Not logged inRybka Chess Community Forum
Up Topic Rybka Support & Discussion / Aquarium / Help with game analysis
- - By cfagyal (*) Date 2008-09-15 02:50
So I have a game which is 72 moves long that I have in my database.  It's a game I played this evening on playchess, and I am trying to use "Game Analysis" to get Rybka to analyze the entire game.  This all works fine, except for the fact that Rybka won't put any notations after move 25.  Unfortunately for me, I made some blunders after move 25 which caused me a stalemate instead of a win..several blunders in fact.  I need the latter part of the game annotated, and no matter what I set the settings for analysis to, Aquarium refuses to annotate the entire game.  I've set the option for "don't analyze if the score is greater than ...." to 99999 to make sure even a mate condition will be analyzed, yet it refuses.

Anyone have any advice?

Parent - - By buffos (Silver) Date 2008-09-15 05:27
when you watch the game analysis procedur, doenst the algorithm assign evaluations to each move?
How much time do you give it for the whole game?
Parent - By cfagyal (*) Date 2008-09-15 11:10
Yes it assigns evaluations to each move, but it doesn't notate each move past a certain point.  I had this happen with this game last night and a different game where it only notated up to move 10.  It seems to stop notating when the evaluation reaches about -4 or -5 for either side.  Unfortunately, as i'm still learning, I *want* the endgame notated as well, so I can see where I have made mistakes in the endgame.  As it stands, I really need to use Arena to annotate my games, because Aquarium doesn't complete the job...
Parent - By KylieFan (*) Date 2008-09-15 12:48
This is very useful information, and a show-stopper for me (if no work-around exists).
I was at the point of buying Aquarium until reading about this limitation in the analysis. I think it is quite a serious bug for beginners like myself who want to analysis the whole game. A shame, because Aquarium otherwise seems good for analysis.

I hope that the Fritz 11 / Rybka combination does not suffer from this problem (unfortunetly the free Arena does not offer a Blunder check).
Parent - - By KylieFan (*) Date 2008-09-15 15:37
Just a thought - are you using the demo version (which I think stop analysis after castling) or the 'full' version ?
Parent - - By cfagyal (*) Date 2008-09-15 21:36
I'm  using the full version which I just got on Saturday, and yea Arena has it's issues as well.
Parent - - By cfagyal (*) Date 2008-09-16 00:23
Oh also, Buffos, I'm setting it to manual evaluation of a set time per move (10 seconds and 10 plies) and a set prolongation per move (5 seconds and 10 plies).  I have the Do not analyze score at 99999 (i've had it anywhere from 300 to 99999 trying different numbers to try to figure out the correlation between that number and when it stops).  The commenting options are set at max variation length of 12, untrusted engine halfmoves (which im not sure what that means) at 7, and maximum variations at 0 - any amount.  I'm using the hugebase.cdp as the opening database.
Parent - - By buffos (Silver) Date 2008-09-16 05:59

> I'm setting it to manual evaluation

you better use the first tab (the automatic) , not because manual is bad, but i have not used it , dont know the behaviour and i cannot help there :)
Parent - - By KylieFan (*) Date 2008-09-16 07:24
Can I just confirm - you have had a full game analysis (using the 'automatic' option) right until the last move ?
Parent - - By buffos (Silver) Date 2008-09-16 07:52
I have analyzed more than 10000 games using game analysis
Every move is analyzed but not every move has variations.
Only moves that change the evaluation of the position have variations.

And when i say change the evaluation of the position i dont mean from +3 to +7. This is from lost to lost.

I dont know if there is an internal thershold on which moves to extend and which note.

I also think that game analysis can  be improved.
Parent - By Vempele (Silver) Date 2008-09-16 09:06
I got a bunch of +156.xx scores (about 10 pawns in reality). It seems one-percent-of-win resolution isn't quite enough when the scores are high...
Parent - - By KylieFan (*) Date 2008-09-16 09:27
I don't quite follow your "from +3 to +7. This is from lost to lost" comment (a difference of 3 points is not a loss at the beginner / low club level).
If I understand correctly, it seems that once the score difference gets over say +3 or +4 it will not analysis (or show blunders) anymore ??

An example - in a game both sides play is perfect, until I blunder a rook or queen at move 25, then will analysis enable me to see further blunders of (for example) +- 1 at move 70 ?

