GPL discussion, sense and nonsense

[Rebel]

I am speaking of 1986 (Munich) and on.

1986 was cologne. i think bob was there.

ed and me were there too.

1993 was munich.

Cologne yes, my bad

But I believe only Harry Nelson of the Cray Blitz team was present.

Could I be wrong twice in one posting?

Ed

[Gerd Isenberg]

I am speaking of 1986 (Munich) and on.

1986 was cologne. i think bob was there.

ed and me were there too.

1993 was munich.

Cologne yes, my bad

But I believe only Harry Nelson of the Cray Blitz team was present.

Could I be wrong twice in one posting?

Ed

Yep, see cpw or your own photos from wccc 1986. Harry Nelson and Albert Gower. The last time Bob probably was in Europe was 1984 - the match against Levy at Advances in Computer Chess 4 in London.

[Gerd Isenberg]

How do you figure that? Don Beal wrote the first paper on null-move. Fruit was the first known instance of LMR (known as history pruning). So I am not exactly sure how you say they contributed the most. To wit:

iterated search? Slate - De Groot ?
Transposition table? Greenblatt
Exhaustive search actually works? Slate - and Adelson-Velsky et al.
Null-move? Beal - Adelson-Velsky et al.
LMR? Fruit was the first I knew of it, although you can find discussions with Bruce Moreland and myself (with a few others) back in 1997 that discussed the idea at length, but we never fooled around with which moves to reduce and which to not reduce.
Endgame databases? Thompson - Ströhlein ?
Large opening book? Thompson
Bitboards? Slate - Samuel in Checkers and Adelson-Velsky and Alexander Bitman of course
Singular extensions? Hsu/Campbell - and Thomas Anantharaman

Etc. I hardly think that "commercials contributed the most" is anywhere within a light-year of reality.

Of course on tournaments were a lot discussions between professionals, semi professionals, as well amateurs - about hoax as well as serious stuff or both. But true, except Donninger's null move paper in 1993 there were no publications from commercials that time (Marty Hirsch's paper on book learning was some years later).

Ideas were rediscovered many times independently, see alpha-beta. Nullmove was already used in Kaissa. Iterative deepening was probably based on Adriaan de Groot's progressive deepening from Thought and Choice in Chess, and also LMR was likely used and tried by many - whith more or less success. I am not aware Rudolf Huber came up with the idea, but of course might be very well possible as Ed told.

Ed, do you have any sources other than "Chinese whispers"?

Gerd

[Chris Whittington]

How do you know he did not discuss it with others at maybe an ICGA event?

This is just ridiculous.
He might have also discussed it with his wife, or brother or whoever, but that is completely irrelevant.
I would even go as far to claim that if Bob doesn't know for a certain "contribution", then real contribution does not even exist.
What does a chess community has out of private discussion of couple of ICGA participants (if that really happened and is not just another Harvey's manipulation) which nobody can actually confirm???
Nothing at all, nada, zero, zip, zilch is the answer.

In the "good old days" we always had a panel discussion at each ACM computer chess tournament, and at some (at least in North America) ICCA events as well. Those were specifically to discuss ideas and such. We often had paper sessions where various more in-depth presentations were made. And the ICCA Journal was _formed_ to provide a written forum for dissemination of computer chess papers. Add 'em all up (ICCA/ICGA papers) and then count the commercial authors. Donninger is the only commercial author that comes to mind. In the late 80's and thru the 90's we had r.g.c.c. Go look at old posts there and see what was discussing actual algorithms and ideas, and who just discussed general topics with no specific ideas of any kind revealed. For all I know, computer chess might have reached the same point we are at today, with nothing but commercial authors. But the key is "for all I know" because they were certainly not heavy contributors to the body of knowledge related to computer chess.

I strongly suspect computer chess would not exist as we know it if not for the early pioneers that did share ideas freely.

