Chess 960 tournaments: variety

Sort:
ichabod801
Atos wrote:

It's not the first time you responded to my posts on 960 in an unnecessarily  combative manner. I have seen such stats but I don't have them at hand right this minute. I might look for them tomorrow, they are somewhere on Google I think.


I was not trying to combat you. I was certainly combative at Kasparov's statement, but that was not meant to be directed at you. I'm sorry if my post offended you so much.

spoiler_alert

Here's a game principle from 960 I've discovered that is very prevalent:  There are repeated opportunities in it for a lone attacking piece to do significant damage early in the game.  But early losses like this can be easily overcome.  invariably you can muster force around this isolated attacking piece,  and in doing so  increase your development while he is moving that one piece around to escape threats and sometimes he gets taken anyway.  There are at least five games I've played that illustrate this.

Also another principle - you very quickly learn to dispense with hackneyed notions about the worth of castling.  Let your opponent think you're forced to castle on one side or another and at the last moment just slide your king behind whatever fortified section exists and be done with it.

The above would be to just counter the idea of Atos that strategy or learning is not possible in 960.  Also note that any 960 opening position shares attributes in common with a large number of other 960 positions that are similar to it. 

Atos
ichabod801 wrote:
Atos wrote:

It's not the first time you responded to my posts on 960 in an unnecessarily  combative manner. I have seen such stats but I don't have them at hand right this minute. I might look for them tomorrow, they are somewhere on Google I think.


I was not trying to combat you. I was certainly combative at Kasparov's statement, but that was not meant to be directed at you. I'm sorry if my post offended you so much.


Okay, I am sorry also if I overreacted. You didn't like Kasparov's suggestion; I thought it was reasonable. (I don't like him too much as a person, but then if I went by persons I wouldn't touch 960.)

Atos

Of course GMs know, and in fact memorize, large numbers of opening lines. What the experiment mentioned suggests is that GMs (I think the experiment involved several GMs, and comparing them with amateurs) do not apply 'rote' memorization but that their memorization powers are greatly enhanced by understanding the positions, and knowing the reasons certain moves are played. In fact, to give an obvious example, even us patzers can understand why after 1. e4 e5 2. Nf3 Nc6 3. Bb5 a6 a Bishop move needs to be played, and it helps us to 'memorize' the lines. But it is not correct to say that GMs do not memorize opening lines, just that they apply a chess-specific mode of memorization.

kokino
Eberulf wrote:

atos:  as far as the propoganda angle,  if you go to the original DB link he supposedly compiled this from, it says "File does not exist".  I'm not saying he's responsible,  its just slightly odd.  Crossed out results,  files no longer existing, and this is the first hit in google. Don't know what it means except that 960 for whatever reason is an extremely contentious issue for a lot of people  (Didn't know this until a couple of weeks ago.)


Eberulf, just trying to bring some light about those links you have posted:

 

 The first DB you mentioned earlier, [posted on the Mark Weeks' blog page (btw, Mark is probably one of the best Chess960 contributors and reading his blogs is always useful for anyone willing to learn more about this chess variant. Besides, you can discuss about this directly with him as he is a chess.com member and admin of one of the Chess960 groups here: bemweeks (Chess960 RandomChess))] comes from one of the best resource in Chess Engine tournaments and testing :

http://www.computerchess.org.uk/ccrl/404FRC/games.html (this one is working)

 

You can download the 68K FRC games available there and draw your own conclusions. (I can anticipate you that this is the most complete DB of FRC games currently available online, but as I said, they are only computer games)

Unfortunately, there are roughly 700 chess960 games ever played (and annotated) by chess masters at top level, so the only sources to get more chess960games for statistics are computer games and amateur games (mostly played on-line under different circumstances which might cause some distrust on them)





The second link you have posted:

http://www.chess-960.org/php/show_entries.php?defaultordering=date

 

It was a good project that unfortunately, wasn't updated for the last 3 years, apart from that you can find:

20% of the games are Master games.

30% of the games are played by amateurs.

50% are played by computers (and we are talking about games from 2001 to 2006... bear in mind that CCRL site has computer games played "yesterday")

therefore, the current reliability of that DB is also "0"

 

With this "panorama", it is completely useless to provide a statistical evaluation of any FRC starting position. 

