Not logged inRybka Chess Community Forum
Up Topic Rybka Support & Discussion / Rybka Discussion / Rybka4 database of games where Rybka doesn't play good endga
1 2 3 4 Previous Next  
- - By dr_zied_haddad (**) [fr] Date 2010-06-29 19:09
Hi all,
in order to help everyone and the Rybka team to improve the endgame knowledge of Rybka (and also may be testing other top engines in endgames compared to Rybka), i invite everyone who finds that Rybka play in endgame was poor to submit his pgn or fen here.

Please tell the time control (short or long time control certainly give different evaluations of positions and choice of the moves)

The main purpose is to build together a database of positions where Rybka evaluation is not good, so all of us (but also when we develop settings) we can see the improvements that Vas will bring with his next update and next versions.

Evaluating endgame play is really far from being easy and i think that this method could help evaluating more easily new features in endgame knowledge.

We can also run tournaments to see which engine plays endgames better.
Parent - - By Vasik Rajlich (Silver) [pl] Date 2010-06-30 06:56
This will be a short thread! :smile:

Vas
Parent - - By yanquis1972 (****) [us] Date 2010-06-30 10:51
not least of all because it's ostensibly the second of its kind. 

i do agree though vas, minus some obvious problems (which may have more to do with evaluation [equally/more important to me as OTB strength] than strength) R4 is a machine in the endgame.

imo the emphasis for next rybka should be -- (and perhaps some of these could be toggles; i know a lot of people have suggested these for a lot of different things, & maybe a tryout of a couple simple ones in 4.1 wouldn't be too bad...[say BUP which iirc you claimed was simple to implement]) -- tactics (believe it or not; still not sure they've improved (!)), king safety (same, probably to a lesser degree), specialized endgame knowledge/positions - zugzwang, fortress, etc. speculative play.

i just read a post of yours saying R4 was an experiment in that last direction. however in my view it is awfully solid & doesn't seem to take unnecessary risks or offer bad sacrifices that fail upon further examination (not a criticism at all, in fact what i would hope for in a top of the line engine, but not what i think of as 'speculative' either).  i recall the idea of a 'wild' personality of some sort; i would still like to see a rybka that emphasizes opening lines of attack against the enemy king, with as close as we can get to an optimal understanding of the critical importance of a vulnerable one...i think R4 has improved incrementally in this regard but it's not the swift-death sort of beast i sorta expected based on early reports.

i recall you saying also sometime ago that you believed you could develop a R4 'kinghunter' type that played 'not bad' (or however you put it) chess; in this case i assume you meant something close to 3000 elo (my sights were set much lower). i do think R4 with a bit of tweaking could do that...R4 dynamic anyone??? i found R3D very useful in a surprising amount  of positions; i think R4D, perhaps with king safety cranked a bit, would be even moreso....the combination of rybka's search & knowledge, with genuine speculative play tossed in, would make for a fantastic alternative opinion in a number of positions, both clearcut & otherwise...i was always astonished at the ingenuity of R3D in even known or relatively quiet positions.

hrm, that got a touch off-topic...carried away by my coffee.
Parent - - By dr_zied_haddad (**) [fr] Date 2010-06-30 19:41 Edited 2010-06-30 19:50
I share many of your ideas.
i think that tweakening the same engine to get a dynamic, king hunter etc. is a bit difficult, thus someone needs to work on different beta versions provided by Vas (even old ones between R3 and R4).
May be by analysing games of different R4 betas we could choose the betas that most fit attacking style, dynamic style, king hunter style and work on them a little bit by changing some scalings of variables to get a good result. This doesn't mean to get engines of the same strength (but rating that ranges +/- 100) and this is possible with some patience and devotion.

I assume that Vas's plans are not exactly these plans, that why someone of his team must make this specific work.

