Still waiting on Ed

General discussion about computer chess...
User avatar
thorstenczub
Posts: 593
Joined: Wed Jun 09, 2010 12:51 pm
Real Name: Thorsten Czub
Location: United States of Europe, germany, NRW, Lünen
Contact:

Re: Still waiting on Ed

Post by thorstenczub » Sun Jul 17, 2011 8:03 am

BB+ wrote:The text of the CW email where he lost his cool has been asked for, and it seems CW is OK with its publication. I happen to have it, and have nothing better to do than copy/paste it here. :roll: I might warn that it is perhaps still a bit out of context (maybe Ed can say more):
the obvious person to validate the email (which you could have done in the first place of course) is Ed Schroeder with whom I have been in email contact using this email for several years now. Since Ed suggested me "taking a look here" and he has been in contact with Hyatt (at least) concerning the very slow and actually quite rude process, you might think it hardly even necessary to ask him, but of course, depending on paranoia level, you may also not believe him as real, even though he and I have been famously in computer chess since about 1980.

Hyatt additionally has been in occasional email contact with this address principally at the time I first modded CCC. He usually claims total storage of everything, does he not?

In what sense do you and the unknown HW think you can make time wasting humiliations in this way? Not a very good start imo.

And no, please do not track my ISP, the talkchess capability of giving immediate IP access/knowledge to mods is one reason I never log in there. Read Jeremy Bernstein's account of behaviour between the "mods" Conkie, Banks and various malicious endusers relating to IP addresses and telephone numbers, for just one example - all done with full knowledge of talkchess and the tcadmin. The danger is not from us humble readers but from site owners/administrators and the occasional little hitler tendency of "mods".

his email is completely ok IMO.

BB+
Posts: 1484
Joined: Thu Jun 10, 2010 4:26 am

Re: Still waiting on Ed

Post by BB+ » Sun Jul 17, 2011 12:59 pm

My opinion (with some other background from the email exchanges) was that a (short) temporal ban was warranted, simply to defuse the situation. However, I'm not sure that MarkL's email response (quoted elsewhere) served to do this, and indeed its end effect appears to be to have alienated CW further. Perhaps I could have been more explicit to CW that my personal feeling was that he should re-apply in a week or so. But at this point, I'm not sure it is all worth revisiting the various recriminations, particularly as CW has said [if I am not mistaken] that in the end he was just as happy not having to chase the ICGA's "political" tails around in circles, just to join their Panel.

Adam Hair
Posts: 104
Joined: Fri Jun 11, 2010 4:29 am
Real Name: Adam Hair
Contact:

Re: Still waiting on Ed

Post by Adam Hair » Sun Jul 17, 2011 3:02 pm

thorstenczub wrote:
BB+ wrote:The text of the CW email where he lost his cool has been asked for, and it seems CW is OK with its publication. I happen to have it, and have nothing better to do than copy/paste it here. :roll: I might warn that it is perhaps still a bit out of context (maybe Ed can say more):
the obvious person to validate the email (which you could have done in the first place of course) is Ed Schroeder with whom I have been in email contact using this email for several years now. Since Ed suggested me "taking a look here" and he has been in contact with Hyatt (at least) concerning the very slow and actually quite rude process, you might think it hardly even necessary to ask him, but of course, depending on paranoia level, you may also not believe him as real, even though he and I have been famously in computer chess since about 1980.

Hyatt additionally has been in occasional email contact with this address principally at the time I first modded CCC. He usually claims total storage of everything, does he not?

In what sense do you and the unknown HW think you can make time wasting humiliations in this way? Not a very good start imo.

