TEST position TCEC5- Houdini 1.03a-DRybka4 1-0

Discussion of anything and everything relating to chess playing software and machines.

Moderators: hgm, Harvey Williamson, bob

Forum rules
This textbox is used to restore diagrams posted with the [d] tag before the upgrade.
tomgdrums
Posts: 736
Joined: Wed Dec 02, 2009 8:48 am

Re: TEST position TCEC5- Houdini 1.03a-DRybka4 1-0

Post by tomgdrums » Tue Nov 30, 2010 6:46 pm

Milos wrote:
Graham Banks wrote:Actually Fruit 2.1 was 2796 and Rybka 1.0 was 2888. Just a slight difference from your false propaganda. :lol:
According to your "very reliable" list which is like +/- 50 elo accurate and from 2010, plus it has Fritz rating adjustment, sure :lol:.
Talking about false propaganda check in dictionary under CCRL, they are synonyms you know ;).

I love when you come to the forum, it is like having Glenn Beck and Rush Limbaugh talking about chess engines all at once!

User avatar
Graham Banks
Posts: 31909
Joined: Sun Feb 26, 2006 9:52 am
Location: Auckland, NZ

Re: TEST position TCEC5- Houdini 1.03a-DRybka4 1-0

Post by Graham Banks » Tue Nov 30, 2010 6:48 pm

tomgdrums wrote:
Milos wrote:
Graham Banks wrote:Actually Fruit 2.1 was 2796 and Rybka 1.0 was 2888. Just a slight difference from your false propaganda. :lol:
According to your "very reliable" list which is like +/- 50 elo accurate and from 2010, plus it has Fritz rating adjustment, sure :lol:.
Talking about false propaganda check in dictionary under CCRL, they are synonyms you know ;).

I love when you come to the forum, it is like having Glenn Beck and Rush Limbaugh talking about chess engines all at once!
Or alternatively like Sarah Palin talking about her world knowledge! :wink:
My email addresses:
gbanksnz at gmail.com
gbanksnz at yahoo.co.nz

Milos
Posts: 3270
Joined: Wed Nov 25, 2009 12:47 am

Re: TEST position TCEC5- Houdini 1.03a-DRybka4 1-0

Post by Milos » Tue Nov 30, 2010 7:07 pm

Don wrote:I am sure that Vas learned a lot from looking at the Fruit code as he would be stupid not to take advantage of what is freely available. And I'm sure he developed as a chess programmer, learning things as he went along.
It's just hilarious. I'm also sure that many ppl these days learned a lot by looking at Ippo sources and also developed as chess programmers, learning things as they went long. And therefore, many new programs came out some of them stronger than Ippo, some stronger even than Rybka 4.
So do you really believe that???
Or you just believe it in case of Vas but not others. If that's the case, than it's pretty obvious that you are impartial.

He did a lot to improve his program from Rybka 1 beta to Rybka 3 (with some side help, Vincent can tell you more about it :)), but that had nothing to do with all the other programmers. Vas kept you guys so many years in the dark, and probably therefore you think of him as some chess god.
He started in the similar way as some of the new guys (you popularly call cloners), but the only thing he brought to chess community is his full pockets. If he shared knowledge or didn't go commercial he would certainly have more respect. Like this he's nothing but a parasite. Quite contrary to Ippo guys.
Being unable to see this is really a sad thing.

luisrodg
Posts: 30
Joined: Thu Oct 07, 2010 2:15 pm

Re: TEST position TCEC5- Houdini 1.03a-DRybka4 1-0

Post by luisrodg » Tue Nov 30, 2010 7:17 pm

@Milos: "with some side help, Vincent can tell you more about it"

Please tell. I'd like to know about it as I don't.

Thanks.
B.S. in Mathematics
M.S. in Mathematics
Ph.D student in Mathematics.