I personally have a DB with around 250,000 chess960 games (and I am sure that it is one of the largest available worldwide, and trying to enlarge it everyday with all the games I can get anywhere)

Even with those games, I cannot get any significant conclusion either. Just bear in mind that for classic chess, you need at least 1 million games to get a fair enough opening book and games reference... so, just extrapolating that into Chess960 and you would need 960 million games.... for the same reliability)

At any rate, I strongly believe that in the near future, there will be more resources available on this matter, and there will be more opening theory coming up (as far as I know, there are at least 3 sites offering "small" opening books /games explorers for FRC and they are feeding them with Amateur games to replace the lack of Master Games)

 

So, IMHO opening theory might no be relevant for playing Chess960 OTB (unless the players knew the SP beforehand), as most of you are posting here but when we talk about correspondence and online turn-based chess, I am absolutely convinced that it will become esential as it did for classic chess.

 

You are invited to check the information available on our group forum (it is public so you don't have to join to participate there):

stuck thread Chess960 RandomChess FORUM GUIDE

As soon as I can, I will create links to a external DB/ opening book I will create(feeded by my personal DB)... unless chess.com decided to "invest" some programming time for a game explorer based on the chess960 games archive. :)

 

(in the forum, there is a thread with links to most of sites which have a chess960 DB "in process")

spoiler_alert
kokino wrote:
Eberulf wrote:

atos:  as far as the propoganda angle,  if you go to the original DB link he supposedly compiled this from, it says "File does not exist".  I'm not saying he's responsible,  its just slightly odd.  Crossed out results,  files no longer existing, and this is the first hit in google. Don't know what it means except that 960 for whatever reason is an extremely contentious issue for a lot of people  (Didn't know this until a couple of weeks ago.)


Eberulf, just trying to bring some light about those links you have posted:

 

 The first DB you mentioned earlier, [posted on the Mark Weeks' blog page (btw, Mark is probably one of the best Chess960 contributors and reading his blogs is always useful for anyone willing to learn more about this chess variant. Besides, you can discuss about this directly with him as he is a chess.com member and admin of one of the Chess960 groups here: bemweeks (Chess960 RandomChess))] comes from one of the best resource in Chess Engine tournaments and testing :

http://www.computerchess.org.uk/ccrl/404FRC/games.html (this one is working)

 

You can download the 68K FRC games available there and draw your own conclusions. (I can anticipate you that this is the most complete DB of FRC games currently available online, but as I said, they are only computer games)

Unfortunately, there are roughly 700 chess960 games ever played (and annotated) by chess masters at top level, so the only sources to get more chess960games for statistics are computer games and amateur games (mostly played on-line under different circumstances which might cause some distrust on them)

...

With this "panorama", it is completely useless to provide a statistical evaluation of any FRC starting position. 

I personally have a DB with around 250,000 chess960 games (and I am sure that it is one of the largest available worldwide, and trying to enlarge it everyday with all the games I can get anywhere)

Even with those games, I cannot get any significant conclusion either. Just bear in mind that for classic chess, you need at least 1 million games to get a fair enough opening book and games reference... so, just extrapolating that into Chess960 and you would need 960 million games.... for the same reliability)


In random sampling you get accurate results regarding a certain population by only sampling an extremly small fraction of it.  You seem to be saying that even if we sample the entire population of 250000 960 games that the results are not significant.  They are signicant regarding those 250000 games.  How could someone say those 250000 games mean nothing.

If we were to compile statistics on the entire population of 960 games played here at chess.com (regarding highest winning pct for white, black etc,  it would certainly be reliable as to the 960 games played specifically here thus far.  I mean as long as you qualify your results accurately when stating them, then they are accurate.  (I'm stating the obvious.)  

But the problem with Weeks results is not population size but just miscalculation - and erroneously biased in favor of portraying white with inordinate winning percentages in chess 960.  I don't know if he's personally biased but his erroneous results are.

