Not logged inRybka Chess Community Forum
Up Topic Rybka Support & Discussion / Rybka Discussion / A question to Larry Kaufman
1 2 3 4 5 Previous Next  
- - By er (**) Date 2009-12-30 07:07
In 2004 Larry Kaufman's great opening book came out:
'The Chess Advantage in Black and White', published by Random House, McKay Chess Library.
It's a clever textbook, covering a complete repertoire for both sides.

How about an updated edition?

Erik
Parent - - By oudheusa (*****) [nl] Date 2009-12-30 15:33
i think Larry is no longer active on this forum.
Parent - - By Felix Kling (Gold) [de] Date 2009-12-30 16:20
maybe he's reading for time to time. You can never be sure :)
Parent - By herakleitos (***) Date 2009-12-31 02:40
He's answered some posts in the past several months. You never know when he might surface.
Parent - - By lkaufman (*****) Date 2009-12-31 03:38
I usually look at the forum once a week or so. Currently I'm working with Don Dailey on the chess program "Doch", rated 7th by both CCRL and CEGT for single core programs (counting only one program per author), and moving up very rapidly. We hope to make the top three soon, unless others besides Rybka and Stockfish make big leaps.
Parent - - By Banned for Life (Gold) Date 2009-12-31 05:04
What effects do you see from the release of open source code of dubious pedigree? The first time this kind of thing happened, there was a rapid rise in the Elo of many other engines over the next year, but only Anthony C. admitted that he had studied the code and even incorporated some concepts in ZM II (e.g. aggressive futility pruning). I would be surprised if this didn't happen again, although I suspect that very few engine developers will acknowledge getting ideas from this source.

In any event, it will be interesting to see if you can substantially raise the bar on the evaluation front...
Parent - - By lkaufman (*****) Date 2009-12-31 05:58
At least one idea discussed on other forums that was attributed to Rybka 3 via the clone is already implemented in Stockfish 1.6, and was apparently the main reason for its large jump in strength over 1.5. So at least this idea will surely be in all top programs soon. Some developers will study the clone code themselves, while others may only use ideas they read about that came from the clone, but sooner or later other programs will show large gains from this information. At the very least, once an idea is implemented in a legitimate open-source program like Stockfish, it becomes accepted as something every programmer may use. As for evaluation, actually right now I'm helping Don as much with the search as with eval. I knew very little about search while working on R3, but that's no longer so.
Parent - - By Highendman (****) Date 2009-12-31 06:39
Dear GM Kaufman, could you kindly share here what that idea is, in a way a layman could understand that too? It's quite intriguing to understand what 'ideas' in engine development are, especially if you believe this is a new idea that has a significant strength impact. Thanks in advance, if you agree to take the time.
Parent - - By Highendman (****) Date 2009-12-31 11:58
Looking at talkchess, I guess you mean 'smooth scaling' which some people there claim is going to add about 150 ELO on top of the current released Stockfish version.
Parent - - By Aliven (***) [us] Date 2009-12-31 12:06
I find the format of that forum absolutely horrid. Why can't they use a standard forum format?
Parent - - By Vempele (Silver) [fi] Date 2009-12-31 12:11
Wait, what? It's phpBB, which is about as bog-standard as it gets.
Parent - - By Aliven (***) [us] Date 2009-12-31 12:28
Okay... Why can't it look more like this forum so I don't have to click all the "RE:blahblah".
Parent - - By Eelco de Groot (***) Date 2009-12-31 14:39
Hi Aliven,

In the upper right hand corner of your screen there is a button "flat view" which you should set to "thread view". If you register you can set this in your profile, but for guests the default I think is still "flat view". It would probably be better to change that.