Milos
Posts: 3270
Joined: Wed Nov 25, 2009 12:47 am

Re: TEST position TCEC5- Houdini 1.03a-DRybka4 1-0

Post by Milos » Tue Nov 30, 2010 7:32 pm

luisrodg wrote:@Milos: "with some side help, Vincent can tell you more about it"

Please tell. I'd like to know about it as I don't.
It's the tuning thing. As you probably know Vas doesn't do any tuning himself. There was a massive automated tuning effort for R3. And I'm not talking about LK since his coefficient are nowhere near what's been implemented in R3. And as you probably know proper tuning means a lot, hundreds of elo. For example in today's standard Fruit coefficients are ridiculous. Just think of how much elo you could get from it if properly tuned (and I'm not talking about relatively amateurish efforts with Toga).
If you look at R4 it's so un-tuned, they couldn't even get TM correctly. Obviously auto-tuning resources are not available any more...
Who and why provided them in the first place, that you would have to ask Vincent (or Vas).

Tom Barrister
Posts: 227
Joined: Tue Oct 05, 2010 3:29 pm

Re: TEST position TCEC5- Houdini 1.03a-DRybka4 1-0

Post by Tom Barrister » Tue Nov 30, 2010 7:46 pm

Since I don't have saved copies of these lists, and I don't know if histories of said exist for perusal anywhere else, I have to use archive.org. Using the standard 40/40 list from CCRL, starting May of 2007, or about the time that Rybka took its place at the top of the chess program world, and taking selected engines, e.g. the best of each variety that were around both then and now:

Code: Select all

5-21-2007

Rybka 2.3.1     3078
Zap Zanzibar    3021
Hiarcs 11.1     2993
D Shred 10      2945
Deep Fritz 10   2943
Naum 2.1        2937
Toga II 1.3     2875
Glarung 1.2.1   2859
Spike 1.2       2827
Bright 0.1d     2739
Crafty 20.14    2713
Now fast forward to today

Code: Select all

11-26-2010

Rybka 4         3264
Stockfish 1.9.1 3222
Naum 4.2        3184
D Shred 12      3127
Deep Fritz 11   3096
Zappa Mex II    3075
Hiarcs 12.1     3010
Toga II         3008
Bright 0.4      3002
Crafty 23.3     2946
Spike 1.2       2849
This is the improvement over the past 3 years and 6 months for each:

Code: Select all

D Rybka        186
Stkfsh/Glrng   363
Naum           247
Shredder       182
Fritz          153
Zappa           54
Hiarcs          17
Toga II        133 
Bright         263
Crafty         233
Spike           22
Using good-old Spike 1.2 Turin (which hasn't been developed since 2005) as the benchmark, we can conclude (right or wrong) that there hasn't been much, if any, artificial inflation in the ratings in the 3.5 years. Whatever ELO improvement there is has been made due to improvements in the technology, design, etc.

The question I have is this:

HOW can anybody in their right mind say that the last 200/300/whatever ELO has been the exclusive property of Mr. Rajlich and Rybka 3->4? .

Since Rybka is closed-source (or whatever the correct term is), they certainly couldn't all have decompiled his program and used his ideas, such as they are. They must have had their own ideas or used/modified freely-available code/ideas of others.

As for the ELO increases, Stockfish (which is a fork of Glarung) went up almost twice as much. Naum, Bright, and Crafy went up about 1.3 times as much, Shredder about the same, Fritz a bit less, Toga a bit less than that. Only Zappa and Hiarcs didn't keep pace, and of course, we can't expect an engine (Spike) that is exactly the same at both points in time to improve, especially since it was included as a benchmark against artificial inflation.

Apparently, a lot of the leading legal engines improved as much or more than Rybka has by mainly fair means. I'm sure that most programmers looked at the various open-source code of others, plus whatever they could glean from conversations, forum posts, emails, etc., took advantage of whatever new technologies were available, and improved their engines by these means. Since Mr. Rajlich doesn't live in a vacuum (as there IS some air, albeit rarified, up there on that high-cloud perch), he probably did that same.

What am I missing here? Why is Mr. Rajlich the way, truth, and the light, and all others pale before him?