Here's his first table: 


SP    FEN    Gms    %Draw    %White    
Highest Overall Score White
24    NBQNBRKR    42    28.6    76.2    
868    QBBRKRNN    40    22.5    73.8    
74    NNRKBBQR    44    34.1    64.8    
424    RBNQBNKR    32    28.1    67.2    
376    NBRKBRNQ    26    11.5    75.0    
Highest Overall Score Black
92    NBNRKRBQ    28    3.6    41.0    
222    NQRKNBBR    48    12.5    37.5    
694    RQBKNBNR    34    17.6    35.3    
82    BNNRKBRQ    34    17.6    35.3    
396    QBRNNKBR    36    22.2    33.3    

And then he went back and revised the table here but still showing 960 opening positions strongly favoring white (in comparison to the top scorers for black):


SP    FEN    #    %W    %B    %D    
Highest Overall Score White
24    NBQNBRKR    42    61.9    9.5    28.6    
376    NBRKBRNQ    26    69.3    19.3    11.5    
868    QBBRKRNN    40    62.6    15.0    22.5    
266    NRKNBBQR    30    66.7    23.3    10.0    
729    RKNBBQNR    28    64.3    25.1    10.7    
Highest Overall Score Black
537    RNKBBQNR    28    17.8    53.6    28.6    
396    QBRNNKBR    36    22.2    55.6    22.2    
764    RBKNNRBQ    30    23.4    53.4    23.3    
201    QNRBBKNR    44    20.5    50.0    29.5    
82    BNNRKBRQ    34    26.5    55.9    17.6    
Traditional Start
518    RNBQKBNR    28    28.6    50.0    21.4    
Traditional Start (K&Q switched)
534    RNBKQBNR    22    27.3    40.9    31.8    


You provided the database to me so I tested it myself and my results are entirely different:

Top scoring 960 positions for white:

position        #   W# B#  D#   W winning %
BQRKRNNB 56 32 15 9 0.571429 0
NBRQKNBR 60 34 21 5 0.566667 1
QNNRKRBB 70 39 19 12 0.557143 2
NBRKBRNQ 72 40 21 11 0.555556 3
RKNBBQNR 56 31 15 10 0.553571 4
RKNRBBQN 58 32 19 7 0.551724 5
NRQKNRBB 78 43 30 5 0.551282 6
QBBRKRNN 80 44 20 16 0.55 7
RKNBNQBR 60 33 14 13 0.55 8
QRKRNNBB 102 56 33 13 0.54902 9
NBQNBRKR 66 36 14 16 0.545455 10
RNKRNBBQ 46 25 18 3 0.543478 11
NNRKRBBQ 46 25 13 8 0.543478 12
QRBKRBNN 70 38 23 9 0.542857 13
RKRBBQNN 74 40 24 10 0.540541 14
QRNNBKRB 52 28 14 10 0.538462 15
BNNBRKQR 80 43 27 10 0.5375 16
RBBNQKRN 58 31 18 9 0.534483 17
RKNBQRBN 60 32 19 9 0.533333 18
RKNQRBBN 62 33 20 9 0.532258 19
BBNRQKNR 62 33 18 11 0.532258 20
NBBQRKRN 64 34 21 9 0.53125 21
BBQRKRNN 72 38 19 15 0.527778 22
RKNQBBNR 76 40 27 9 0.526316 23


Top scoring 960 positions for black:

position        #   W# B#  D#   B  winning %
BBRQKNRN 62 22 34 6 0.548387 0
BRNQNKRB 52 18 28 6 0.538462 1
BRKNQBRN 70 20 37 13 0.528571 2
RNKRBBNQ 64 17 33 14 0.515625 3
RBNKBRNQ 68 27 35 6 0.514706 4
QBRNNKBR 70 21 36 13 0.514286 5
BRNBNKRQ 74 24 38 12 0.513514 6
BRNKNBQR 98 37 50 11 0.510204 7
RKNRBBNQ 54 17 27 10 0.5 8
BNRBQKRN 76 27 38 11 0.5 9
NBNRBKRQ 72 24 36 12 0.5 10
RKBNQBNR 66 22 33 11 0.5 11
RKNBNRBQ 56 21 28 7 0.5 12
BRQNKNRB 74 25 37 12 0.5 13
QNBRKRNB 76 26 38 12 0.5 14
QRBKRNNB 74 27 37 10 0.5 15
RBKNBQNR 82 32 40 10 0.487805 16
NQRKNBBR 76 28 37 11 0.486842 17
RQNKBBRN 76 29 37 10 0.486842 18
BNRNKBRQ 70 27 34 9 0.485714 19
BRKBRQNN 68 25 33 10 0.485294 20

[I could list all 960 positions for each, btw]

Even his game counts are different than mine, so who knows what happened with his results.

