Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: next deep blue

Author: Robert Hyatt

Date: 14:59:10 01/22/00

Go up one level in this thread


On January 22, 2000 at 12:43:05, Tom Kerrigan wrote:

>On January 22, 2000 at 08:15:10, Albert Silver wrote:
>
>>That is the whole point. On the one hand I read insistently that DB has more
>>knowledge _by far_ than any program around; yet on the other hand, there is the
>>matter of some of its play that seems to contradict this. One could conclude
>>like many, that this was all a lie, or that other factors were involved. Having
>>no reason to doubt the first, I have to seek an alternative explanation. Such a
>>great amount of knowledge would require a huge amount of testing and adjusting,
>>as a figure of say ten times (or whatever) will require an exponential amount of
>>balancing, since each extra item can effect each and every other item involved
>>in judging and finding moves. Even if Hsu had received the final chips months in
>>advance (which is what I based my theory on), this would still seem to be too
>>little.
>
>Supposedly they had GM Benjamin (?) playing DB every day for months, correcting
>its mistakes. How is this possible without the chips, which supposedly arrived
>at the very last minute? Either they wasted months of a GM's time just screwing
>around, or they were tuning a software version of the evaluation function.
>
>With this amount of work and expertise going into the project, DB should have
>been tuned like a mofo.
>
>But then we hear about how the mobility weight was all screwed up for one
>game... how did a mistake like this happen?? If they'd been tuning the mobility
>weight to exactly the right value for months and months, you'd think they could
>get it right during the big game.
>
>-Tom


Never done any hardware design/testing, eh?  They could test a _lot_ of things
without the new chips. There were lots of things they could _not_ test, for
the same reason.  DB2 wasn't a revolutionary design change from DB1.  It just
included more eval hardware that Hsu thought they needed.  Much of the new
eval was never used at all, due to time constraints. Some things were used and
were poorly tuned due to time.

I added one new term to my eval centered on pawn majority/candidate passed
pawns.  And I wrecked the thing for a month, because that was interacting with
other things in ways I had never imagined.  I could _easily_ see how something
like that could happen.  And I don't _know_ that the queen move in game one was
caused by mobility scoring...  Joel Benjamin said it was.  Hsu and/or Campbell
did not.  So it is possible Joel was "assuming"...



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.