Author: Robert Hyatt
Date: 21:55:59 09/24/01
Go up one level in this thread
On September 24, 2001 at 22:07:48, Vincent Diepeveen wrote: >On September 23, 2001 at 23:03:12, Robert Hyatt wrote: > >>On September 23, 2001 at 19:33:42, Vincent Diepeveen wrote: >>> >>>Thesis? >>> >>>comon you think i'm so stupid to publish what i did? >>>You think what i made was made and based upon >>>one afternoon experiment, like all the other parallel numbers in >>>ICCA are ? >>>>Dave >> >>\ >>Two things. >> >>1. Your comments are _really_ insulting to everybody posting in the JICCA. >>I can _guarantee_ you that you have not done 1/5 of the work I have done on >>a parallel search. The numbers _I_ posted in the JICCA represented two full >>years of full-time research, doing _nothing_ else whatsoever but trying to >>get the DTS algorithm designed, written, debugged and tested. To say that >>those results represent an afternoon of work means _your_ numbers represent >>about 30 seconds. >> >>2. I don't believe you have some revolutionary new parallel search algorithm. >>You asked way too many questions about DTS a couple of years ago. So don't try >>to hide behind a "I am not talking" wall and try to imply you have done >>something new and more efficient than anything done by anybody else. That is >>ridiculous. > >I'm not saying i invented a new wheel Bob. > >But i'm sure no one combined the wheels i used. I didn't mean to >insult your work. But you write here in CCC with a different view >than when you wrote down your numbers for Cray Blitz. > >Cray blitz gives 1.9 as speedup. Why wouldn't i be able to get 2.0 then >with some further enhancements made to it? I'm not questioning 2.0. I am questioning > 2.0, which is simply not possible except for occasional positions. > >You are a pioneer! > >First there was a man who created an airplane. Then there was someone >who started with that first airplane and >improved that airplane and made it fly faster. > >Inventing the airplane took 2000 years. Improving it to a fighting >machine in WWII took less than 10 years. > >You are saying here that the man who invented the airplane is by >definition having a better airplane? I'm saying he knows at _least_ as much about it, after doing it for 20 years, as someone that has done it for 10% of that time, yes... > >If i insulted you with my JICCA statement, then i sincerely apologize. > >I wasn't out to insult you. I was out to insult the *average* research >in JICCA. > >It is big crap. I also told Jaap that the latest advances in ICCA i found >much better than the big crap which he published in the past. > >But Jaap lives in netherlands. Dutch universities get a bit >more money if they publish more in official magazines. I do not know >other countries. Who am i to blame that they publish? > >Most important statement i want to make is that we live in 2001 now. >Something which worked back in 1986 at a cluster with 1 Mhz transputers >with an i/o speed of 1 MB, that these algorithms look pathetic nowadays. Again, not necessarily. If the connection speed has scaled equally with the cpu speed, then things should work just as well today as they did 15 years ago. Network speeds _have_ kept up. 10 gigabit tests are now active all over the US. 1985 saw 10mbit ethernet. 2001 sees 10gigabit networks. a factor of 1,000. Processors have not come that far by a long shot... > >That doesn't mean that in *those* days it was pathetic. In that respect >i want to insult no one. > >But for example i remember a few years ago, >when everyone was searching 9-13 ply. >Some even 17 ply at the world champs, then someone published how >well hashtables worked with 1,2,4 probes using at most 8 ply depths! > >The APHID results based upon 8 ply depths, that was in 1997. I mean how >many Ghz were all those machines together and then basing conclusions >upon 8 ply? That is simply wrong. In 1997 they had already done the Crafty port and I can assure you that the Jakarta version of Crafty was not doing 8 ply searches. The ACC paper you are looking at might have been 8 plies. But the more recent APHID stuff was _not_ 8 plies. I chatted with them multiple times as they got it up and running and ran test suites... > >The examples are numerous. Most articles with a new algorithm do not >even have pseudo code, not to mention source code link, where their algorithm >is implemented (where's that pseudo code from Hsu about singular extensions, >i couldn't find the pseudo code in advances in artificial intelligence!). The JICCA paper was perfectly clear to me. His original paper on SE was what I used to implement them myself... I believe that is the reference that Kittinger used to implement just the PV-singular extension he worked on in Wchess back around 1996 or so. > >This is why made big compliment for the latest AICG 9, because even though i >disagree with some conclusions, they at least show some source code. > >In short it wasn't my meaning to insult you, but 99% of all parallel >experiments are either completely unreadable, or they are >completely incomplete quoted, or they are showing complete frog results. > >Obviously i didn't want to include your article about DTS here! Then you should be careful of wording. You said "all ICCA papers..." which covers a lot of ground. There has _always_ been some "suspect" stuff in the JICCA. But there has also been a lot of good stuff there as well. It is just a matter of separating wheat from chaff...
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.