Let's re-phrase this shall we? In the pre-1980's when there were no available microprocessors the only people with access to computers were the various university nerds and socioopaths sharing basement space with the cockroaches and whirring discs nobody knew what to do with. They thought "Hey, wow, awesome, now there's meaning to my life of pizza, late nights with the cockroaches and no girls and maybe I can make some headway in the university departmental building game that has so far eluded me!" The goal was a commercial career at the university, nobody having thought what miniaturisation would bring, so publishing 'papers' and 'sharing' nonsense became all the rage. Then those bastard commercials with their microprocessors and low cost of entry came along, surprisingly enough occupying slightly more of the real world space where people and ideas compete for limited wealth and then anyone could do it (beard, body odour, stammering, never washing and total clamminess with the girls being the only requirement).

[hyatt]

Null-move is about +80 in Crafty. Measured carefully on the cluster. LMR is right at +100 today. By itself with no null-move. Again measured carefully on our cluster. One year ago it was +80, but some recent changes boosted this a bit. But even more importantly, if you already have null-move, LMR is only about +50 more. It is nowhere near +200 or beyond...

You speak of Crafty while I had Rybka, Stockfish and Ippo.* in mind. Apparently they seem to profit a lot more.

Not from null-move and LMR. I tested all of 'em, with and without. Although it is easy enough for anyone to run the tests. The two together are nowhere near +200 Elo...

How do you figure that? Don Beal wrote the first paper on null-move. Fruit was the first known instance of LMR (known as history pruning). So I am not exactly sure how you say they contributed the most.

There is no LMR in Fruit only "history reductions" based on counters. This idea originally came from Rudolf Huber (SOS) and was successfully implemented by SMK in Shredder. But the idea became known and it got the attention of everybody because of the Fruit sources.

I think that Shredder was the first program to use LMR. It's why Shredder ruled the world for a couple of years. Then (again) the idea of LMR became known (Stephen likes to talk apparently) and Tord eventually gave it the public attention.

References below. Talking commercial programmers

Ed

By Anthony C Date 2008-02-04 14:47 The true inventor of LMR was Stefan Meyer-Kahlen, of course, which is why S7-S9 completely dominated the computer chess world. But Fabien gets credit because he published his work. Stefan didn't, and random message board discussions certainly do not count.

...

cheers,

anthony

and:

By Vasik Rajlich Date 2008-02-08 09:35 Actually, I talked to Stefan about it and yes, he was using LMR in almost exactly the modern way already with Shredder 7. This means re-searches, exceptions for late-move captures and other special moves, etc.

In my view, Tord clearly deserves the main credit for LMR. Fabien probably deserves the #2 spot, since Fruit is what really showed beyond any doubt that this works.

Vas

[Rebel]

Null-move is about +80 in Crafty. Measured carefully on the cluster. LMR is right at +100 today. By itself with no null-move. Again measured carefully on our cluster. One year ago it was +80, but some recent changes boosted this a bit. But even more importantly, if you already have null-move, LMR is only about +50 more. It is nowhere near +200 or beyond...

You speak of Crafty while I had Rybka, Stockfish and Ippo.* in mind. Apparently they seem to profit a lot more.

Not from null-move and LMR. I tested all of 'em, with and without. Although it is easy enough for anyone to run the tests. The two together are nowhere near +200 Elo...

And yet you again are talking only about the performance of LMR & null-move in Crafty. In mine the gain is even less, maximum 20-30 elo each. Just look at the sophisticated LMR implementation of Stockfish & Ippo and add a 50-100-150 elo points to Crafty. Where else do you think Stockfish's big elo jumps are coming from other than sophisticated and new LMR idea's?

BTW, a little bird told me Crafty recently made big elo-jumps as well, 150-200 elo. Is that bird right? Can you elaborate in which area's you made this kind of progress?

Ed

[Rebel]

Of course on tournaments were a lot discussions between professionals, semi professionals, as well amateurs - about hoax as well as serious stuff or both. But true, except Donninger's null move paper in 1993 there were no publications from commercials that time (Marty Hirsch's paper on book learning was some years later).