Concerning endgame knowledge, Rybka understanding is already terrific, but not perfect as it tends to spoil some positions where some more knowledge could be helpful. in the next update version, it could be interesting to have two different updates. A Rybka 4.1 with bug fixes +++, and another Rybka lets say Rybka 4 beta16 (as the latest know beta is 15) with bugfixes but also implemented with new endgame knowledge especially for rook endgames.
Parent - - By keoki010 (Silver) [us] Date 2010-06-30 22:08
Gotta say:, "Dream on; We'll be lucky if/when we get 4.1". :twisted:
Parent - By dr_zied_haddad (**) [fr] Date 2010-06-30 23:43
"i have a dream!". It's Vas who can make it a reality.
Parent - - By Vasik Rajlich (Silver) [pl] Date 2010-07-01 06:31
The Rybka 4 bugfix will be only a bugfix, I am not going to get carried away with that. The only really crucial bug in Rybka 4 is the MultiPV bug, but I'll fix a couple of other smaller bugs while I am at it.

The general setup is that we'll have two versions, the commercial version and the remote version. New features will be introduced in the remote version and will trickle down to the commercial version. (For Rybka 4, there was no pipeline yet.) The playing styles of the commercial and remote versions are quite different. I'll keep it that way. In general, the commercial version will tend to have bigger changes in playing style from version to version (ie. from Rybka 3 to Rybka 4 and again to Rybka 5), I think this is a good thing and something which normal users like. The changes in the remote version will be more incremental - the remote version is aimed at a smaller, more expert audience, who cares more about accuracy than things like coolness and novelty. If you are reading in our forum, you are probably in this second category. :smile:

I think it's best for us to launch the remote version and then I can write about everything. Rybka 4 will make more sense then. :smile:

Vas
Parent - - By Mark (****) [us] Date 2010-07-01 10:04

> the remote version is aimed at a smaller, more expert audience, who cares more about accuracy than things like coolness and novelty.


accuracy _is_ cool and novel!
Parent - By Vasik Rajlich (Silver) [pl] Date 2010-07-02 05:51
That's your "expert user bias" talking. :smile:

Vas
Parent - - By Labyrinth (*****) [us] Date 2010-07-01 13:14

>The playing styles of the commercial and remote versions are quite different


How do they differ?
Parent - By Vasik Rajlich (Silver) [pl] Date 2010-07-02 05:51
I'll write this up later - so far the remote version has been too much talk and not enough action. :smile:

Vas
Parent - - By keoki010 (Silver) [us] Date 2010-07-01 14:50
When will we get the bugfix? That's what most of us are waiting for.  MPV is a pain without it. :smile:
Parent - - By Vasik Rajlich (Silver) [pl] Date 2010-07-02 05:53
I hope in July - I'm taking care of some personal things ATM (and handling the forum :smile:).

Vas
Parent - By keoki010 (Silver) [us] Date 2010-07-02 13:12

> <SPAN class=htt>Response to</SPAN> Vasik RajlichI hope in July - I'm taking care of some personal things ATM (and handling the forum :smile:).


Ok! Just impatient. :grin:
Parent - - By yanquis1972 (****) [us] Date 2010-07-02 13:52
not to rub it in (really) but its truly amazing how this couldve gone unnoticed. i dont think i notice it with r4 1cpu for some reason, but with a multicore computer it hangs for me in almost every position after a very short while.
Parent - By Vasik Rajlich (Silver) [pl] Date 2010-07-03 07:28
I think some testers noticed it but I decided that it was normal. :smile:

Vas
Parent - - By stephanie (**) Date 2010-07-01 18:45
No, endgame play is real bad and has bugs.

Test position:

