Chess960-FRC Setup Data is Excellent

[GeneM]

.
The chess960-Frc statistics at the following Http URL are fantastic:

http://www.computerchess.org.uk/ccrl/40 ... y_eco.html


I am looking for good hard data about which non-traditional Frc setup would have....
[A] a relatively low draw rate, and
a relatively low victory rate for White, and
[C] no bishops starting on any corner square, and
[D] the two white knights starting on the same shade of square (in contrast to the traditional setup).

As of 2012/Feb/11, the best candidate setup seems to be:

RBNKBRNQ , DrawPct= 10.40% , WhiteWinPct= 48.60% , ID= S-664 (R-561) , Based on 106 speed games between computers.

Happily, the "reciprocal" to setup R-561, meaning QNRBKNBR or R-061, has similarly low draw and white victory rates. These similarities tend to confirm and validate the statistical data gather for R-561.

QNRBKNBR , DrawPct= 16.70% , WhiteWinPct= 47.60% , ID= S-207 (R-061)

Unfortunately, a comparison of the reciprocal pair R-862 & R-362, meaning the traditional setup and its reciprocal, shows the two setups with wildly different rates for draws and white wins. This tends to reduce our ability to make conclusions from reciprocals, unfortunately.


When your speed tournaments are completed someday, ideally you could run a longer time-control tournament among a much smaller number of Frc setups, just the most promising (according to criteria like mine above). Then your results should be published on ChessBase.com.

I do not often re-visit this website, but I am interested in any chess960-Frc developments along these lines.
Please feel free to send email to me.

Thank you.

GeneM (Gene Milener)
genem@castlelong.com

http://www.CastleLong.com/

Chess960-FRC book, on Amazon.com
Chess960-FRC book, on Amazon.de

Possibly related thread on this forum?
.

[Ray]

.

Thank you for your kind words about our chess960 work ! I run all the games for this list, and Kirill has designed the excellent website with those statistics.

.
When your speed tournaments are completed someday, ideally you could run a longer time-control tournament among a much smaller number of Frc setups, just the most promising (according to criteria like mine above). Then your results should be published on ChessBase.com.

I do not often re-visit this website, but I am interested in any chess960-Frc developments along these lines.
Please feel free to send email to me.

Please post a proposal. I've always been interested in some smaller scale longer time control work, but have been short of ideas as to what to do.

[GeneM]

.

Please post a proposal. I've always been interested in some smaller scale longer time control work, but have been short of ideas as to what to do.

QNRBKNBR 84 16.70% 47.60% 0.06 MB R-061 S-207
RBNKBRNQ 106 10.40% 48.60% 0.07 MB R-561 S-664


RNBQKBNR 125 19.20% 54.40% 0.09 MB R-862 S-518
RNBKQBNR 74 28.40% 46.60% 0.05 MB R-362 S-534


RNBBKNQR 120 19.20% 57.10% 0.08 MB R-745 S-549
RQNKBBNR 106 17.00% 61.30% 0.07 MB R-245 S-506


NBRKBRNQ 120 12.50% 66.30% 0.08 MB R-579 S-376
QNRBKRBN 106 11.30% 46.20% 0.07 MB R-079 S-301


==== ==== ==== ====


QUESTIONS TO TARGET WITH EXPERIMENTS:


1. Are the surprising and unsettling differences, in draw pct and White win pct, between the following two reciprocal setup pairs robust --- meaning would the pct statistics within each pair still show these large discrepancies if the time controls were increased to perhaps 30moves/30minutes repeating?

a) R-579 & R-079
b) R-862 & R-362 (traditional and its reciprocal)


2. The setup R-561 has a low draw pct and a low ratio of White wins / Black wins. Thus R-561 could be fantastic for slow chess if its speed chess statistics were tested and found robust even with longer time controls.


3. It would be interesting to gain PGN game score data for games involving the two setups of a reciprocal pair where the White win / Black win ratio differs a lot between the two setups. By human examination the games' moves, it might be plausible to articulate more concretely what subtle factors have a surprisingly large effect on the W/B wins ratio. Potentially these factors could be abstracted to match factors in known traditional setups with various known ratios.


4. Do the statistical similarities within the reciprocal pair R-745 & R-245 remain robustly similar even as time controls are increase to long controls?


5. One of my favorite questions to ask would be --- Which chess960-Frc setups have the highest rate of Opposite wing castling (one color castles the king to column-c, while the other color castles the king to column-g in the same game). I suppose you already have the raw information to answer this question, at least for speed chess. But half the effort in science is mining all the raw data.

Thank you.
.

[Ray]

Thanks for the suggestions, I'll take a look

For point 5, I'm sure CQL (chess query language) can help
http://rbnn.com/cql/

but, quite complex to understand and use.

EDIT: I have emailed the CQL authors to see if they can provide the script to pull out games with opposite side castling. For standard chess I'm sure it is a 5 min job for them, but if they don't understand chess960 castling rules then they may not be able to help. I may have to try and figure it out myself.

[GeneM]

.

... if they don't understand chess960 castling rules then they may not be able to help. I may have to try and figure it out myself.

King and the castling rook end on exact same squares as in traditional chess1. SAN notation of castling is no different; O-O or O-O-O (letters Oh, not digits zero).
The *only* castling rule difference is the squares that the king a rook start the game on.

In chess960 (and ideally in chess1 also), FEN encodes castling eligibilities more robustly, as 'AHah' instead of 'KQkq'.
.