Hi Gerd,

I think the absence of papers by commercials is merely related to the absence of being asked. I never been asked, if they had I would have done. I have been asked by "Computerschaak" and I have contributed. You know the guys yourself, they are shy in the positive sense of the word, they don't put themselves in the forefront asking to publish, they liked to be asked.

Ideas were rediscovered many times independently, see alpha-beta. Nullmove was already used in Kaissa. Iterative deepening was probably based on Adriaan de Groot's progressive deepening from Thought and Choice in Chess, and also LMR was likely used and tried by many - whith more or less success. I am not aware Rudolf Huber came up with the idea, but of course might be very well possible as Ed told.

Ed, do you have any sources other than "Chinese whispers"?

Gerd

Unfortunately I have forgotten the source but not the content. Note that Huber did not invented LMR but "history reductions" based on history counters. But then from one thing came another, programmers (to begin with SMK) started to smell the real thing, double / triple reductions if a move is late and in CUT mode. But in a way you could call Huber's history reductions the forerunner of LMR.

Ed

[hyatt]

Null-move is about +80 in Crafty. Measured carefully on the cluster. LMR is right at +100 today. By itself with no null-move. Again measured carefully on our cluster. One year ago it was +80, but some recent changes boosted this a bit. But even more importantly, if you already have null-move, LMR is only about +50 more. It is nowhere near +200 or beyond...

You speak of Crafty while I had Rybka, Stockfish and Ippo.* in mind. Apparently they seem to profit a lot more.

Not from null-move and LMR. I tested all of 'em, with and without. Although it is easy enough for anyone to run the tests. The two together are nowhere near +200 Elo...

How do you figure that? Don Beal wrote the first paper on null-move. Fruit was the first known instance of LMR (known as history pruning). So I am not exactly sure how you say they contributed the most.

There is no LMR in Fruit only "history reductions" based on counters. This idea originally came from Rudolf Huber (SOS) and was successfully implemented by SMK in Shredder. But the idea became known and it got the attention of everybody because of the Fruit sources.

I think that Shredder was the first program to use LMR. It's why Shredder ruled the world for a couple of years. Then (again) the idea of LMR became known (Stephen likes to talk apparently) and Tord eventually gave it the public attention.

References below. Talking commercial programmers

Ed

Let me say this again. The idea is to _share_ information. And the most usual way to share information is thru written explanations, such as the _many_ published in the ICGA (back then ICCA) Journal. Please cite SMK's paper that explains LMR.

Fruit clearly was "the beginning". Within a year of Fruit, I had already discounted the history approach and went to a more direct idea. Tord certainly coined the term "LMR". Who gets credit for it may not be answered. But I don't count commercial programmers that use it, keep it secret, and then when it becomes generally known, says "I was doing this 3 years ago."

By Anthony C Date 2008-02-04 14:47 The true inventor of LMR was Stefan Meyer-Kahlen, of course, which is why S7-S9 completely dominated the computer chess world. But Fabien gets credit because he published his work. Stefan didn't, and random message board discussions certainly do not count.

...

cheers,

anthony

Which is my point, of course...

and:

By Vasik Rajlich Date 2008-02-08 09:35 Actually, I talked to Stefan about it and yes, he was using LMR in almost exactly the modern way already with Shredder 7. This means re-searches, exceptions for late-move captures and other special moves, etc.

In my view, Tord clearly deserves the main credit for LMR. Fabien probably deserves the #2 spot, since Fruit is what really showed beyond any doubt that this works.

Vas

Again, using it is one thing. Telling everyone about it is something else. And SMK certainly did not do that. So I fail to see what your point is, in the above..

[hyatt]

When did Frans Morsch supposedly give us recursive null-move? It was certainly in Cray Blitz in 1989 as the version on Carey's "computer chess history" web site will show. The first reference I am aware of is "Selective search without tears" by Don Beal which laid this out nicely.