k5b1/1p6/p7/P3B3/1KP5/8/8/8 b - - 0 0


Rybka 4 plays Bxc4 and loses. Rybka 3 does not play it and evaluates it right. Can you please explain, is this a "feature", accidental bug, bug on purpose or an elo strength cap? And also there are lots of passed pawn games where rybka is in a centipawn advantage and ends up in a draw.
Parent - - By Wayne Lowrance (***) Date 2010-07-01 19:34
Wow, this is really bad !  A major bug to me anyhow.  R4 finds
the bishop capture at depth 32 on my computer and stays with it
despite hitting deep in TB's which tells me I gotta add another 6
man position in my 6 man TBs.
Follow up at depth 52 still capturing with over 100,000 TB hits.
Wayne
Parent - - By Wayne Lowrance (***) Date 2010-07-01 19:38
Follow up. R4 realizes as soon as she catures the move leads #14.
How in heck can that be. gotta be a hole some where in R4.
Wayne
Parent - By Wayne Lowrance (***) Date 2010-07-01 19:43
Heheheheh, follow up to follow up.I have kbpkpp TB and it works.
So again, what is happening here with the lil girl. ? Her TB access
is telling her NO and she says, yes yes/
Wayne
Parent - By Maxiator (***) [de] Date 2010-07-04 09:33
Funny: I become victim of the Aquarium panel bug described elsewhere, when analyzing the position with R4. Having stopped another game and R4, setup position and start analyze with R4 but still the tab with the old analysis of R4 is visible->unload engine.... but now clicking on R4 Tab shows the notation of the game before (and board synchronously changed) and clicking on notation pane restores the position of stephanie; after another stop R4 and restart all went in correct manner...
VISTA 32,DualCorePC,R4 w32,Aquarium 4.05
btw R4 prefers the capture on c4 at d=46 and start position is 8 Pieces not in TB6-online (?) but there must be a great hole in Nalimov usage because R4 obviously don´t use when looking further; the bug maybe simple, for example R4 doesn´t look when the starting position is out of table base
Second I suspect R4 puts a greater emphasis on moves that alters material/scores in short than quiet but equal scored moves
max
Parent - By Fulcrum2000 (****) [nl] Date 2010-07-01 19:55
Parent - - By dr_zied_haddad (**) [fr] Date 2010-07-02 01:59
Hi Stephanie
A very good example.

A few other examples where Rybka in my opinion chooses not to play the variations which give him the best chances.

8/6b1/7p/P2N3R/2k2P2/7P/r7/7K w - - 0 49


8/8/R3pk2/7K/P7/8/2r3PP/8 w - - 0 41


3k4/5p2/3P1p2/p2BpP2/P1p1P2P/4K1P1/1r6/8 b - - 0 45


8/3B4/8/5pp1/P2p4/r6P/2PK1kP1/8 b - - 0 41


8/1pp4p/8/2p2p1k/2PP1P2/2P4P/P5K1/8 w - - 0 39


6k1/2b2p2/6pp/pB3P2/8/8/5P1P/6K1 b - - 0 46
Parent - By dr_zied_haddad (**) [fr] Date 2010-08-05 20:47
Another very interesting position from a correspondance game.
8/1b3p2/1P3kp1/P1R5/3N4/8/1r6/4K3 w - -

I think this position is a draw, even if white seems to have a huge advantage with two very advanced passed pawns, but the fact that one of the pawns is a Rook pawn, and Knight vs Bishop (in fact in this ending the Bishop is far much stronger than the knight in my opinion) will probably leed to a draw.

Rybka4 evaluates this position as White having a clear advantage, For the moment, the sole engine that shares my understanding of the position is Naum 4.2, even Zappa II Mexico is not as good as usual in this position.

What are your opinions?
Parent - By Vasik Rajlich (Silver) [pl] Date 2010-07-02 05:54
This looks like a side-effect of the blind bishop knowledge. Rybka 4 assumes that the side with only a bishop and the wrong rook pawn (and nothing else) can't win.

Vas
Parent - - By Banned for Life (Gold) Date 2010-07-02 03:59
Yes! We are interested in coolness and novelty and don't give a shit about accuracy! :eek:
Parent - - By Vasik Rajlich (Silver) [pl] Date 2010-07-02 05:56
Assuming that you're not being sarcastic :smile:, this means that you are not an expert user. Sorry. :razz:

Vas
Parent - By yanquis1972 (****) [us] Date 2010-07-02 06:42
heh i mistook what you were saying at first too.
Parent - - By Nelson Hernandez (Gold) [us] Date 2010-07-03 20:32
Accuracy?  We don' have no accuracy.  We don' need no steenkin' accuracy!

I can hardly believe Vas really believes this coolness and novelty bullshit!  He doesn't get that you are being bitter and sardonic.  I just can't believe Vas really thinks we give two figs about "coolness and novelty"!  Accuracy is the ONLY thing that matters, godammit!  How can anyone make any progress in chess research without frigging accuracy?!  I am flabbergasted.  WTF!?!
Parent - - By Vasik Rajlich (Silver) [pl] Date 2010-07-04 05:44
Just wait a bit please. :smile: Our original plan was to release Rybka 4 and then the remote version one right after the other, but reality intervened again.