[Ray]

The *only* castling rule difference is the squares that the king a rook start the game on.
.

Yes - for clarity, what are those ?

king ends up on g1 - where can it have come from => b-f ?
rook ends up on f1 - where can it have come from => c-e ?

king ends up on c1 - where can it have come from => d-h ?
rook ends up on d1 - where can it have come from => a-c ?

I'm currently only getting < 100 games out of over 100,000 where opposite side castling has occurred.That can't be right, I must have the syntax wrong

EDIT: Well I've spent several hours on this and made no further progress. Unless someone comes along with some CQL and FRC knowledge, this isn't going to get solved. As an example of something which doesn't work:

(match
:pgn FRC107100.pgn
:Output frc_Oppcastling.pgn

(position
:movefrom K[b-f1]
:moveto .[g1]

:movefrom R[g-h1]
:moveto .[f1]

:movefrom k[d-h8]
:moveto .[c8]

:movefrom r[b-c8]
:moveto .[d8]

:flipcolor

)
)

"K" = white king and "k" = black king, i,e. white pieces are capitalised.

I'm getting a lot of kingside O-O castles but very few queenside O-O-O castles

[Ray]

Bumping this so I don't forget about it. I've been very busy

[Ray]

Yes, I haven't got to it yet. Statistical analysis of FRC is of great interest to me. I need to think through the suggestions above, how to approach each one.

[Adam Hair]

It appears that CQL does not understand castling in Chess 960. In fact, I am not certain it reads FEN strings. I tried it out and the PGN it produces from the query does not contain the FEN tag. The PGN is meaningless without the FEN tag.

The only information I can derive about castling from the FRC database is that O-O has occurred, irrespective of color, 84213 times and O-O-O has occurred 53681 times.

[Ray]

My own lack of knowledge of FRC castling was also holding me back with CQL. I know where the king and rook end up after castling, but I wasn't sure about the range of squares that each could come from to achieve the new positions. What was your script for the above ? Just looking for the "after" position is not good enough, because the king and rook can end up there later in the game through normal moves.

[Ray]

I'm sure for a programmer this would be easy - read the pgn, and search for white O-O and black O-O-O and vice versa, and output the games to a new file. My programming skills are nowhere near good enough though. Having got the pgn, I could then run through Kirill's scripts to give us the position where opposite side castling has occurred the most

[Adam Hair]

My own lack of knowledge of FRC castling was also holding me back with CQL. I know where the king and rook end up after castling, but I wasn't sure about the range of squares that each could come from to achieve the new positions. What was your script for the above ? Just looking for the "after" position is not good enough, because the king and rook can end up there later in the game through normal moves.

I may have to try again. For the life of me, I can not remember how I specified the before position of the rook. As for the king, it has to be between the rooks in the initial position. So, it can range from b1 to g1 (b8 to g8).

I have to do some work at the moment. I'll be back in an hour or two with an idea I have forming in my head.

[Adam Hair]

CQL definitely does not read the FEN string. If you do the following position search:

(match
:pgn BBNRKNQR.pgn
: output file.pgn
(position
Ke1
)
)

CQL finds the king at e1 in all 118 games.

If instead you try this:

(match
:pgn NQRKRNBB.pgn
: output file.pgn
(position
Kd1
)
)

CQL outputs 0 games, even though all 124 games begin with the king at d1.

As far as actually searching for examples of castling, I tried this out:

(match
:pgn NQRKRNBB.pgn
: output file.pgn
(position
:movefrom Kd1
:moveto .[g1]
:movefrom Re1
:moveto .[f1]
)
)

CQL did not find any games that matched the query, even though White castled O-O several times in that PGN.

What I have found from studying the output PGNs from CQL is that it flags (as an error) the first move it finds that could not have occurred if the initial position is the standard one.

[Adam Hair]

This is not the detailed information that GeneM was asking about, but I do have some more precise information concerning castling in Chess960.

Same side, O-O castling: 24,293 times
Same side, O-O-O castling: 11,994 times
White O-O, Black O-O-O: 4530 times
White O-O-O, Black O-O: 3987 times

I found a version of Scid that can handle Chess960, though it may have some bugs. Also, finding castling moves with Scid is slightly ambiguous. So these numbers are not exactly correct. I hope to check with a different program soon. Also, I do not have a way to directly count when one side castles and the other side does not. The program I hope to use can do this.

[Adam Hair]

Here are the correct statistics (thanks to Ed Collins):

Castling Statistics:
--------------------
White castled Kingside and Black castled Kingside: 26,787 (25.0%)
White castled Kingside and Black castled Queenside: 6,088 (5.7%)
White castled Kingside and Black did not castle: 12,679 (11.8%)

White castled Queenside and Black castled Kingside: 5,577 (5.2%)
White castled Queenside and Black castled Queenside: 11,422 (10.7%)
White castled Queenside and Black did not castle: 10,142 (9.5%)

White did not castle and Black castled Kingside: 10,735 (10.0%)
White did not castle and Black castled Queenside: 9,030 (8.4%)
Neither player castled: 14,640 (13.7%)

[GeneM]

Here are the correct statistics (thanks to Ed Collins):
Castling Statistics:

Great data guys. Thanks so much.

I am surprised and a bit disappointed in one fact revealed by your data:
The rate of OPPOSITE wing castlings in chess960 (10.7%) seems to be no higher than in traditional chess1.

Such games have a lower draw rate, and are too infrequent just for proper variety.

GeneM