PS: I wouldn't doubt it if Stockfish overtook Rybka before the next version of the latter is produced, supposedly in the fall of 2011. I also wouldn't be surprised, if Stockfish does whiz by Rybka , that Rybka updates/versions start cranking out a lot faster than they used to. Without the 100 ELO cushion----indeed if not even number one, and with no way to pressure the ratings lists people to exclude Stockfish, Mr. Rajlich may feel more compelled to improve his product (faster), so that he can possibly stay on top of the engine list, and thus give the public a reason to buy his product.

Any odds on Mr. Rajlich incorporating some of the ideas from Stockfish into Rybka 5?
This production is being brought to you by Rybka: "The engine made from scratch.™"

User avatar
Don
Posts: 5106
Joined: Tue Apr 29, 2008 2:27 pm

Re: TEST position TCEC5- Houdini 1.03a-DRybka4 1-0

Post by Don » Tue Nov 30, 2010 7:54 pm

Milos wrote:
Don wrote:I am sure that Vas learned a lot from looking at the Fruit code as he would be stupid not to take advantage of what is freely available. And I'm sure he developed as a chess programmer, learning things as he went along.
It's just hilarious. I'm also sure that many ppl these days learned a lot by looking at Ippo sources and also developed as chess programmers, learning things as they went long. And therefore, many new programs came out some of them stronger than Ippo, some stronger even than Rybka 4.
So do you really believe that???
Or you just believe it in case of Vas but not others. If that's the case, than it's pretty obvious that you are impartial.
The point you are choosing to completely miss is that Vas did what no other program has done since - he put a huge amount of ELO distance between himself and the others. These cloners are not doing that.

Pick your favorite clone and tell me how many hundred of ELO it is ahead of Rybka - and then we can talk. I will respect any program author that can gain 200 ELO on Rybka 4 and I can be sure that isn't because they took all the good ideas from Rybka.

There is nothing wrong with taking the good ideas that are freely available by studying open source programs and incorporating them into your own programs, I think most of us do that. But catching everyone else is a whole different world than passing everyone else by hundreds of ELO.

I think the last time this happened was with Richard Langs programs. He put a huge amount of distance between every other program. If his program was at a tournament, he usually won. Once in a while another program would beat his, but that was always an upset - just like any top 10 program can beat Rybka but not consistently.

He did a lot to improve his program from Rybka 1 beta to Rybka 3 (with some side help, Vincent can tell you more about it :)), but that had nothing to do with all the other programmers. Vas kept you guys so many years in the dark, and probably therefore you think of him as some chess god.
I've always suspected this was a vigilante justice thing. When a program gets close to Rybka there is a squeal of delight reporting that the "king is almost dead" as though this was some kind of deliverance from evil or with a "give him another kick" attitude. It's just plain childish and immature to carry on that way. Shame on Vas for the evil he has done us.
He started in the similar way as some of the new guys (you popularly call cloners), but the only thing he brought to chess community is his full pockets. If he shared knowledge or didn't go commercial he would certainly have more respect. Like this he's nothing but a parasite. Quite contrary to Ippo guys.
Being unable to see this is really a sad thing.
I'm not quite as hard core on this as some. I respect the guys coming in who admit up front that they used good ideas from the clones and from those who came before them and are able to put together a strong program based on their own code base. I just don't consider them hero's or trail blazers. Some of them may later prove they are and if they do then I'm sure you will probably turn on them. Right now they are hero's in your eyes because you think they are "hurting" someone you don't like or held back riches of some kind from you.

Tom Barrister
Posts: 227
Joined: Tue Oct 05, 2010 3:29 pm

Re: TEST position TCEC5- Houdini 1.03a-DRybka4 1-0

Post by Tom Barrister » Tue Nov 30, 2010 8:16 pm

Don wrote: (excerpts)

The point you are choosing to completely miss is that Vas did what no other program has done since - he put a huge amount of ELO distance between himself and the others.

Pick your favorite clone and tell me how many hundred of ELO it is ahead of Rybka - and then we can talk. I will respect any program author that can gain 200 ELO on Rybka 4

I'm not quite as hard core on this as some.
Point 1: Can you show us the archived list (i.e. a date from archive.org or a place we can all look at the old list) in which Rybka has a "huge amount of ELO distance" over the next competitor, and in which all of the leading competitors of that time were on the list? What do you define as "huge"? 40? 100? 200, as you seem to imply that Rybka once had over others (since you use that standard against the alleged clones today)?