Vas
Parent - - By Nelson Hernandez (Gold) [us] Date 2010-07-04 11:35
You still don't get it.  We all want accuracy.  Thinking we want coolness is like assuming we have the mentality of teenagers.
Parent - By Akbarfan (***) Date 2010-07-04 12:16
.
Parent - - By Felix Kling (Gold) [de] Date 2010-07-04 13:43
okok. so what you want is e.g. a higher node count, right? ;)
Parent - - By Nelson Hernandez (Gold) [us] Date 2010-07-04 16:51
Let me ask you.  Do I harass Vas about BUP because I think BUP is cool, or because having it is more accurate?  Can you imagine me swaggering down the street in sunglasses and Italian shoes, telling my entourage "my chess program allows bishop underpromotion"--ooooo! ahhhh! you don't say! wow! you are The Man, Nelson! we'll never be cool like you!

Ditto for my annoyance at the pawn-up rook endgame problem.  Accuracy is everything!
Parent - By Vasik Rajlich (Silver) [pl] Date 2010-07-05 08:02

> Can you imagine me swaggering down the street in sunglasses and Italian shoes, telling my entourage "my chess program allows bishop underpromotion"--ooooo! ahhhh! you don't say! wow! you are The Man, Nelson! we'll never be cool like you!


Remove the "sunglasses and Italian shoes" part, replace "allows bishop underpromotion" with "has a cool playing style", and you're pretty might right on. :cool:

Vas
Parent - By Akbarfan (***) Date 2010-07-04 21:12
RYBKA - for the serious chess player  ....   :mad:
Parent - By Geomusic (*****) Date 2010-07-06 19:31
R4 remote hopefully will have a trial free membership or discounted one with restricted access so users can see what they are getting :)
Parent - - By mikal (**) [us] Date 2010-07-02 06:15

>I think this is a good thing and something which normal users like. The changes in the remote version will be more incremental - the remote version is aimed at a smaller, more expert audience, who cares more about accuracy than things like coolness and novelty. If you are reading in our forum, you are probably in this second category. :smile:


Can you please be more specific of what type of 'normal' users are you talking about? For what is worth the chess world is already small compared to majority of the gaming or sport industry, the number of people who use chess softwares are even much more smaller and most of them use simple user friendly ones like chessmaster etc. Now the people who regularly use top engines like rybka are almost in the utmost minority which most are simply just searching for the best software for analyzing and playing the most accurate and strongest chess entity.

And how long do you expect this business model to last? What if some engine or so called 'clones' comes out and outplays your remote service, what will you do?
Parent - - By Vasik Rajlich (Silver) [pl] Date 2010-07-03 07:32

> Can you please be more specific of what type of 'normal' users are you talking about? For what is worth the chess world is already small compared to majority of the gaming or sport industry, the number of people who use chess softwares are even much more smaller and most of them use simple user friendly ones like chessmaster etc. Now the people who regularly use top engines like rybka are almost in the utmost minority which most are simply just searching for the best software for analyzing and playing the most accurate and strongest chess entity.


I should have used different terms. Substitute "elite users" for "expert users" and "expert users" for "normal users".

There is a big difference between the average Rybka 4 customer and for example the average customer who posts on this forum.

> And how long do you expect this business model to last?


Decades - I'll just set it up and let it roll. :smile:

> What if some engine or so called 'clones' comes out and outplays your remote service, what will you do?


Admit defeat and do something else. :smile: I am aware that this is extremely sensitive to performance.

Vas
Parent - - By Nelson Hernandez (Gold) [us] Date 2010-07-03 20:44
So there are three tiers.

Elite users - Super GMs and people with very deep pockets -- incline toward "accuracy"
Expert users - sad sacks who post in this forum and wish they were rich and cool enough to be elite users -- incline toward "novelty"
Normal users - the truly clueless who buy the commercial product, don't post in this forum and wouldn't remotely consider becoming elite users -- incline toward "coolness"

I would add two more tiers.

Sub-normal users - morons who pirate the product and then play 1/0 games on Yahoo! with it -- incline toward "bullet strength"
Brain-dead non-users - unquestioning buyers of Fritz and Chessmaster who may or may not have heard of Rybka, and think it must not be any good because it's a weird name
Parent - By Geomusic (*****) Date 2010-07-06 19:34
lol at NH, Elite users = "people with very deep pockets" so true isn't that how the world works :)
Parent - - By Nelson Hernandez (Gold) [us] Date 2010-07-03 20:49
Just to be clear: I have no problem at all with your employing a market segmentation approach to marketing your products.  That makes very good business sense.