Just to review as my columns are a little different -  I have the top config for white as 57% and for black 54.8%.  Down the columns the difference between white and black seem consistently 2 or 3 % points or so. Compare that to Weeks results - in his first table its 76% white vs. 41% black. Even in his second table he has 61% white vs. 53% black.  That's WAY off from my results, and my results are correct.  Not to rub it in as anyone can make a mistake I guess.

spoiler_alert

(cont.)

Just anticipating possible objections  - you might ask, how reliable are the results from those 250000 games in predicting the winning pct for white/black in future games.  And I can think of an easy test.  Take statistics on the earliest half and compare it to statistics from the later half.  I'll look into it in a bit on that 64000 game DB.

kokino

@Eberulf,

everything has an explanation... :)

Well, regarding the differences in the figures: as I said, the CCRL database is updated almost daily... and you are comparing the calculation done with the current games done by yourself and the one done by Mark 1,5 years ago.

Let's take for instance the SP-024:

NBQNBRKR - when Mark did that chart, there were only 42 games available. Now, you say there are 66.

 


Regarding the miscalcullation, I believe you are not reading his posts correctly.

You are comparing %white wins Vs. %white overall score, that's why you get different outcome.

%white overall score = % wins + 1/2 % draws.

%white wins = #white wins / total # games


And finally, you are here providing the facts to me when I say there is not significance in creating stats and % to extrapolate for future chess960 with so few games. The uncertainty is so high that is completely useless (I am sorry, but when it comes to use technical words, my English is limited but I hope you understand what I mean)

A good example, taking the same SP - 024 (NBQNBRKR):

 

When Mark did the calcullation, there were a 61,9% of wins for white (sample 42 games). Now, according to your calcullations, there is a 55% of wins (sample 66)

what reliability can you get when just adding 24 games you get a 7 points @ winning rate difference?

 

(at any rate, this is quite interesting to me, as I am deeply working on that... so, I find this thread very instructive and it is always nice to get different points of view ... :))

kokino

(Cont.) by the way, I agree there is a miscalcullation on the Mark's 2nd post regarding the standard position figures:

Traditional Start
518 RNBQKBNR 28 28.6 50.0 21.4 39.3

clearly, this is wrong as he is giving 50% wins for black and obviously, it should be for white with an overal score of 61%.

spoiler_alert

"You are comparing %white wins Vs. %white overall score, that's why you get different outcome."

No, I'm not doing that - I only looked at white wins which he also has a column for.  I didn't understand white overall score and in his first table he crossed out all the results for white overall score anyway.

But he has a column for white wins.

 

--------------

 

But I will also rerun the test only up through the dates he used.

kokino

I have run a query in Chessbase again for that SP-024 (NBQNBRKR) and if you take the 42 games played before Dec 11, 2008 (according to the date Mark posted the first entry in the blog) I get this:

WHITE WINS : 26 games (62%) - overall (76%)

BLACK WINS : 04 games (10%) - overall (24%)

DRAW          : 12 games (28%)

[exactly the same figures as he mentioned on his second (and corrected) post]


taking the current sample available (66 games):

WHITE WINS : 36 games (55%) - overall (67%)

BLACK WINS : 14 games (21%) - overall (33%)

DRAW          : 16 games (24%)

[of course, exactly the same outcome you mention in your reply as we are taking the same figures]

[Observe how black wins % chance so significantly]

 


Taking the sample of my DB for that SP - 024 (159 games which include a significant change back as 95 games are "human" played):

WHITE WINS : 106 games (67%) - overall (73%)

BLACK WINS : 33 games (21%) - overall (27%)

DRAW          : 20 games (13%)

[Observe how the number of draws drop significantly compared to taking only the computer games... (human behaviour?) and again, the figures are changing too much to get any conclusion here which may apply to future games played with that SP ]

spoiler_alert

OK fair enough - I'll have to revisit this later today as I'm kind of fatigued with it now.