(this scenerio is possible (and very helpful) in example Fritz 5.32, etc. This is hugely important for beginners)
Parent - By buffos (Silver) Date 2008-09-16 09:36
Yes i think that the user should be able to tune the thershold of where moves get analysis.
Parent - - By cfagyal (*) Date 2008-09-16 13:00
Kylie, your assumption is accurate.  Around -4 or -4.5 in the games I am looking at is where annotations (other than rybka's evaluation of the score of the position) stops.  Yet I need the annotations/analysis in these situations just as much as in the opening, maybe more so, as I work to learn tactics.  I NEED to be shown where I missed good moves in the endgame, and I NEED to be shown blunders that caused a loss, or in the case of the game I am looking at, caused me to lose a won position, because I took a QK vs K position and incorrectly moved the king into a stalemate. 

The analysis of these games is the same whether it is "automatic" analysis versus manual.  Now I can, using Infinite analysis, go move by move and analyze these endgame scenarios myself, but this is far more tedious than necessary. 
Parent - By buffos (Silver) Date 2008-09-16 13:07
i completely understand your point. I guess the programmers decided that the time of analysis should better be spend at moves that really do matter in the result of the game (this is also what i would do if i was doing such a function, that wanted to stay simple without much user parameters), but i completely understand the beginners point of view, so i guess this is a very valid user request , i mean an option for a threshold to which moves get annotated and which not , so an 300 value would annotate every single move .
Parent - - By Dadi Jonsson (Silver) Date 2008-09-16 13:32

> I NEED to be shown blunders that caused a loss

This is always shown.

I think the philosophy behind the approach used in Aquarium is that you learn most from the moves which change the outcome of a game. Studying those positions deeply will advance your game faster than spreading your time over every blunder you make.

> caused me to lose a won position, because I took a QK vs K position and incorrectly moved the king into a stalemate.

You say it caused you to lose a won position, but a stalemate is a draw. Anyway, a blunder that leads to a draw from a won position should always be annotated. Was it not annotated in this case?
Parent - - By cfagyal (*) Date 2008-09-16 14:14
Dadi, I guess I wasn't overly clear...when I say "caused me to lose a won position.. " which resulted in a stalemate (a draw), to me that is a loss..because I had an obvious win and screwed it up. 

But to be clear, yes the game in question was 77 moves.  Around move 25 the score rybka indicates is that I am up 450 centipawns (-4.5 evaluation for white - I was playing black).  At this point not a single move is annotated from that point forward, only evaluation scores are indicated.  I have hand analyzed using Rybka's infinite analysis from move 25 through the end of the game and there are cases (two blatant ones where I turned a mate in X into a -4 or -5 evaluation by blundering, and another blatant one where I turned a mate in 1 into a stalemate due to blundering) where I made massive blunders, and nothing is annotated.  I agree Dadi that the study of positions where you blundered causing a loss is of vital importance.  I also think, in this case, study of the positions (which i've started using infinite analysis) where you made moves which significantly increased the duration of the game, are important too.  Around move 30 or 35, I had mate in 7.  The game ended on move 77 in a stalemate...  That is a VERY poorly played endgame by me, and I need to understand why.

Buffos:  In my opinion you should be able to have "blunders" shown, based on the settings in the show blunders dialog box as to what a "blunder" is (is it a 200 centipawn eval difference? is it 300 centipawns? That configured value should be what is used to determine a blunder).  Thus for an entire game, no matter what the evaluation of the position is, if a blunder is made, it should be annotated with rybka's evaluation of the line that should have been played.  These analysis are invaluable for a player who is newer to chess.  I have understood how to play chess at a basic level for years, but I have never done any study or any work at trying to improve until the last few weeks.  Thus for me, it is invaluable learning to be able to see where I made mistakes, replay the line(s) that Rybka, or some other engine suggests, and see what I could have done to play better.

Parent - - By KylieFan (*) Date 2008-09-16 16:08
Nicely put - I agree 100%.

I think I will still buy Aquarium, and hope that the problem is addressed in the future, but it did make me think twice about perhaps buying Fritz 11 GUI instead.
Parent - - By tigershark (**) Date 2008-09-16 18:22
See my other thread on Rybka only showing me 2 moves. I don't think it's a GUI problem; it's a bug in Rybka. Basically, unless I've misunderstood, Rybka is no good for analysing games once the position has gone beyond + or - 3 or 4; it just won't show much.  It's a shame but there you go. 