And no, please do not track my ISP, the talkchess capability of giving immediate IP access/knowledge to mods is one reason I never log in there. Read Jeremy Bernstein's account of behaviour between the "mods" Conkie, Banks and various malicious endusers relating to IP addresses and telephone numbers, for just one example - all done with full knowledge of talkchess and the tcadmin. The danger is not from us humble readers but from site owners/administrators and the occasional little hitler tendency of "mods".

his email is completely ok IMO.
I have to agree with Franklin and Thorsten. It seems to me that the "little hitler" reference has been misrepresented in earlier accounts. It seems to be a general statement and not one that is directed specifically towards the Secretariat.

veritas
Posts: 111
Joined: Thu Jun 16, 2011 2:35 pm

Re: Still waiting on Ed

Post by veritas » Sun Jul 17, 2011 5:47 pm

Adam Hair wrote:
thorstenczub wrote:
BB+ wrote:The text of the CW email where he lost his cool has been asked for, and it seems CW is OK with its publication. I happen to have it, and have nothing better to do than copy/paste it here. :roll: I might warn that it is perhaps still a bit out of context (maybe Ed can say more):
the obvious person to validate the email (which you could have done in the first place of course) is Ed Schroeder with whom I have been in email contact using this email for several years now. Since Ed suggested me "taking a look here" and he has been in contact with Hyatt (at least) concerning the very slow and actually quite rude process, you might think it hardly even necessary to ask him, but of course, depending on paranoia level, you may also not believe him as real, even though he and I have been famously in computer chess since about 1980.

Hyatt additionally has been in occasional email contact with this address principally at the time I first modded CCC. He usually claims total storage of everything, does he not?

In what sense do you and the unknown HW think you can make time wasting humiliations in this way? Not a very good start imo.

And no, please do not track my ISP, the talkchess capability of giving immediate IP access/knowledge to mods is one reason I never log in there. Read Jeremy Bernstein's account of behaviour between the "mods" Conkie, Banks and various malicious endusers relating to IP addresses and telephone numbers, for just one example - all done with full knowledge of talkchess and the tcadmin. The danger is not from us humble readers but from site owners/administrators and the occasional little hitler tendency of "mods".

his email is completely ok IMO.
I have to agree with Franklin and Thorsten. It seems to me that the "little hitler" reference has been misrepresented in earlier accounts. It seems to be a general statement and not one that is directed specifically towards the Secretariat.
i see probable disrespect in use of surnames without title or christian name but its something certain" types" in the U.K. do ( gets right up my nose ) , but other than that i find his mail perfectly O.K , even reasonable

mjlef
Posts: 43
Joined: Thu Jun 10, 2010 6:51 pm
Real Name: Mark Lefler

Re: Still waiting on Ed

Post by mjlef » Mon Jul 18, 2011 10:14 pm

BB+ wrote:My opinion (with some other background from the email exchanges) was that a (short) temporal ban was warranted, simply to defuse the situation. However, I'm not sure that MarkL's email response (quoted elsewhere) served to do this, and indeed its end effect appears to be to have alienated CW further. Perhaps I could have been more explicit to CW that my personal feeling was that he should re-apply in a week or so. But at this point, I'm not sure it is all worth revisiting the various recriminations, particularly as CW has said [if I am not mistaken] that in the end he was just as happy not having to chase the ICGA's "political" tails around in circles, just to join their Panel.
Within a day I sent this email to Chris, offering to let him be on the panel, provided he would be more polite. But Chris never responded to this offer. Please also note I offered to call him on a land line (and some people scanned in ID cards to prove identity), but he never responded to these offers either.

Mark

Chris,

I do not know what to say. I do not think we have been in any way "rude" to you. Yes, the approval process has been slow for you and others we do not know well. But since we have several examples of people falsely claiming to be others and in the process stealing source code, requesting extra information is totally reasonable. If anything, you have been rude. Just look at the snarky writing below! "wasting humiliation"? "occasional little hilter"? Do you really think acting like this is in your best interest?

David Levy and Harvey felt we should approve you. I have been waiting on Bob Hyatt's decision. I was going to approve you as well, but the histrionics of your emails and sheer paranoia suggest I should not decide in your favor. We need professional, reliable people on the panel. Are you willing to change and be more polite to serve there?