Eelco
Parent - - By Vempele (Silver) [fi] Date 2009-12-31 19:20
You seem to be confused. The view type is the one that the button isn't. Flat view is the one with multiple posts per page (and threaded view is the default).
Parent - - By Eelco de Groot (***) Date 2009-12-31 19:58
Yes, you are right of course Vempele. I don't use threaded view much, the indentation on Talkchess is too high so it quickly becomes impossible to see who responded to whom, if the quotes don't make it clear. Here in the Rybka forum I set the indentation to 1 as the minimum, that works better. Also here you can go directly to the parent post if it is not clear which post is being replied to. In the old CCC there was only threaded view, but it was more clear which posts were new since the last time you visited. But "flat" view is alright for a better overview, it does seem to lead to more oneliners though, but a more conversational style.

Eelco
Parent - By Aliven (***) [us] Date 2010-01-01 00:25
Ty for that information, I seriously had no idea, lol. This should help for sure.
Parent - - By Uri Blass (*****) [il] Date 2009-12-31 12:24
I do not believe that smooth scaling is going to add 150 elo to stockfish.
This was the first estimate of Dann Corbit but it is clearly wrong.

Rybka3 is still significantly stronger than stockfish based on all the results that I read.

Uri
Parent - - By Akbarfan (***) Date 2009-12-31 13:19
Yes, I also think so, but i guess larry means the step from 1.5.1 to 1.6
Parent - - By Uri Blass (*****) [il] Date 2009-12-31 14:07
I responded to highendman and not to Larry.
Parent - By Akbarfan (***) Date 2009-12-31 15:23 Edited 2009-12-31 15:35
Yes I know, and HEM ask larry >I guess you mean 'smooth scaling'...< which is not right.
Larry means - according to other forums - the step from 1.5.1 to 1.6 was partly from DELETED.
smooth scaling was made by Dan Corbit.
Whatever,
HAPPY NEW YEAR

Edit:
Oops, Larry has responded to HEM, in the meantime I was on the phone.
Parent - By lkaufman (*****) Date 2009-12-31 15:00
No, I haven't confirmed any gain from that, and if there is a gain it will be tiny.
Parent - - By Androcles (**) [fr] Date 2010-01-01 17:40
Do you mean yet another 150 ELO jump from the current released (ie 1.6.0 ja) version?
Or do you mean 150 ELO from the superceded 1.5.1 version?
Parent - - By turbojuice1122 (Gold) [us] Date 2010-01-01 17:54
The 150 elo jump was implied to be referring to version 1.6, i.e. 150 elo stronger than Stockfish 1.6.  This has since been found to be an overestimate, though it is still possible that this version is stronger than Rybka 3.
Parent - By ernest (****) [fr] Date 2010-01-03 02:53

> though it is still possible that this version is stronger than Rybka 3.