Point 2: Do you honestly believe that you're not as hardcore as others? Somebody who just walked in off the street could easily believe that you're a Rybka shill, or at least (pardon the pun) a "Vas-kisser", not that I personally believe that, although I'm not far removed from being just in off the street myself.
Last edited by Tom Barrister on Tue Nov 30, 2010 8:31 pm, edited 1 time in total.
This production is being brought to you by Rybka: "The engine made from scratch.™"

User avatar
Don
Posts: 5106
Joined: Tue Apr 29, 2008 2:27 pm

Re: TEST position TCEC5- Houdini 1.03a-DRybka4 1-0

Post by Don » Tue Nov 30, 2010 8:30 pm

Tom,

The only thing that matters is how close you are to the top program. If you just measure improvement of each individual program it is based on how crappy it was at the point you start measuring.

There was a point in time when Rybka 3 first came out that there was a huge gap in performance between him and anyone else.

You picked Rybka 2 as your starting point and Rybka 2 only had a modest superiority but then you compared all of this to Rybka 3 and programs which were given a few years to catch up to Rybka 3. This is prime example of cooking the numbers. I don't dispute the fact that these other programs have caught Rybka, what I dispute is the notion that is was an independent development or that Rybka did not show the way.

I'm not really sure what point you are trying to make, but you seem to be making is sound like Rybka never was any better than anyone else and the only reason it was is because it was cloned from the inferior Fruit. Is that supposed to be your point here?

Or is your point that the other programs eventually (after a few years) caught Rybka? I don't think anyone is contesting that. I'm totally confused by your post.
Tom Barrister wrote:Since I don't have saved copies of these lists, and I don't know if histories of said exist for perusal anywhere else, I have to use archive.org. Using the standard 40/40 list from CCRL, starting May of 2007, or about the time that Rybka took its place at the top of the chess program world, and taking selected engines, e.g. the best of each variety that were around both then and now:

Code: Select all

5-21-2007

Rybka 2.3.1     3078
Zap Zanzibar    3021
Hiarcs 11.1     2993
D Shred 10      2945
Deep Fritz 10   2943
Naum 2.1        2937
Toga II 1.3     2875
Glarung 1.2.1   2859
Spike 1.2       2827
Bright 0.1d     2739
Crafty 20.14    2713
Now fast forward to today

Code: Select all

11-26-2010

Rybka 4         3264
Stockfish 1.9.1 3222
Naum 4.2        3184
D Shred 12      3127
Deep Fritz 11   3096
Zappa Mex II    3075
Hiarcs 12.1     3010
Toga II         3008
Bright 0.4      3002
Crafty 23.3     2946
Spike 1.2       2849
This is the improvement over the past 3 years and 6 months for each:

Code: Select all

D Rybka        186
Stkfsh/Glrng   363
Naum           247
Shredder       182
Fritz          153
Zappa           54
Hiarcs          17
Toga II        133 
Bright         263
Crafty         233
Spike           22
Using good-old Spike 1.2 Turin (which hasn't been developed since 2005) as the benchmark, we can conclude (right or wrong) that there hasn't been much, if any, artificial inflation in the ratings in the 3.5 years. Whatever ELO improvement there is has been made due to improvements in the technology, design, etc.

The question I have is this:

HOW can anybody in their right mind say that the last 200/300/whatever ELO has been the exclusive property of Mr. Rajlich and Rybka 3->4? .
Most of the modern improvements have been a result of taking ideas from the Rybka clones, specifically Ippolitto and family. These ideas have mostly come from Rybka.

These ideas are not the exclusive property of anyone - nobody is saying that. However credit where due - the ideas originally came from Vas and we are now using them.

Does this mean no program has any original ideas? Of course not. I go back to the 4 minute mile - when Roger Bannister first ran the 4 minute mile he showed the way - very shortly later everyone was doing it. In this case Rybka blew everyone away completely and not only did we find out that our programs sucked but we also get a glimpse into what we must do to fix it. And whether you like it or not you have to give Vas the credit for this. Even if you think he represents everything wrong with capitalism.

