Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Q&A with Feng-Hsiung Hsu

Author: Vincent Diepeveen

Date: 08:31:35 10/14/02

Go up one level in this thread


On October 14, 2002 at 04:53:28, Andreas Guettinger wrote:

>On October 13, 2002 at 21:20:32, Robert Hyatt wrote:
>
>>On October 13, 2002 at 12:00:39, Andrew Williams wrote:
>>
>>>Several people asked:
>>>
>>>Question: What does "12(6)" mean in Deep Blue's logs?
>>>
>>>Dr Hsu said, "12(6) means 12 plies of brute force (not counting the search
>>>extensions & quiescence). 6 means the maximum hardware search depth allowed.
>>>this means that the PV could be up to 6 plies deeper before quiescence."
>>>
>>>Unfortunately questions were being fed through a third-party, so it wasn't
>>>possible to get a follow-up.
>>>
>>>Andrew
>>
>>I saw the question and answer tonight when I got back home.
>>
>>Too bad the question wasn't phrased reasonably to clear up the ambiguity.
>>
>>The above doesn't contradict the email I got from DB team members, nor does it
>>contradict the article they published.  It is just too vague...
>
>
>Seems pretty clear to me what he said. The PV can be up to 6 ply deeper than 12
>ply before quiescence search starts, this make a maximal PV length of 18 (not
>including extensions).
>
>regards
>Andreas


It is very clear. It is 12 ply in total including hardware depth.
That doesn't contradict with anything.

If they would have gotten 18 ply they would have shouted it loudly
around at this planet and sued everyone with commercial interest in
the field of computerchess who said it was 12 ply.

Instead when the subject search depth comes on the agenda, the deep blue
team starts to talk about nodes a second, instead of search depth.

Reality is that in 1997 everyone had a branching factor of between 6 and 10.

They had around 6.0 to 10.0, so that's not so bad as it looks like. They
just got more nodes a second. the improvements in b.f. started later when
hashtables and nullmove started to work well together and R=2 was changed
to R=3 by most.

The statement is very clear when asked whether it was 12 ply or 18 ply
in total the answer was: 12.

Very clear.

And that he better could say so, because it is impossible to search
18 ply fullwidth:

According to Knuth, it is
  2* squareroot(40^18) = 524.288.000.000.000 nodes

Or do you disbelief the proof of Knuth about what the minimal tree is
that you have to search with alfabeta?

If you do not, cut the nonsense please, and get back to normal.

Most searched like 8-9 ply in 1997. They get 12 ply. They didn't get
twice the search depth at all of course. Stories about 18 ply nominal
search depth is complete nonsense.

Even Knuth proved it to be impossible.

Now they have this machien that's just busy getting a hell of a lot of
nodes a second.

You have a statement from Hsu that he cares less for searching deeper
than doing a bunch of extensions.

Bob claims they lost 2 ply to extensions.

And you still believe it is 18 ply?

So without those extensiosn that would be 20 ply.

Want to know what the minimal tree is for 18 ply search when searching
fullwidth without hashtable?

2 * squareroot(40^20) = 20.971.520.000.000.000

Or 2.0 * 10^16 nodes

MINIMAL tree.

So not realistic tree. Add another huge factor to that for the
realistic tree.

How naive are we still today?

WHY THE HELL IS HYATT STILL PUTTING OIL IN THE FIRE?

He'll realize as the best that it's 12 ply and that 12 ply was great in 97.

Whether he got that too at his cray is not important.




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.