Shogi rating and ELO

[Adam Skalný]

Hello shogi fans,

I've created simple page for tracking rating of Slovak shogi players.
We started to use it from our last session - few of our club game
results are stored there. Rating system used is ELO similar but bit
simpler to one used by FESA. While I was coding, handicaps become
impassable obstacle. Do you know how FESAs rating system cope with
this problem please?

Are handicaps assigned according to grade system mentioned on FESA
page? E.g. 15th vs 5th kyu => 2 pieces. Providing it's true, is this
"table" available somewhere?

I also found table proposing standard handicap according to rating
burrowed somewhere in shogi.net ( http://www.shogi.net/mso2000/mso-shogi-rules.html#anchor2236759
)
H ERD
L 100
B 200
R 250
RL 350
2P 450
4P 550
5P 750
6P 1000

Is this used by FESA? Or it was just rule of MSO at 2000?

Thank you for your advice.

Cheers,
Adam

PS: Rating page url is http://rating.shogi.sk if you are interested
but there's not that much to see, yet.

[Larry Kaufman]

The FESA system works like this for handicap games: The ratings of the
handicap giver is converted to a "rank" using decimals (i.e. 3.2 Dan). The
rank is then reduced by the rank equivalence of the handicap given, from a
table. Then the revised rank is reconverted back to a rating to be used for
calculating the rating difference to be used in the rating formula.
Note that this is different from just assigning a rating value to each
handicap, because the ranks grow in width as you go up in strength. This is
logical because an extra piece might not mean very much in a game between
beginners but it is a big advantage in high-level play.
As for the table, my recollection is that the rank values are 0.8, 2.0,
2.8, 3.6, and 4.8 for lance thru two piece handicap, but I'm not so sure
about this and I've asked for confirmation. These values are in the ratio of
2-5-7-9-12, which were the rank values of the handicaps in pro play prior to
WWII.

Larry Kaufman

[Bernhard]

Hi,

thanks for the answer, Larry.

On 15 Jul., 06:30, "Larry Kaufman" <lkauf...@comcast.net> wrote:
>      As for the table, my recollection is that the rank values are 0.8, 2.0,
> 2.8, 3.6, and 4.8 for lance thru two piece handicap, but I'm not so sure
> about this and I've asked for confirmation. These values are in the ratio of
> 2-5-7-9-12, which were the rank values of the handicaps in pro play prior to
> WWII.

At the FESA-page (http://www.shogi.net/fesa/) you can also few the
sourcecode of there rating program:
http://www.shogi.net/fesa/turnering.lisp

I'm not sure, if this code is up to date or if the values have changed
again.

The following lines of the code are important for the thread-opener:

(defvar *m-handicap* 15)
(defvar *l-handicap* 100)
(defvar *b-handicap* 200)
(defvar *r-handicap* 250)
(defvar *rl-handicap* 350)
(defvar *2p-handicap* 450)
(defvar *4p-handicap* 550)
(defvar *5p-handicap* 750)
(defvar *6p-handicap* 900) ; 1000 in 2000
(defvar *s-handicap* 150)
...
(defparameter *m-handicap-r* 0.20)
(defparameter *l-handicap-r* 0.60)
(defparameter *b-handicap-r* 1.50)
(defparameter *r-handicap-r* 2.10)
(defparameter *rl-handicap-r* 2.70)
(defparameter *2p-handicap-r* 3.60)
(defparameter *4p-handicap-r* 5.00)
(defparameter *5p-handicap-r* 6.50)
(defparameter *6p-handicap-r* 8.00)
(defparameter *s-handicap-r* 1)

Because your answer, Larry, I'm understand now, why there are two of
this tables.
Thanks.
(Still I'm not absolutely sure, why this doubleconversation is made. I
see, that this two tables are not linearly dependent, but than I ask,
why a 6P Handicap have a ratingdiff of 900 points, but only of 8
grades?)

Bernhard

[George Fernandez]

In the U.S., rated handicap play has been regularly used for over 25 years.
Since the merger of the rating systems [U.S. & European] in 2000,
Pan-Atlantic Ratings Chairman Asle Olufsen, Larry Kaufman and others have
monitored, the ever increasing handicap game results, and made slight
adjustments to the "Suggested Handicaps" along the way.
I don't think that an official Pan-Atlantic Rating System "Suggested
Handicaps" table exists. Also, as far as I know, FESA currently does not
rate handicap games.

The current "Suggested Handicaps" at the New York Shogi Club are as follows:
0-20 Furigoma [Even]
21-60 Sente [Move]
61-180 Kyo [Lance]
181-300 Kaku [Bishop]
301-420 Hissha [Rook]
421-540 Hissha-Kyo [Rook+Lance]
541-660 Ni-Mai [2P]
661-780 Yon-Mai [4P]
781-900 Go-Mai [5P]
901-1020+ Roku-Mai. [6P]

Regards,
George

[Adam Skalny]

Hi,

thank you for help Larry and Bernhard it seems a bit clearer now ^_^

Seems a bit complicated but I'd like to have handicaps as an option in rating system because handicaps

So algorithm is basically:
(I'll use game of players P1 (rating 835, bishop handicap) vs P2 (rating 403) as an example)

1) Get handicap giver rating.
E.g. P1 => 835points
2) Look at table in paragraph 2 of http://www.shogi.net/fesa/rating.intro and calculate players decimal grade
E.g. 10th kyu is from 800 to 880 thus result is 10 - 35 / (880 - 800). Which means that player P1 has grade of 9.57kyu
3) Get handicap grade penalty and subtract it from handicap givers decimal grade
E.g. B => 1.5 according to Bernhard's code fragment. Thus effective grade of P1 is 9.57 + 1.5 = 10.07kyu
4) Convert back grade to rating using table from step 2.
E.g. 11th kyu is from 720 to 800 so effective rating of P1 is 10.07kyu => 800 + (800 - 720) * 0.07 = 805points
5) Finally calculate ELO rating difference using diff = k * (1 - res ^ ... ) formula.