I'm now using Deep Rybka 3 human inside Fritz GUI for Deep Position Analysis

and I'm using HIARCS for full game analysis
Parent - By buffos (Silver) Date 2008-09-16 19:24
i think i have noticed that when this happens Rybka returns just a single move , the winning one. No variations.
Parent - By buffos (Silver) Date 2008-09-16 16:13

> In my opinion you should be able to have "blunders" shown

i dont think we disagree at all. I was saying exactly the same thing. :)
I some time find myself prefering the old game analysis method used in CA. But i guess this is a totally different story.
Parent - By Dadi Jonsson (Silver) Date 2008-09-17 09:00

> turned a mate in 1 into a stalemate due to blundering

This one should have been annotated with a variation, as all moves affecting the outcome of the game.
Parent - - By KylieFan (*) Date 2008-09-16 23:57
It seems to be a problem with the GUI - since the engines Crafty and Delfi also seem to have this problem when performing a 'Game Analysis' or Blunder Check under Aquarium (If it was a Rybka specific engine problem then at least another engine could be used for Blunder checks)

Has anyone tried Fritz 11 GUI / Rybka for a Blunder check or Game Analysis in this situation - same problems ?
Parent - - By buffos (Silver) Date 2008-09-17 05:12
Not annotating above +4 is not a "problem" but a decision.

The above post talked about rybka not sending variations when score is big.
Parent - - By KylieFan (*) Date 2008-09-17 09:06
But the same problem occurs with Crafty and Delfa under Aquarium (not only Rybka) !
Parent - - By buffos (Silver) Date 2008-09-17 09:08
i am not talking about annotating i am talking about what the engine shows in infinite analysis.
Parent - - By ppipper (*****) Date 2008-09-17 09:53
Not showing more than 2 plys in infinite analysis by Rybka 3 when score is +5 (-5), is considered a decision / bug as it is already discussed in other threads.

But other engines that have not reported this bug should be able to annotate whatever the evaluation is.

IMO, game analysis feature must be definetly improved as soon as possible: one of the clear advantages of using Aquarium is for analysis purposes, and only IDeA seems to be working properly (might be because we have nothing to compare with). Graphical annotation, chart's visualization when blunder checking, etc etc are some other features that need to be revisited.

Parent - By buffos (Silver) Date 2008-09-17 09:54

> Graphical annotation, chart's visualization when blunder checking, etc etc are some other features that need to be revisited.

we totally agree.
Parent - - By cfagyal (*) Date 2008-09-17 11:58
Yea I don't like the fact that Rybka 3 won't show the entire line it has come up with when the score is above a certain point.  I ended up having to use Toga 1.4 Beta 5c to annotate my games, because Rybka just doesn't cut it at the moment, and unfortunately because the game analysis feature won't completely analyze the game with any engine (I tried several aside from Rybka), I had to do it manually as well.
Parent - - By KylieFan (*) Date 2008-09-17 13:54
I'm just totally amazed that more people don't mention this important problem (like putting your head in the sand and pretending that the problem doesn't exist). I guess that Game Analysis and Blunder Check is not that important here!?

I guess I (and other beginners) have to buy Fritz 11 GUI + HIARCS 12 package for whole game and blunder analysis (not in itself bad!)
Parent - - By tigershark (**) Date 2008-09-17 14:25
Exactly what I'm doing.  Now I've bought Rybka 3 I'll still use it for the occasional deep position analysis where the position is probably under the + or - 3.  But I'd never have bought it if I'd known about this bug. And HIARCS will be my engine of choice for general analysis now (HIARCS is easily strong enough to analyse my games which is the main reason I buy a chess engine; I was attracted to Rybka 3 because of the "Human" engine but HIARCS also tries to play Human like moves)

It seems that Rybka 3 has been developed solely for engine vs engine players (and those players who never make mistakes or play against people who never make mistakes of the order of + or - 3).  The masses who use it for analysis have not been catered for.  This is a mistake and goes against what Vas said he was trying to do.

I'm pretty disappointed
Parent - - By ppipper (*****) Date 2008-09-17 20:00
Well, maybe Ribka issues should be discussed in other forum, and IMO we cannot claim about whether she has been developed solely for engine vs engine, or not. We have this forum as very good source of information and we all know what are the weakness of this engine, and some of the bugs.