What doesn't make sense, and prompts scornful remarks, is your presumption that those of us without the deep pockets value anything other than accuracy.  Accuracy is everything.  Coolness is nothing.  Novelty is a natural by-product of accuracy.
Parent - - By Vasik Rajlich (Silver) [pl] Date 2010-07-04 05:49
The basic remote version won't be more expensive. In fact, if you consider home electricity and hardware costs, it will be cheaper, especially for users like you. :smile:

Of course, if you are so inclined, it will be a bit easier to spend money. It will take just a few clicks of the mouse, instead of assembling some complex hardware in your home as is currently the case.

When you see it, I am sure you will agree with what I have been telling you: it's win/win/win.

Vas
Parent - - By Uly (Gold) [mx] Date 2010-07-04 06:29

> if you consider home electricity and hardware costs, it will be cheaper


But I'll be using the same electricity as if I was using the engine locally, and my hardware can be used to run other things (for instance, other chess engines, I really doubt remote Rybka will be the answer to everything considering I only rely on Rybka 4 for 20% of my games, so keeping hardware for the top engines that are being released still makes sense).
Parent - - By Vasik Rajlich (Silver) [pl] Date 2010-07-05 08:04
You'll use much less electricity if your engine isn't blasting away at 100%. (See Lukas' comments in the thread about hyperthreading for an example.)

You can also use remote Rybka and Rybka 4 at the same time, another reason to have different playing styles between the two.

Vas
Parent - - By rookie (**) [us] Date 2010-07-05 23:11
I wish you great success with remote rybka; I have no opinion about whether it will fly, but for your sake, as one who has had your intellectual property ripped off, I hope it works.
  That said, the same thought occurs to me that I've expressed at dreary length elsewhere on the forum: remote e-mail and your other examples did not represent hobbies. Hobbies are different in that they are illogical. Few care a whit about saving electricity--hobbies aren't about efficiency. Hobbies are about process and illusion, and remote rybka scores a bull's eye in gutting that process and illusion. Applying logical reasoning to an illogical activity that has no rational goal like computer chess is irrational. Hobbies are not, in essence, the sum of the activities that make them up; they are about the superstructure of fantasy and imagination that each persons builds around those activities. I think that for most computer chess hobbyists, remote rybka will crash those fantasies.

  But this only applies to the crazies that hang about this forum. For that part of the market for whom computer chess is only a means to achieve a goal, that is, for those who only want the best analysis and could care less about the process, I have no reason to think the remote idea wouldn't work. Provided the price is right and the product is clearly superior to anything else available.
Parent - - By Uly (Gold) [mx] Date 2010-07-05 23:18
The email analogy doesn't work anyway, it's akin to charging for each click you do while checking your emails, how much would you be willing to pay for the best email service available?
Parent - - By Geomusic (*****) Date 2010-07-06 19:37
pay for email service who does that anymore?
Parent - By Uly (Gold) [mx] Date 2010-07-06 20:58
Nobody, that's why the analogy wouldn't work unless the access to remote Rybka was free, then I'm sure people wouldn't mind never owning the engine, but the model wouldn't make sense, so an engine online and email shouldn't be mixed, they have no relation whatsoever.
Parent - - By Nelson Hernandez (Gold) [us] Date 2010-07-04 11:49
There is a fallacy in your thinking.  If I am happy with my present hardware, and I am, then subscribing to the remote version represents 100% incremental cost to me.

Further, if you consider the fact that I use 16 threads 24 hours a day, doing that on your computers means spending more over the course of three or four years than I would for a lot of tangible things.  At the end of which, I have nothing of tangible value that I can point to and say "that's mine".

It simply does not work for me and never could.  But, never mind that, I still want accuracy in those things I can afford.
Up Topic Rybka Support & Discussion / Rybka Discussion / Rybka4 database of games where Rybka doesn't play good endga
1 2 3 4 Previous Next  

Powered by mwForum 2.27.4 © 1999-2012 Markus Wichitill