Author: Graham Laight
Date: 01:04:45 02/03/02
Go up one level in this thread
On February 02, 2002 at 12:01:53, Robert Hyatt wrote: >As I mentioned in the past, on more than one occasion they had said "we did >an N ply search there" and when I saw their output it said N(X). I didn't give >it any thought until I finally asked "what is that X value?". And Murray or >Hsu explained it. I then asked "why did you say N ply?" and one of them >responded "the hardware search is much simpler than the real software search >and we tend to think of that as a unique form of a 'static evaluation' even >though it does include a search + capture search." > >IE I guess that if I had had that kind of design, I would probably shorten >the reference to depth to just include the software depth since (at least for >deep thought) it seemed that the hardware depth was generally statically fixed >to 4 or 5 (I don't recall exactly which I saw). DB seemed to be a bit more >dynamic in that they could do an 9(5) iteration and then either a 10(5) or >9(6) iteration next. > >That was why I specifically asked the question about the depth, because I had >seem them use "depth" in what I considered to be an ambiguous way and after we >all started examining their logs it seemed important to understand their stuff. > >There are other "hints" about this. Remember that the chess hardware has >absolutely no way to return a PV move of any kind. So _every_ PV you see >produced by their program was absolutely searched by the software part of >the machine _only_. When you see 10(6) you can now _know_ that in addition >to whatever PV you see (and it may well not be a full PV as they could only >get the PV from the hash table which is not 100% reliable in producing moves >particularly near the end of a PV) there _must_ be at _least_ 6 more PV moves >(non-captures) plus the q-search moves. The Chess processors didn't do SE, >but it did do classic extensions like in-check and recapture, because it was >copied directly from Belle which did the same things. > >In short, _every_ PV you see has at least (N) more non-capture moves on the >end of it, plus their q-search... If you look at their output carefully, >you begin to get the idea of their search, because it is _definitely_ a fact >that the hardware provides no PV information of any kind, period. Look at the >PVs you see and when you realize that they can only come from the X(Y) (the X >part only) part of the search, things begin to make sense. Vincent sees a >8(4) depth and 12 moves and says "aha, that is obviously a 12 ply search and >PV" even though it should now be obvious that that 12 ply PV came from the 8 >part of the software search, not from the 4 part of the hardware search. If the chess chips did not produce a PV (Position Valuation), then what was their output, and how did it contribute to the move selection, please? -g >So far I have not seen anything that is inconsistent with anything they have >said, _except_ that their branching factor seems lower than I would expect. >This _could_ be caused by using alpha/beta triggered extensions that get >hit more on PV searches and less on other searches. Or due to the not-very-well >understood "hardware futility pruning" done in the chess processors. Clearly >something is going on in their search that is "interesting"...
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.