However, although I agree with the fact that Aquarium is a very good and very different tool for analysis, it turns out that most of us find it very powerful tool for advanced users and really difficult one for simple purposes as analyzing a game, check blunders or just have a chart that let you easily know how the trends of the game have gone (I am not talking about the lots of bugs discovered by users who have paid to use Aquarium but have not been paid for discovering them; fortunately Aquarium's team really seems to be wotking on fixing them and they are releasing updates every "two" weeks)

I am not still disappointed because I have decided to give one more chance to Aquarium, but too much time invested by this moment....
Parent - By Zarkon (***) Date 2008-09-18 03:23
I also am not confident that the blundercheck and game analysis works well and finds what I would expect it to find, which is why I suggested that it be tested by comparing its output to Fritz's.

The advantage of a simple blundercheck where you specify a delta for ? (and ideally a second delta for ?? as with babaschess) and a depth or time is that you know exactly what you are getting.

I hope these two features (and of course the database support) are improved, because I very much like the interface.
Parent - - By vroger007 (**) Date 2008-09-17 23:12
yes, it is due to aquarium...
back in the early beta-testing days I've sent the same remark to the gui designers:

> Just reminding you that game analysis in beta5 still gives only annotations when a player's advantage is less than about 3 pawns,
> even if we are setting the parameter to 7 pawns... hoping you won't forget to have this corrected !

and got the reply:

> This will be rather difficult...

I still fail to see what could be so difficult about it :-)
it's partly the reason why I switched back to chessbase gui,  a second reason being the huge complexity of the aquarium gui... :-(
Parent - - By tigershark (**) Date 2008-09-18 09:04
aquarium gui will come good in the end I am sure. The team are so focussed on liasing with users I';m confident it will be a success

The problem seems to be with Rybka though. I use Rybka in Chessbase gui and it still won't show the variations when the score is plus or minus 3
Parent - By ppipper (*****) Date 2008-09-18 09:13

> I use Rybka in Chessbase gui and it still won't show the variations when the score is plus or minus 3

Rybka doesn't show more than 2 plys when eval is + - 5. See Rybka forum for further details.

Parent - - By ppipper (*****) Date 2008-09-18 09:10
I cannot see why it can be so dificult, and why is so diferent to have +3 rather than +8 for these purposes.

None of these bugs / features have been fixed / implemented in any update since Aquarium release; maybe this is the right time.

Parent - By ppipper (*****) Date 2008-09-18 13:36 Edited 2008-09-18 13:58
this is also already said in other threads, but it looks really strange to me when in the analysis & notation pane the engine analyzes a move that is check and still '+' sign does not appear at all.

Please, fix it.


Parent - - By Arkansaw (***) Date 2008-09-18 16:18
Get Fritz.
Parent - By ppipper (*****) Date 2008-09-18 17:30
Fritz is quite buggy free, but is very limited in terms of analysis. Aquarium is really a good approach, but currently is just that: a good approach.

It needs serveral updates to become a serious GUI.
- By gregor (**) Date 2010-10-24 19:20
Looks like this issue has not been resolved yet!!

I found the same issues on my system: Deep Rybka 4 Aquarium (4.0.5).

Sorry to say that, but this is very annoying ... 

It's very important for me to get a "full game" analysis, not only half the story.
To me this does not look as a Rybka issue, similar behaviour with other engines, eg. stockfish: game analysis stops after evaluation around +/- 5.00.
- - By andrey lezhnev Date 2010-10-27 12:22
I have recently purchased Rybka Aquarium 4 and I have one question question. Which of three engines Deep Rybka 4960 w32, Deep Rybka SSE42 w32 or Deep Rybka 4 w32 is the strongest in terms of a match analysis while playing regular chess and why?
Parent - By nebulus (****) Date 2010-10-27 12:40
Deep Rybka SSE42 version will run about 5-10% faster on CPUs with SSE4.2 (Intel's i-series) or SSE4a (AMD's Phenoms) instruction sets, thus a bit stronger than Deep Rybka 4.

Deep Rybka 4 960 supports Fischer Random.
- By gregor (**) Date 2011-01-25 17:26
Problem still evident   -   Aq 4.06 - Build 386

I had a clear win with KQN vs K and only got a stalemate because of a blunder.

Aq "blunder check" nor "game analysis" found the mistake.
Analysis stopped after winning position was reached not taking the user defined threshold (99999) into account.

Not good.
Up Topic Rybka Support & Discussion / Aquarium / Help with game analysis

Powered by mwForum 2.27.4 © 1999-2012 Markus Wichitill