...you mean it is still possible that this version 1.6s is not at all stronger than version 1.6 (what a joke!!!) :lol:
Parent - By lkaufman (*****) Date 2009-12-31 15:06
I don't think this is the place to publicize ideas from Rybka that were supposed to be secret, but the Stockfish 1.6 notes make it obvious that the big change in this version was a new way of doing "singular extension", improved from the way it was done in Deep Thought/Deep Blue.
Parent - By Androcles (**) [fr] Date 2010-01-01 17:43
@Higendman
Do you mean yet another 150 ELO jump from the current released Stockfish (v 1.6.0ja) or
150 ELO from the superceded version (1.5.1)?
Parent - - By Banned for Life (Gold) Date 2009-12-31 16:05
It's interesting to see how sausage is made here. Nobody really wants to fess up and admit that ideas were obtained by reverse engineering, so they have to be laundered first, preferably through an open source program from respectable authors. This is in contrast to many other fields, where having your product reverse engineered is the highest form of flattery...
Parent - - By lkaufman (*****) Date 2009-12-31 16:16
I'm not sure I follow you here. Clearly the reverse engineering of Rybka is indeed "the highest form of flattery"; as to why the author(s) of the clone don't openly admit it I can't say. But the specific idea I'm talking about here is attributed to the clone (and hence by implication to Rykba) in the forum where it was discussed, so I don't think Stockfish or any other program which uses this idea is denying its origin. Once a good idea becomes common knowledge, regardless of its origin, programmers must use it (if it helps) to remain competitive.
Parent - - By Banned for Life (Gold) Date 2009-12-31 16:31
I'm quite sure that the ideas that have been gleaned will quickly show up in other engines. I am not nearly as sure that there will be much attribution though (kudos to you and Don here, but I haven't seen the same from the authors of Stockfish). One can only hope that in the future, when a book on the history of ideas in computer chess is written, these algorithms will be properly credited.
Parent - By lkaufman (*****) Date 2009-12-31 17:07
To be fair to Stockfish, the authors may not be certain whether the idea in question came from Rybka or from the clone, though I don't have any doubt on that point myself.
Parent - - By zwegner (***) Date 2009-12-31 17:14
I assume you are talking about singular extensions? It's hardly a new idea, or even a new implementation (in the RE code). Stockfish didn't have it before, true. But as GCP pointed out, most commercial authors will have already played extensively with the idea. I know of several that have.
Parent - - By lkaufman (*****) Date 2009-12-31 17:23
I thought that the way it was implemented in Stockfish and discussed on forums was supposed to be a new way to do singular extensions attributed to the clone. If others knew about this idea before the clone that is news to Don and I and to whoever credited the clone in the forums.
Parent - - By zwegner (***) Date 2009-12-31 18:22
It's not really that new. The idea of only doing with only doing SE if you have a hash move is perhaps not well known, but I've heard of similar restrictions (having a good move, hash or good capture, etc), though I don't do that. The other thing that's "new" is the weird (score - depth / 2) bound that is used. I assume that all of these values are well tuned in Rybka, but that doesn't make much sense to me, and there are many other (already existing) ideas that should work just as well.
Parent - - By lkaufman (*****) Date 2009-12-31 18:41
I'm referring specifically to the hash move limitation; perhaps other implementations included too many  other categories of moves and so failed to prove to be of much benefit.
Parent - By BB (****) [au] Date 2010-01-17 18:27
Since Gerd Isenberg seems have to conflated SF/RL ideas in a CCC post regarding this point (at least I am confused by his terminology, as applies to RL), I perhaps should add something (at the risk of it being buried in this thread):

> GI: The kind of singular extensions restricted to lower bound hash hits is a much stronger symptom, as recently stated by Larry Kaufman in the Rybka forum, which might be simply proved by looking with Ida-pro to the Rybka binary, if one don't trusts Larry for whatever reasons.


I'm not sure what "lower bound hash hits" means here. That seems to be the SF method (which Marco Costalba noted was different). Here are the R3/RL pseudo-codes for singular extensions at CUT nodes when not in check:

Rybka 3 Chess [CUT/ALL/Exclude node for white is 0x487ac0 in Linux gdb]
  if (depth>=18 && trans_move && move_is_ok(trans_move)) // 0x48864a-0488682
  // also checks that this is indeed a CUT node, and that no move is already excluded
  {compare_value=Exclude(scout-30,depth-10,trans_move,...); // 0x4886d5
   if (compare_value < scout-30)
   {SINGULAR=2; if flags were already fiddled, increase SINGULAR some more, possibly, depending upon the height perhaps;}}


IP/RL/IG/IH/FB
  if (depth>=16 && trans_move && move_is_ok(trans_move))
  {compare_value=Exclude(scout-depth,depth-MIN[12,depth/2],trans_move);
   if (compare_value < scout-depth)
   {SINGULAR=1; if (height*4<=depth) SINGULAR++;
    compare_value=Exclude(scout-2*depth,depth-MIN[12,depth/2],trans_move);
    if (compare_value < scout-2*depth)
    {SINGULAR++; if (height*8<=depth) SINGULAR++;}}}


I think the claim is that both demand trans_move to be non-null? Considering that there is going to be an "Exclusion search" against the trans_move, it seems almost necessary for it to exist?! When I spoke with GCP, he seemed to suggest that the whole "exclusion scheme" was new to him, but I couldn't make out what he gave as an alternative ("search-within-search" rather than "exclusion-search"?).