 

-----------------

 

I'll have to look into ChessBase as well, as I first tried ChessDB but it knows nothing of 960.  Arena is pretty slick and it understands 960 but wasn't able to figure out how to do queries.  Ended up coding about 100 lines of C++.

kokino

And I am sorry we are going away from the initial purpose of this thread by Ilmago. :)

So, my opinion about the different 4 options:

a) this is the current situation here at chess.com tournaments and I agree this should be changed into probably option "b", although online turn-based chess960 will be soon affected by games DB and probably will not make any differences as will happen to "b" and "c" options.

 

b) this is the system used for chess960 tournaments OTB (at least most of the one I have noticed of) Probably, it would be the best choice, however, I agree that you can simply "wait" for other players participating in the tournament to copy the moves and / or learn about possible mistakes... (moreover for next rounds, when you have a good number of games to study)... can provide you certain advantage beforehand... but we are talking about correspondence chess, and as I mentioned in previous posts, pretty soon, there will be a fair number of games for each SP available for everybody before starting a game  with any SP so, going to option "a", "b" or "c" will not make any differences.

 

c) this is also a good system, but as I mention in option "b", it will lose all the significance pretty soon when talking for correspondence chess.

 

 

d) I completely disagree with this system as it doesn't provide equal opportunities to everyone. I believe that both games against the same opponent should be played in equal conditions. (just to prevent what we are discussing here, whether some SP provides "certain" better pre-disposition to white or black)

kokino
Eberulf wrote:

OK fair enough - I'll have to revisit this later today as I'm kind of fatigued with it now.


 it is ok... :)

kokino
Eberulf wrote:

 

-----------------

 

I'll have to look into ChessBase as well, as I first tried ChessDB but it knows nothing of 960.  Arena is pretty slick and it understands 960 but wasn't able to figure out how to do queries.  Ended up coding about 100 lines of C++.


 Chessbase v.10 can work with chess960. (but using Shredder FEN only)

Regarding that point, there is also a lot behind the notation standards... :(

Arena can play chess960 and manage chess960 books but only when using FEN notation (not even X-FEN) so, there might be mistakes when reading some chess960 positions.

 

To run a querie in Arena you can import the PGN for the games (CNTRL + P):

and then in PGN\List Games

You can filter the games by FEN and then click "ELO STATS".... go to "general data" and you can get the information we are discussing here.

 

I hope this helps!

spoiler_alert

Kokino:

First of all as you say Weeks was using a smaller earlier version of the database so let that be noted again and I apologize for reading further into the error.  But as far as that DB, the more computer played games you add you don't have random fluctuation but rather convergence towards something that seems quite clear.  Look at my tables again from the previous page.  There are configurations that are strongly favored for black as well (not just white) and looking down the win % column in the two tables the differences in respective rows is like 2-4% between black and white - roughly the same advantage for white getting to move first.  In earlier iterations of the DB, that difference between black and white is greater but it gets smaller with more games being added.   Even in earlier versions there are still board configs that strongly favor black.  This all must be already known, explaining why in 960 tournaments here opponents have to play both colors.  But the original comment in this thread was that there were board configs that strongly favored white in 960 and that is only half the story.

Atos

Actually my comment was that: "That some initial positions seem to favor White strongly, others less or not at all." I don't really understand how there can be initial positions that favor Black.

spoiler_alert

If you just look at the results for the traditional start, its wildly incorrect to begin with (in that computer game DB)  and the more computer games you add the closer it gets to correct value.

Here it is through games played up through 1/6/2009:

                                              w%         b%

RNBQKBNR 32 10 15 7 0.3125 0.46875 878

And through the present:

RNBQKBNR 68 28 29 11 0.411765 0.426471 511

spoiler_alert

Just think about it for a second, Atos - how could it not be reciprocal between black and white, save the known advantage to white for moving first.

spoiler_alert

Also the other trend I personally want to look at is as follows:

If you looked at the DB through say only 10 games  and ranked starting positions by how much they favored black (which color we choose doesn't matter for reasons already noted)  there would be wild extremes. At the top you would have positions that win virtually all the time for black and at the bottom postions that virtually never win.  Well, we know those extremes will have to get modulated the more games we add.  But the question is by how much will these extremes diminish.  Actually as it stands now with the computer games the traditional setup is roughly in the middle, (I guess that can't be coincidence) but with a lot of other setups roughly in the middle as well.  So in the final analysis will it really be true there are positions that strongly favor black - we know there will be a convergence towards greater uniformity with more games added, but how much convergence.