Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: next deep blue

Author: Robert Hyatt

Date: 20:54:06 01/28/00

Go up one level in this thread


On January 28, 2000 at 23:50:45, Robert Hyatt wrote:

>On January 28, 2000 at 17:35:56, Robert Hyatt wrote:
>
>>On January 28, 2000 at 10:50:40, Dave Gomboc wrote:
>>
>>>On January 28, 2000 at 04:12:38, Jeremiah Penery wrote:
>>>
>>>>On January 28, 2000 at 01:35:20, Dave Gomboc wrote:
>>>>
>>>>>On January 27, 2000 at 21:20:46, Christophe Theron wrote:
>>>>>
>>>>>>On January 27, 2000 at 14:21:19, Dave Gomboc wrote:
>>>>>>
>>>>>>>On January 27, 2000 at 13:00:39, Christophe Theron wrote:
>>>>>>>
>>>>>>>>On January 27, 2000 at 00:06:19, Dave Gomboc wrote:
>>>>>>>>
>>>>>>>>>On January 25, 2000 at 21:26:19, Christophe Theron wrote:
>>>>>>>>>
>>>>>>>>>>On January 25, 2000 at 13:33:36, Dave Gomboc wrote:
>>>>>>>>>>
>>>>>>>>>>>It seems weird to me that when Ed Schroder says Rebel does better without
>>>>>>>>>>>null-move than with it, people believe it, but people criticize the DB team for
>>>>>>>>>>>not using it (e.g. from your text above: "by not using a good, known pruning
>>>>>>>>>>>system...").
>>>>>>>>>>
>>>>>>>>>>If the DB team did not have enough time, they could simply take the null move
>>>>>>>>>>algorithm because there is documentation available on it.
>>>>>>>>>>
>>>>>>>>>>However null move is not the final say. Rebel does very well with ANOTHER
>>>>>>>>>>pruning system. Junior does very well with ANOTHER pruning system as well. And
>>>>>>>>>>there are other programs that do fine without null move, one of which I know
>>>>>>>>>>very well.
>>>>>>>>>
>>>>>>>>>Yes, and DB does very well with ANOTHER pruning system too.  What's your point?
>>>>>>>>>
>>>>>>>>>Dave
>>>>>>>>
>>>>>>>>
>>>>>>>>My point is that they claimed that they did not use one. Several of us, looking
>>>>>>>>at the apparent branching factor shown in the DB log files, have doubts about
>>>>>>>>this.
>>>>>>>>
>>>>>>>>
>>>>>>>>    Christophe
>>>>>>>
>>>>>>>Where did they claim that they did not use one?  In their published work they
>>>>>>>clearly stated the contrary.
>>>>>>>
>>>>>>>Dave
>>>>>>
>>>>>>
>>>>>>What did they say?
>>>>>>
>>>>>>
>>>>>>    Christophe
>>>>>
>>>>>Argh, I can't find the paper!
>>>>>
>>>>>In the software, there are some full-width plies, then some selective plies.  In
>>>>>the hardware, there are some full-width plies again.
>>>>>
>>>>>That is from memory though.
>>>>>
>>>>>Maybe Ernst can pull out his copy of Search Control in Deep Blue?
>>>>
>>>>I think the 'selective plies' are the extensions.  The full-width plies at the
>>>>end of the search, in hardware, coincide more with the evaluation function than
>>>>anything.  Rather than doing a static evaluation at the end of the search, they
>>>>did that evaluation based on a 4-ply search done in hardware, to lessen any
>>>>error involved.
>>>
>>>That was not the impression I got from reading the article.  Normally when one
>>>speaks of selective plies one speaks of pruning some moves out.  I'm not saying
>>>that they're not extending some lines like crazy, though.
>>>
>>>Also, the hardware didn't have to do a 4-ply search.  It was configurable.
>>>Looking at the logs, where you can clearly see the (3), (4), (5), (6) after the
>>>plies, it makes me want to guess that the number inside the brackets is the
>>>full-width depth that the hardware searched.  But that's just speculation.
>>>
>>>Dave
>>
>>
>>No.. that I know didn't happen.  I can explain the technical details but they
>>are messy.. but the deal is that the chess hardware has a pretty narrow range
>>of depths it can search to without (a) searching too quickly for the SP to
>>keep it busy or (b) searching so slowly that the SP is waiting all the time
>>and not helping.
>>
>>Hsu settled on 4 and stuck with it so far as I know, although it is possible
>>this was bumped up in endgames (I haven't seen anything suggesting that it was
>>however...)
>
>
>My guess was wrong.  From the DT team, the number _is_ the hardware depth.
>In deep thought, it apparently could vary from roughly 3 to 5 depending on
>whatever, although the shallower the better, since there was no hashing in
>the hardware and the resulting search trees would get bigger quickly.
>
>But now the actual search depth should be treated as the sum of the two
>numbers, something we didn't know for certain before.


Just looked.  And this gadget is beginning to look even more frightening that
I first thought.  Typical software search depth was 10, and at that depth
the hardware depth always seemed to be 6.  For a total depth of 16 plies, which
is a monster.  Many searches were 11/6 or 17 plies.

Think about that in the concept of todays micro programs...  that is scarey...



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.