Other than that, I don't see all that much the same here. RL uses a "2-step" process to find singular moves with a variable window, while R3 does it all at once, with a fixed one. Both have "supersingular" extensions (that is, beyond a 1-ply extension -- I redacted these in the above R3 code, and I don't understand them exactly), though in RL they are purely height-based, while in R3 they seem to involve, for instance, series of singular moves.

The picture is similar at CUT nodes when in check. At PV nodes, there are more differences, as R3 puts "extensions" (passed pawn pushes, moves that give check, piece swaps) in the singularity code, while RL does it later.
Parent - By BB (****) [au] Date 2010-01-05 02:57

>LK: I'm referring specifically to the hash move limitation


This does seem to be in R3. The depth bound (16 half-ply) is also the same I think.

>ZW: The other thing that's "new" is the weird (score - depth / 2) bound that is used. I assume that all of these values are well tuned in Rybka,


I fail to find this "weird" (score - depth / 2) bound in R3. And I think R3 uses a different depth reduction when doing the comparison.
Parent - - By Geomusic (*****) Date 2010-01-01 12:36
Soon most competitive programs will have a bit of rybka in it :) This is indeed flattery!
Parent - By lkaufman (*****) Date 2010-01-01 15:47
Probably they already do, due to "Strelka" and other leaks from earlier Rybka versions.
Parent - - By BB (****) [au] Date 2010-01-05 04:22

>Nobody really wants to fess up and admit that ideas were obtained by reverse engineering,


To quote a wiki:
Fact: fine ideas cultivate for many gardens ...ideas splurged for miles in ovals
And:
Code with ****** buys own same tunings, accumulating ideas from all turnip patches. For comparisons, Russia Champion Karpov achieved thoughts from yoghurts, so and challenger shed brainwaves upon parapsychographical apparitions. Effulgent ideas sprout in totality of garden shrubs, so too exposed in capitalist ships!

But on the other hand:
Decompilation burns the intrepid puddles with the lutefisk. Impressment has not the bones. Conclusivity: Wasting of Comrades mobilities.

There you have it - is that an admission? :-P
Parent - - By Banned for Life (Gold) Date 2010-01-06 04:51
LOL, No translator could generate such terrible gibberish. This person should pursue a career as a speechwriter!

But actually, I was referring to acceptable engines not wanting to admit that they were using ideas from R3, rather than Nelson's comrades.
Parent - - By BB (****) [au] Date 2010-01-06 05:57

>This person should pursue a career as a speechwriter!


Well, they/it claims to be 2 different people between those 3 quotations...

>But actually, I was referring to acceptable engines not wanting to admit that they were using ideas from R3, rather than Nelson's comrades.


You realise that he is now a "Suspicioned Capitalist", due to "starring in a Capitalist Propaganda video" (or something like that)? :-P
Parent - - By Banned for Life (Gold) Date 2010-01-07 01:15
They have just discovered that Comrade NelsonHernandez is a libertarian with a secret admiration for Trotsky. This should be enough to confuse anybody...
Parent - - By Nelson Hernandez (Gold) [us] Date 2010-01-07 04:02
Vyshinsky is another of my favorites.  What a piece of work.
Parent - - By Dr.X (Gold) Date 2010-01-07 04:40
Now they have some joker posting as a representation of the Vietnam Communist party! It would have been more impressive if he had posted as a rep from the The Dragon Capital!
Parent - - By Graham Banks (*****) [nz] Date 2010-01-07 04:55

> Now they have some joker posting as a representation of the Vietnam Communist party! It would have been more impressive if he had posted as a rep from the The Dragon Capital!