Since Rybka is closed-source (or whatever the correct term is), they certainly couldn't all have decompiled his program and used his ideas, such as they are. They must have had their own ideas or used/modified freely-available code/ideas of others.
All the program were mediocre until Ippo came out and the source code was immediately exposed. They were the only ones who had to decompile Rybka - everyone else now had an open source clone to work from.

Let's be reasonable about this. With or without Rybka programs are going to continue to improve and they will still do so long after Rybka. The point is that the appearance and subsequent scavenging operation was a big temporary boost to the strength of all programs being actively developed.

You should take note that nobody is passing Rybka like it's standing still - funny how that works isn't it? But Rybka 3 passed everyone like they were standing still because it was just a much better program than every other program including the program that it is claimed to be a clone of.

Every few years it happens, someone with a bunch of talent will pass everyone else and be the one to beat for a short period of time, perhaps even a few years and then everyone will catch up and this guy will drop off the scene either suddenly or gradually.

As for the ELO increases, Stockfish (which is a fork of Glarung) went up almost twice as much. Naum, Bright, and Crafy went up about 1.3 times as much, Shredder about the same, Fritz a bit less, Toga a bit less than that. Only Zappa and Hiarcs didn't keep pace, and of course, we can't expect an engine (Spike) that is exactly the same at both points in time to improve, especially since it was included as a benchmark against artificial inflation.

Apparently, a lot of the leading legal engines improved as much or more than Rybka has by mainly fair means. I'm sure that most programmers looked at the various open-source code of others, plus whatever they could glean from conversations, forum posts, emails, etc., took advantage of whatever new technologies were available, and improved their engines by these means. Since Mr. Rajlich doesn't live in a vacuum (as there IS some air, albeit rarified, up there on that high-cloud perch), he probably did that same.

What am I missing here? Why is Mr. Rajlich the way, truth, and the light, and all others pale before him?

PS: I wouldn't doubt it if Stockfish overtook Rybka before the next version of the latter is produced, supposedly in the fall of 2011. I also wouldn't be surprised, if Stockfish does whiz by Rybka , that Rybka updates/versions start cranking out a lot faster than they used to. Without the 100 ELO cushion----indeed if not even number one, and with no way to pressure the ratings lists people to exclude Stockfish, Mr. Rajlich may feel more compelled to improve his product (faster), so that he can possibly stay on top of the engine list, and thus give the public a reason to buy his product.

Any odds on Mr. Rajlich incorporating some of the ideas from Stockfish into Rybka 5?
I'm sure if there is something good he will use it, just as Stockfish and Komodo and many other programs have done with Robbo and family. I just don't understand this middle finger in the air attitude against Vas.

Dann Corbit
Posts: 9280
Joined: Wed Mar 08, 2006 7:57 pm
Location: Redmond, WA USA
Contact:

Re: TEST position TCEC5- Houdini 1.03a-DRybka4 1-0

Post by Dann Corbit » Tue Nov 30, 2010 9:05 pm

Tom Barrister wrote:
Don wrote: (excerpts)

The point you are choosing to completely miss is that Vas did what no other program has done since - he put a huge amount of ELO distance between himself and the others.

Pick your favorite clone and tell me how many hundred of ELO it is ahead of Rybka - and then we can talk. I will respect any program author that can gain 200 ELO on Rybka 4

I'm not quite as hard core on this as some.
Point 1: Can you show us the archived list (i.e. a date from archive.org or a place we can all look at the old list) in which Rybka has a "huge amount of ELO distance" over the next competitor, and in which all of the leading competitors of that time were on the list? What do you define as "huge"? 40? 100? 200, as you seem to imply that Rybka once had over others (since you use that standard against the alleged clones today)?