Mark

Prima
Posts: 328
Joined: Tue Dec 14, 2010 6:12 am

Re: Still waiting on Ed

Post by Prima » Tue Jul 19, 2011 3:45 am

mjlef wrote:Within a day I sent this email to Chris, offering to let him be on the panel, provided he would be more polite. But Chris never responded to this offer. Please also note I offered to call him on a land line (and some people scanned in ID cards to prove identity), but he never responded to these offers either.

Mark
Oh, this must be the part conveniently omitted by those against the ICGA rulings on Rybka/Vas' unethical behaviour. Not that I believed their ICGA-is-bias theory.

From the talks in Rybka forum, most (supporters of Rybka's illegal & unethical acts) stated directly or indirectly the ICGA was biased in its ruling, including the removal of CW, the "only Vas supporter" from the panel.

hyatt
Posts: 1242
Joined: Thu Jun 10, 2010 2:13 am
Real Name: Bob Hyatt (Robert M. Hyatt)
Location: University of Alabama at Birmingham
Contact:

Re: Still waiting on Ed

Post by hyatt » Tue Jul 19, 2011 6:52 am

Prima wrote:
mjlef wrote:Within a day I sent this email to Chris, offering to let him be on the panel, provided he would be more polite. But Chris never responded to this offer. Please also note I offered to call him on a land line (and some people scanned in ID cards to prove identity), but he never responded to these offers either.

Mark
Oh, this must be the part conveniently omitted by those against the ICGA rulings on Rybka/Vas' unethical behaviour. Not that I believed their ICGA-is-bias theory.

From the talks in Rybka forum, most (supporters of Rybka's illegal & unethical acts) stated directly or indirectly the ICGA was biased in its ruling, including the removal of CW, the "only Vas supporter" from the panel.

You can find a _lot_ of nonsense there...

diep
Posts: 3
Joined: Thu Jul 28, 2011 4:11 pm
Real Name: Vincent Diepeveen

Re: Still waiting on Ed

Post by diep » Thu Jul 28, 2011 4:47 pm

Hello!

I understand the thoughts of Ed very well. First of all he understands something that all strong programmers understand
and that's that all this huge progress is impossible to achieve without huge teams that share a lot of information.

However i feel Ed attacks the wrong person. You shouldn't attack Bob there. Sure he also avoids being online and chatting
over there. Sure the last source code posted is 23.4 from nearly a year ago and each time i have problems compiling the code
over here using gcc and visual c++.

Yet the feeling that Ed has is true: sure a lot of cash from the government agencies has been thrown into computerchess.
No they didn't support me. I did do it all myself.

Ed more than others realizes how the different projects that seemingly progress now are in fact government projects where
a lot of people put effort in. To quote Stefan MK: "if you get back all those massive testing results, then you're not
finished yet, the hard work just starts then". With the obvious and provable amount of programmers that contributed to
different projects (you can see huge difference in editting styles in the different engines) not to mention all the parameter
tuning, you need even more people. Managers, teamleaders and persons who simply only maintain the source code branch.

Huge overhead in manpower.

Ed realizes all this. He has been a professional in computerchess. Please don't throw mudd at him.

Do things less sneaky i'd argue.

As for the reduction type idea. This is pretty trivial if i may say so for this type of researcher that works onto these projects.
The first time i tried reductions was around end 1998. The world champs 1999 version of Diep used reductions.

It didn't reduce the next type of moves:
a) important pawn pushes such as to the center and passed pawns
b) attacks to higher pieces (like a pawn attacking a piece or a knight attacking a bishop
c) captures
d) hashtable move

With exception of the hashtable move all the above moves only didn't get reduced when the actual move itself
doesn't lose material.

This is old 1999 code of diep to be honest and it worked ok for blitz at the time but crap in general.
In 2000 i kicked out all extensions again as well as all reductions and instead investigated Conspiracy Number Search
in depth. Also implemented this in diep. That took a lot of time. I have to really do this statement about it:
the researchers who also investigated similar searches, i tried to communicate with them. Yet in reality i got 0
emails back. This where i did do 6 months worth of effort to investigate CNS. Talking about it was impossible,
the guys in question kept their mouth shut and didn't share any results/ideas/thoughts they had on it.