How anybody could possibly think that this whole thing is anything else but a sick and twisted joke, is beyond me.
Parent - - By Dr.X (Gold) Date 2010-01-07 05:09
The proletariat's take this very seriously! Oh! The members take this very seriously- I've check out their  main chess forum- they have a lot of members, some from here!
Parent - - By Graham Banks (*****) [nz] Date 2010-01-07 05:11

> The proletariat's take this very seriously! Oh! The members take this very seriously- I've check out their  main chess forum- they have a lot of members, some from here!


Suckers - LOL
Parent - - By Dr.X (Gold) Date 2010-01-07 05:12 Edited 2010-01-07 05:20
Oh! You'd be surprised! I was! With that said, I think I'd be real careful about labeling anyone- People have a right to decide who they want to support and that is something that they should be able to do openly and honestly without fear of recriminations taking place. However, if you try and play both ends against the middle that's when you end up in trouble. That is just plain double dealing.
Parent - By Graham Banks (*****) [nz] Date 2010-01-07 05:56

> I think I'd be real careful about labeling anyone- People have a right to decide who they want to support and that is something that they should be able to do openly and honestly without fear of recriminations taking place.


That's true.
Parent - - By AWRIST (****) Date 2010-01-01 10:24 Edited 2010-01-01 11:24
I take this message as base for my short comment, but I could have taken almost every other from you today -

you are very good at chess for certain, but how deep are you in programming?

There is a point in your presentation that makes me suffer in my ethical mind frame. Something that stinks, sorry if that might insult you, but I have seen something that goes beyond your person.

I fear we must re-define the role of what is a crime in software copyright (for fields like computerchess where everything is about ideas and how to code it in a most effective manner).

For me it's a crime if someone (I'm not talking about LK, how could I without informations) robs out of a code that is NOT for free but closed for commercial usage, puts it into his robbed rob, then publishs the robbed code for free and some Stockf or whatever have it in their code too. Then Stockf isnt legal (what you are proclaiming here) in my eyes but the same robbed rob in Stock veils. It's then also a crime, just IMO, if a Doch appears with your help and suddenly has all the robbed stuff and soon is among the first three in SP, then you give this for free (just quoting what you wrote) again, and then announce that you want to make money with MP Doch. Do you at least realise, sir, that you are standing on robbed rob code from Vas?

I fear that the whole concept of open source if you take MS and then the inofficial, free OpenOffice is unimportant. But it's still a crime IMO, but who cares if official institutions all use MS? But in a small area in computerchess it cannot be legal, it IS a crime, if after every new invention in Rybka, over the robbed implementation into open source like Crafty and others, suddenly over Doch someone wants to establish a competition for Rybka, but in truth it's basically all robbed from Rybka itself. That's unbelievable chuzpah method of crime, no?!

IMO the open source stuff over Crafty as a source of inspiration is working as long as known competitors in business then take everything after a _polite_ period of respect, but if suddenly someone appears without any basis, robs the whole open source and turns it in competition against the friendly donator Rybka, then this is a crime, no?

IMO the open source like Crafty is no longer tolerable if it seemingly legalizes a robbed rob code from Rybka in the last instance. We must investigate if such a pretended  engagement out of universities without money interests isnt basically supporting the former robbery of stolen code. That should imediately be investigated by the State authorities in the USA. People always speak of the "crazy" Russians, but in truth absolutely innocent chaines of naive distribution of code is the main threat in computerchess for any possible business at all. Because we speak here about poor Rybka as victim, but what about the misery for the old competitors like Junior or Fritz in the business and _their_ living?

IMO we must reform our understanding of what is legal and what is crime in this field of software production. It cannot be true that robbery is publicly applauded and plans are announced how to make money on that basis.
Up Topic Rybka Support & Discussion / Rybka Discussion / A question to Larry Kaufman
1 2 3 4 5 Previous Next  

Powered by mwForum 2.27.4 © 1999-2012 Markus Wichitill