Computer Chess Club Archives


Search

Terms

Messages

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

Author: Robert Hyatt

Date: 18:25:16 10/16/02

Go up one level in this thread


On October 16, 2002 at 07:01:27, Vincent Diepeveen wrote:

>On October 14, 2002 at 21:39:43, Dave Gomboc wrote:
>
>Look they got 12 ply in *total*.
>
>If you would have had a chessprogram in 1997 you would be
>not believing they got 12 ply fullwidth.
>
>You don't have a program however.
>
>In 2002 standards are different. If you say now that most
>expensive car in 1910 drove 15 miles an hour, then you would
>not believe it either i bet.
>
>chess progressed faster.
>
>12(6) means 12 ply in total from which up to 6 ply in hardware.
>I hope you realize the number they give to the processors is
>variable.
>
>But you'll never understand that i bet. You will believe any
>vague ambigious statement of them, but of course they will never
>loudly say: "we searched 18 ply fullwidth" as that's a lie.
>
>They got 12 ply but managed all those years to formulate it only
>clear in artificial intelligence proceedings (average nominal
>search depth: 12.2 ply it clearly mentions).
>
>In 1997-1999 it was not so unclear. Bob clearly posted 12 ply
>here. They posted in IEEE1999 very clearly 12 ply,
>but only after 1999, and in 2000, some idiots like you
>who do not know fuck from math made that to 18 ply.
>
>Look Fritz with nullmove and forward pruning and hashtables and
>millions of nodes a second doesn't even get to 18 ply.
>
>And the minimal tree of c*2.8^18 is a lot smaller than 6.0^18
>
>Because WITH forward pruning deep blue achieved b.f. = 6 which is
>very good in 1997.

Why do you keep posting that number?  When the log files _clearly_ show it to be
_wrong_?

Ever think about doing a bit of work before putting your fingers to work???





>
>I remember my own thing from then. I remember schach which WITH nullmove
>and without all those checks in qsearch and WITh hashtable which gets
>also to that 6.0 hardly.


A program doesn't need null-move to get to 6.  That is a pretty common value.

For even plies, raw alpha/beta and nothing else will vary quite a bit, but the
basic
idea is that the effective branching factor is going to hover around 6.0 unless
move
ordering goes south.  Null-move drops that.  Even non-recursive R=1 drops it.



>
>I remember Genius which even today doesn't get 12 ply fullwidth.
>Not even 10.
>
>And that's with just at most 12 ply tactical extensions in *total*
>(including giving checks). A peanut compared of course to what gets
>extended by singular extensions, which extend every capture!
>
>In the paper of 2001 they very clearly show that their average search
>depths WITH extensions and qsearch and without the transpositions in
>software, that those were about 17-18 ply.
>
>That's however something different. That's where all this confusion
>gets from.
>
>Now you claim that's their nominal search depth?
>
>Note my maximum search depth is 128 ply with DIEP.
>
>From that it achieves in an average game about 70 ply.
>
>Majority of lines that do not get nullmove pruned nor transposition
>pruned are way deeper than the nominal search depth of course
>(singular extensions, guess why) especially if i add qsearch to it
>which is unlimited by the way in diep, where DB has a limit of
>32 ply for every line including nominal h-search.
>
>So i must claim in the future that diep gets about 70 ply?
>
>You simply do not see that in 2002 the machine is very outdated but
>the only way they can do as if they still count in todays computerchess
>is by saying they were better than todays software.
>
>Amazing how bad reading their technical documents is giving so much
>confusion. Even a clear statement  with a DOT behind it, it wasn't
>convincing you. Amazing.
>
>You give something that played pathetic chess a status which is not
>realistic.
>
>In 1988 they got 500k nodes a second single cpu with chiptest.
>
>You believe that those 500k nps which got them to 8 ply fullwidth,
>after 3 minuts, that this is a better 8 ply than the 8 ply i get?
>
>No even worse you believe it was 12 ply. As from that 8 ply 4 ply
>was in hardware...
>
>However if you calculate a fullwidth branching factor (in software+hardware)
>from that 500k nps chip to the way better 1997 chips,
>then you will see that 100+ MLN nodes is indeed a minimum to get that
>12.2 ply.
>
>but you of course do not see that. you closed your eyes for anything
>relevant. Let me ask you this. WHAT must hsu say more than the already
>published (and non-published) 5 papers and books and statements he has
>done about his search depth to convince you?
>
>If he says: 12 ply from which 6 ply was hardware search depth,
>that's sufficient for you?
>
>No he said after the question:
>
>was it 18 ply or 12 ply?
>
>he answerred: 12 ply.
>
>DOT behind it.
>
>That's not enough for you.
>
>You really believe that if it was 18 ply, even if he pruned 2 ply
>of it, that he would answer 12 ply?
>
>He would be insane stupid.
>
>He's not. He's a clever guy. He answerred:
>
>EeEk(DM) kibitzes: kib question from ardee: Does "12(6)" mean 12 total ply or
>12+6=18 total ply?  This has the been source of huge arguments for years!
>aics%
>CrazyBird(DM) kibitzes: 12 total in terms of brute force. 6 is just the max
>partition in hardware.
>aics%
>
>So only idiots now believe it's 18 ply. He answerred clearly 12.
>
>
>Best regards,
>Vincent
>
>>On October 14, 2002 at 09:44:06, Vincent Diepeveen wrote:
>>
>>>So it is clear his answer was very clear : it is 12 ply.
>>>
>>>So is it clear now?
>>
>>No, it's 12 ply _in_software_, why that isn't clear to you, I don't know.
>>
>>
>>>Apart from that. For 18 ply you would need to see squareroot(40^18)
>>>nodes to search it using fullwidth. That's a minimal tree, excluding
>>>qsearch and other overhead.
>>>
>>>You really believed all that time that
>>>deep blue searched 40^9 = 262144000000000 nodes ?
>>
>>The 18 ply is not full-width.  Don't be an idiot.
>>
>>>It is very obvious now that one of the deep blue team members has
>>>spreaded misinformation to you and to Hyatt.
>>
>>So now you're calling him a liar?  Vincent, I have heard more than
>>enough out of you over the years, but calling the guy a liar is
>>beyond pathetic.
>>
>>>Note that they spreaded during the match more misinformation. For
>>>example they said that during the KRRPPP KRRPP endgame of kasparov,
>>>where deep blue managed to draw in a neat way,
>>>that deep blue played it perfect because the whole endgame was
>>>in its EGTBs.
>>
>>B.S.  They never said such a thing.
>>
>>Dave



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.