In 2004/2005 i restarted the quest after searching deeper in a selective manner using reductions. This time variations
of LMR. Not sure it already had that name.

An obvious thing i also tried was exactly what is in crafty 23.3, the reduction. I did communicate this to other authors.
My biggest amazement was not so much that it didn't work for Diep - the biggest amazement was that it reduced the
total search tree with so little nodes.

This is not so trivial that it doesn't reduce much; if you first reduce the entire search by 1 ply and then only in some
cases research it, and alternated by trying specific moves always without reducing or not at all, one would argue
you can search really a lot deeper. Yet that wasn't the case. Maybe 20% up to 50% reduction in nodes - too little for the
huge risks.

This for taking the huge risk of missing the entire line. Another experiment i did do to search deeper, from which i
feel it still might work with more additional work for Diep, failed completely, just weeks before the world champs 2005,
namely forward pruning experiments. I posted before about that onto CCC.

One of the experiments i did do, and also played games with, was always reducing by 1 ply except for the hashtable move
and mixtures of the above rules.

Another interesting attempt is first determining with a very shallow search which moves are the best in this position,
if the margin is big then you can reduce all moves except 1, or simply all moves. and variations of that reducing 1 or 2 ply.

It all didn't bring much for Diep.

Careful statistics done onto Fruit soon revealed why history reductions worked so great for it.
In positions where a specific move X is best, if i turned on reductions in diep, diep needed up to 5 ply extra
to find that move X, meanwhile reductions winning 3 ply of search depth on average.

This whereas Fruit 2.1 with history pruning turned on, lost positional spoken at most 1 ply search depth, usually even 0,
with its reductions. The simple explanation has to do with Fruit's evaluation which is largely based upon its piece square
tables. So its history table to a large extend reflects its piece square table, which makes it nearly 100% safe to do those
reductions for non-tactical moves.

For brain dead evaluation functions it's very easy to forward prune without missing too much within its search that a
brute force search wouldn't miss. I'm convinced that more knowledge in evaluation is very useful (besides that of course
the big progress of the past 10 years is the improved evaluation functions of engines), yet if we speak about search,
you also must do a huge effort for search to adjust the search to that more knowledgeable evaluation function,
in order to profit from it.

Add to that, that if you get more nps, that branching factors improved really a lot past years. So where in 90s diep was
20+ times slower in nps than the fast beancounters, that difference is "only" factor 10 now, yet a factor 10 is a lot more plies
by now than factor 20 back then.

Additional if i may refer back to my 2005 attempt to forward prune last few plies in Diep using a fast 'brain dead search'
to avoid big prune errors, that still doeosn't work for Diep, whereas todays rybka & clones are forward pruning last 3.5 plies or so
and i guess crafty is doing some futility type stuff last few plies in a cheap manner as well.

This gives quickly a few plies extra whereas i can't do that with Diep.

That results in massive search depth differences from which you can wonder whether that's what you want.

Finally, ever cycle has been squeezed out of crafty to have it search faster; with the ever faster time controls
that get played it's still the case that only a number of seconds times your nps is the total number of nodes you're
allowed to search.

One can question whether those ever faster time controls can be seen as progress. I sure do not.

diep
Posts: 3
Joined: Thu Jul 28, 2011 4:11 pm
Real Name: Vincent Diepeveen

Re: Still waiting on Ed

Post by diep » Thu Jul 28, 2011 5:34 pm

Rebel wrote:This is going to be long I am afraid....

I downloaded the latest Crafty (version 23.4) and the search exactly behaves as the Ippo's, Robo's, Fire's etc. quite contrary to previous versions.

And you said it yourself, just yesterday in CCC, the relevant quote in question:
S. Taylor wrote: So the ippo or robbo (in official events, till now) would be rybka itself, wouldn't it?
Bob Hyatt wrote: Unproven, but suspected, so probably yes. Of course, there is the same idea for Fruit/Rybka.:)
So you took contaminated ideas you suspected they were the origin of the hacked Rybka and used them in Crafty.