This gets interesting. You never claimed this until now. You only have said (and multiple times BTW) you and Bruce have tossed with the idea of "null-move" but eventually could not find an improvement and dropped the idea. That is no surprise because the strength of null-move comes from its recursive use. And that was new. The inventor: Frans Morsch. Frans talked to Donninger who wrote an article in the ICCA journal (1994/95). Then in 1996 the topic exploded in RGCC, recursive-null-move was heavily discussed and finally implemented in every chess program. Talking commercials.

I assume you made a typo claiming recursive-null-move as your own, you probably meant null-move.

Ed

As "my own?" I thought I made it quite Clear that my implementation came after Don Beal published "Selective search without tears." My very first implementation allowed just one null move. We went to recursive at some point, but I had thought it was around 1992 until we found the 1989 version of the Cray Blitz source that Carey has on his web site. When I looked at the code, I thought "I don't remember doing this" but when I looked at the change-log that was carried along with the source code, there it was. It was most definitely R=1 and not 2 or adaptive at that point in time. But it was certainly there, and it also was certainly not my idea.

Burt Wendroff sent me a copy of Beal's paper, and I had null-move in and working in Cray Blitz an hour later. I just checked to confirm the date and this was 1986 and not 1989. So I was actually off by 3 years as to when we first used null-move. And beal used it before that.

So, apparently, the time-line (since I don't have anything but some paper logs of CB output, and then this one source version with a changelog that ended around the end of 1989 was found and sent to me along with the now public version of the source that does run on a PC) was:

1986. null-move, R=1, one null-move in any path.

somewhere between 1986 and 1989, null-move R=1, only restriction was two consecutive null-moves were not allowed.

And then somewhere in 1995-1996 I worked with varying R from 2 to 1, after discussions on R.G.C.C with John Stanback (he was playing with this as well).

About 1.5 years ago I went to pure R=2 everywhere.

And I don't claim originality in any of the above. Don "opened pandora's box".

[hyatt]

Quite simply, in that I know a number of competitors at the time that went to those events. Bruce Moreland was one, and we had talked about a "reduction scheme" during the 1996-1997 (both) ICCA WMCCC events. Bruce talked quite a bit with SMK and he didn't reveal a thing about "reductions" to Bruce, as one example. And in the above window, both Bruce and I were experimenting with this idea as previously mentioned, we just never reached a "happy point".

Your bias is unbelievable

Have you (yourself) ever spoken with a commercial?

How about Lang? The Spracklens? Kittinger? Do they count?

I am asking because I never met you on any tournament. You just were never present. I am speaking of 1986 (Munich) and on.

You must mean 1986 in Cologne? Harry went. I never missed an ACM event however, but never saw you at one although Louman ran your program a couple of times. I was at the 1989 WCCC. I was at the 1977 WCCC. 1983 where we won. I've been around. Just not at the old WMCCC events when we were still having the WCCC every 3 years.

Programmers talk on tournaments, the commercial ones included. From several commercial programmers I got: Alpha/Beta, killer heuristic, iterative search, aspiration search, Q-search. I returned as well. They are not much different than you, if you have a passion it's hard to keep your mouth shut especially when you are on an event with your kindred spirits. Just enter a topic and words starts to roll. You should have tried but were not there, so how can you judge so mean?

Ed

You do realize that alpha/beta is 50+ years old and is explained in every AI book on the planet? You might have heard discussions about "killer moves", iterative search, etc. But I can tell you where they came from... "Chess Skill in Man and Machine" in the chapter written by Slate.

Again, talking about something "after the cat is out of the bag" is not particularly interesting. Hsu explaining singular extensions when nobody had heard of them is more interesting. Or Beal describing his null-move experiments. Or Donninger with his null-move threat detection idea. Or greenblatt with trans/ref table. Or Thompson with endgame tables. Or Newborn, Schaeffer and myself on parallel search approaches. I am talking about "revealing new ideas" not "discussing old ones."

Those you might have a tad more trouble identifying as coming from commercial programmers.