person who tries to present things better >>>>than they are is Hsu. >>> >>>Hsu is the one who said 12 ply - how is that trying to present things better >>>than they are? More like worse. >> >>Based on other things that he said(deeper blue is better than deep Fritz) it >>seems that Hsu is not the person who tries to present things worse than they are >>so if he says 12 plies he cannot mean more than what is considered as 12 plies >>by people except extensions. >>> >>>>The documents which clearly state non-ambigious 12 ply are written >>>>by 3 persons and proofreaded. >>> >>>The paper that states '12.2 ply' says that is the iteration depth, which is >>>software. >> >>No >> >>I did not see where they said that the iteration depth is the software >>I remember that they even said that iteration x means depth >>x-4 in the software(I understood that it was about deep blue Junior and not >>about deeper blue but I guess that the word iteration has similiar meaning in >>deeper blue). > >Ok, i-4 is the MINIMUM software depth. In Crafty (or Fritz, or Diep, or >anything else), when it reports a depth of 12, the MINIMUM can be several plies >lower due to null-move (or some other) pruning. Should we now say that Crafty >is really getting 8 plies instead of 12 because of this? Well, you can if you >want. please don't get on cocain too fast. transposition cutoffs (which happens in my software way more than they happened in deep blue of course) and nullmove cutoffs are pretty correct cutoffs. that isn't taken into account for 'iteration' depth. In all programs iteration depth means the same. Just it happens to be the case that some programs use fractional search depths and thereby either 'add' units to the length of a move (which otherwise is 1 ply deep) like junior is doing all the time. Or you can decrease the number of units, which means the move gets extended. However this is not a fair way to look at the facts. Of course junior is doing it so drastic that it overwhelms completely the iteration depth. positional moves are like 3 ply (3 * plyunit) possibly, that's outpowering things. In deep blue we know they used for captures and such tactical moves fractional search depth and the extension system is further explained in the papers. So we can *assume* that deep blue was tactical powerful. Most lines show clear extensions at nearly all tactical lines it shows. I do not doubt it was a powerful machine, even to todays standards, to solve tactics at 3 minutes a move. If you look to how its extension system works and what it is doing in quiescencesearch. yet that doesn't take away the fact that 12 ply iteration depth for the positional lines means obviously that it didn't get deeper than 12 ply when showing 12(6). that's the biggest weak spot of it simply in terms of search. We do not know how Hsu did his forward pruning in hardware. Whether it was 1 ply from that 12 ply or 2 ply or 3. A good guess would be 1 or 2. Most likely only when evaluation was already outside some kind of big windows, then only trying tactical moves. That's what he described at the microsoft meeting. The exact implementation of that forward pruning doesn't interest me much. Even the claim it reduced him 90% of the total nodes in hardware. That's all so very logical, when not using nullmove, that the only thing that reduces your depth, that it prunes *a lot*. But iteration depth of deep blue was very clear. it was on average as claimed (a bit more positive even than it was in reality, as endgame is counted to this too and i basically do not care for search depth in endgames at all too much; i care for middlegame and there it was like 11.0 ply or so on average). On average the thing got 12.2 ply in iteration depth. That's very impressive for a fullwidth machine in fact. If you have made yourself software run on so many processors and if you some years ago toyed yourself with just fullwidth engine, then you *know* how hard it is to get 12.2 ply fullwidth. >It doesn't change the fact that DB's iteration depth is meaning software. >Whether the number is i-4, i, or i+4 doesn't change that fact.
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.