And you know what?

It does not matter.

It's not the issue.

The hacker is hacked and his secrets are in the open for everybody to see. I can even see the irony of that. Use it by all means. The whole thing has been so ugly let's make the best of it. Programmers can profit. It's good for the progress of computer chess.

What I very much dislike is the moral side and by that I mean you. The verdict on Vas is right. And the penalties are harsh. While every other programmer involved is silent and show some empathy for the total abasement in public among his colleagues, his fans, his co-workers, his consumers and on top of that in the mainstream media. It's a punishment for life. That kind of shame. It's enough.

But not for you. You will not rest to trample Vas until you have squeezed the last drop of blood from his vessels while in the meantime using parts of his stolen legacy. I don't know what is more ugly.

After all this whole issue is about moral and ethics.
Ed i know how utter confusing all this is. Yet the similary between all those robo's and ippo's and rybka's is the same thing.

Let's first distinguish between near similar cloned engines. Eiterh total similar, or datastructure total similar and material eval similar. of the non-obvious clones: That's Rybka/Pandix/Thinker/Naum.

Having seen thousands of games of rybka i recognize these 4 engines as playing material in the same manner (never exchange anything except if it gives a big advantage and other oddities of rybka that all these 4 have).

Crafty is the exception here. Nothing of it has been cut'n pasted.

Yet what confuses you and which let those engines behave similar is the automatic way how they got tuned/tested.
In case of crafty this is also not so transparant (where is the code that allows you to modify the parameters in crafty huh? - only NSA probably has that code and as their code is military code and military code is secret it isn't getting published huh).

Yet the way how the testing principle of crafty works seems total different from the Rybka/Pandix/Thinker/Naum engines.

This testing to death principle causes however similar behaviour, as from fundamental viewpoint, the evaluation functions of all those engines are not so complicated. This in the long term causes similar tunings to coexist. Sure it took years extra for crafty.

Even more proof that crafty isn't doing the same there.

This where all the rybka/pandix/thinker/naum engines BOOM they were there at 3000+ elo.

rybka - the big engine
pandix - the quadcore version of it, so weaker than the big thing
thinker - if you download it lobotomized to 32MB hashtable
naum - 32 bits version of rybka.

The question you should ask yourself Ed, is how much money they threw into those projects to tune them. Of course once you have the entire tuning system in place, it's easy to quickly produce newer versions of rybka/pandix/thinker and naum at nearly the same time.

Yet where these 4 engines play in the same style and have the same oddities - crafty from my viewpoint as a chessplayer really plays different.

The thing crafty shares with them is that things slowly have become tested to death. Even a 128 core cluster running nonstop for some years doing that, seems to manage to tune crafty pretty well.

This where the edits in crafty reveal 3 programmers, which still is a lot less than the stockfish engine where if i remember, at least 4 programmers showed up with evidence that at least half a dozen of people are busy with it.

Also in those cases the code missing is the code that can modify parameters from the engine. Excuses are not relevant there.

If in open source projects specific very important code, i would argue THE MOST IMPORTANT CODE, NAMELY THE TESTING CODE,
is missing, you know the NSA is involved.

Regards,
Vincent

hyatt
Posts: 1242
Joined: Thu Jun 10, 2010 2:13 am
Real Name: Bob Hyatt (Robert M. Hyatt)
Location: University of Alabama at Birmingham
Contact:

Re: Still waiting on Ed

Post by hyatt » Thu Jul 28, 2011 11:28 pm

The code to tune parameters in crafty is the "eval" command. You can change most any eval term, even a specific value in a piece/sq table, you can tune reductions, futility margins, depth controls on futility/reductions, etc.

I simply have my test platform emit a command and tack it on the end of the .craftyrc file, to change the parameter of interest. And I run a match. And then change it and run again. I tell the test platform the range of values to use, and the increment, or the specific set of values to use, and then let it grind. When it is over, I pick the one with the best result, make that default, and move on...

No NSA/CIA/DOD/DOE people involved at all...

Post Reply