Point 2: Do you honestly believe that you're not as hardcore as others? Somebody who just walked in off the street could easily believe that you're a Rybka shill, or at least (pardon the pun) a "Vas-kisser", not that I personally believe that, although I'm not far removed from being just in off the street myself.
In the first SSDF list containing Rybka, Rybka was 75 Elo above #2:

Code: Select all

      THE SSDF RATING LIST 2006-06-08    106696 games played by  277 computers
                                           Rating   +     -  Games   Won  Oppo
                                           ------  ---   --- -----   ---  ----
   1 Rybka 1.2  256MB Athlon 1200 MHz        2931   38   -35   473   79%  2699
   2 Hiarcs 10 HypMod 256MB Athlon 1200 MHz  2856   24   -23  1076   76%  2661
Rybka 2's first SSDF list shows 91 Elo over the first Non-Rybka entrant:

Code: Select all

      THE SSDF RATING LIST 2007-04-21    111048 games played by  289 computers
                                           Rating   +     -  Games   Won  Oppo
                                           ------  ---   --- -----   ---  ----
   1 Rybka 2.3.1 Arena 256MB Athlon 1200 MHz 2962   47   -44   252   67%  2837
   2 Rybka 1.2  256MB Athlon 1200 MHz        2909   26   -25   845   72%  2748
   3 Hiarcs 11.1  256MB  Athlon 1200 MHz     2871   51   -51   189   52%  2858
Rybka 3's debut shows 90 Elo:

Code: Select all

       THE SSDF RATING LIST 2009-04-10   %119166 games played by  311 computers
                                           Rating   +     -  Games   Won  Oppo
                                           ------  ---   --- -----   ---  ----
   1 Deep Rybka 3 x64  2GB Q6600 2,4 GHz     3224   31   -28   803   84%  2941
   2 Naum 4 x64  2GB Q6600 2,4 GHz           3134   29   -27   714   74%  2955
When Don references 300 Elo, I guess he is referring to this:

Code: Select all

      THE SSDF RATING LIST 2010-03-21   %120316 games played by  311 computers
                                           Rating   +     -  Games   Won  Oppo
                                           ------  ---   --- -----   ---  ----
   1 Deep Rybka 3 x64  2GB Q6600 2,4 GHz     3227   27   -25  1005   83%  2962
  30 Fruit 2.2.1  256MB Athlon 1200 MHz      2833   19   -19  1385   62%  2750
394 Elo difference between Rybka 3 and the commercial edition of Fruit. {Clearly, a good chunk of this is due to the Q6600 verses the Athlon}

SSDF has not yet created a list which contains Rybka 4, however we do have this from CEGT and CCRL:

Code: Select all

CEGT:
no Program Elo + - Games Score Av.Op. Draws 
1 Deep Rybka 4 x64 6CPU 3249 23 23 500 71.2% 3092 43.6% 
347 Fruit 2.3.1 2796 20 20 698 48.5% 2807 38.3% 

CCRL:
Rank Name Rating Score Average Opponent Draws Games 
1 Rybka 4 64-bit 4CPU 3264 +19 -19 71.1% -141.0 44.1% 895 
  Fruit 2.3.1 2888 +14 -14 50.0% -0.4 46.1% 1721 
for differences of 453 Elo on 6 CPUs and 376 on 4 CPUs respectively.

So I think it is safe to say that the strength increase for high end hardware of Rybka over Fruit 2.3.1 is about 450 Elo {give or take} which is fairly substantial.

We all seem to speculate over who might have been involved in wrongdoing. Vas has admitted to study of Fruit, in his initial Beta release notes. Decompilation of Rybka has shown Fruit similarities.

However, the very fact that Rybka is bitboard shows that a rewrite was required. Did Vas do something wrong? I have no idea. Did people who decompiled Rybka do something wrong? I am not even sure about this.

At any rate, I think that there has been a lot of innovation by recent chess programmers. Especially Fabian and Tord spring to mind. I also think that everyone should take note of Dr. Hyatt's test methods. Crafty's strength has been growing by leaps and bounds, clearly due to his superior test methods. This is another kind of innovation.

Clearly, what Vas has done is also very innovative. You do not pull hundreds of Elo out of thin air.

Everyone wants to run to judgement. Don't we have something better to do with our time?

Post Reply