Is this correct?

@George: Thank you for table update I'll use that one instead of FESA suggested one.

Once again thank you for help.

Cheers,
Adam

On Wed, Jul 15, 2009 at 8:01 AM, Bernhard <shog...@arcor.de> wrote:

Hi,

thanks for the answer, Larry.

On 15 Jul., 06:30, "Larry Kaufman" <lkauf...@comcast.net> wrote:

>      As for the table, my recollection is that the rank values are 0.8, 2.0,
> 2.8, 3.6, and 4.8 for lance thru two piece handicap, but I'm not so sure
> about this and I've asked for confirmation. These values are in the ratio of
> 2-5-7-9-12, which were the rank values of the handicaps in pro play prior to
> WWII.

--~--~---------~--~----~------------~-------~--~----~
You received this message because you are subscribed to the Google Groups "SHOGI-L" group.
To post to this group, send email to sho...@googlegroups.com
To unsubscribe from this group, send email to shogi-l+u...@googlegroups.com
For more options, visit this group at http://groups.google.com/group/shogi-l?hl=en
-~----------~----~----~----~------~----~------~--~---

[Adam Skalny]

Sorry for sending incomplete message. Missing sentence follows :)
* because handicaps are used quite often in our "club" games.

[Fabien Osmont]

FESA DOES rate handicap games.
Asle Olufsen would be probably the best person to explain it as rating
officer from this side of Atlantic.
Asle, any comments to put there?
Rgds
Fabien

[Larry Kaufman]

----- Original Message -----
From: "Fabien Osmont" <fab...@osmont.org>
To: <sho...@googlegroups.com>
Sent: Wednesday, July 15, 2009 4:28 AM
Subject: Re: Shogi rating and ELO



FESA DOES rate handicap games.
Asle Olufsen would be probably the best person to explain it as rating
officer from this side of Atlantic.
Asle, any comments to put there?

I was summarizing Asle's comments sent to me when I described the way
handicap games are rated, so no need for him to comment.

Rgds
Fabien

Larry

[Larry Kaufman]

----- Original Message -----
From: "Bernhard" <shog...@arcor.de>
To: "SHOGI-L" <sho...@googlegroups.com>
Sent: Wednesday, July 15, 2009 2:01 AM
Subject: Re: Shogi rating and ELO

I think that this part of the code is left over from before we switched
to rank-based handicap values, and is no longer used.

...
(defparameter *m-handicap-r* 0.20)
(defparameter *l-handicap-r* 0.60)
(defparameter *b-handicap-r* 1.50)
(defparameter *r-handicap-r* 2.10)
(defparameter *rl-handicap-r* 2.70)
(defparameter *2p-handicap-r* 3.60)
(defparameter *4p-handicap-r* 5.00)
(defparameter *5p-handicap-r* 6.50)
(defparameter *6p-handicap-r* 8.00)
(defparameter *s-handicap-r* 1)

Yes, these must be the current values. They are in the ratio I
indicated (2-5-7-9-12), but 25% lower across the board than I remembered.

Because your answer, Larry, I'm understand now, why there are two of
this tables.
Thanks.
(Still I'm not absolutely sure, why this doubleconversation is made. I
see, that this two tables are not linearly dependent, but than I ask,
why a 6P Handicap have a ratingdiff of 900 points, but only of 8
grades?)

The second table is a revision of the old one, they don't need to
correspond precisely. The rank values are lower for the obvious reason that
on average ranks are wider than 100 Elo (if you give equal weight to Dan and
kyu ranks).

Note that the rank values are consistently lower than the official
standard of one rank per handicap (or two ranks in some cases). This is
because experience shows that the official handicaps are not enough to
equalize the chances.

Bernhard

Larry

[Larry Kaufman]

----- Original Message -----

From: Adam Skalny

To: sho...@googlegroups.com

Sent: Wednesday, July 15, 2009 3:07 AM

Subject: Re: Shogi rating and ELO

Hi,

thank you for help Larry and Bernhard it seems a bit clearer now ^_^

Seems a bit complicated but I'd like to have handicaps as an option in rating system because handicaps

So algorithm is basically:
(I'll use game of players P1 (rating 835, bishop handicap) vs P2 (rating 403) as an example)

1) Get handicap giver rating.
E.g. P1 => 835points
2) Look at table in paragraph 2 of http://www.shogi.net/fesa/rating.intro and calculate players decimal grade
E.g. 10th kyu is from 800 to 880 thus result is 10 - 35 / (880 - 800).

 

 I believe we use the     midpoints, so that 10 kyu is exactly 840, and so 835 would be (10 + 5/80) kyu or 10.06 kyu.

 

Which means that player P1 has grade of 9.57kyu
3) Get handicap grade penalty and subtract it from handicap givers decimal grade
E.g. B => 1.5 according to Bernhard's code fragment. Thus effective grade of P1 is 9.57 + 1.5 = 10.07kyu
4) Convert back grade to rating using table from step 2.
E.g. 11th kyu is from 720 to 800 so effective rating of P1 is 10.07kyu => 800 + (800 - 720) * 0.07 = 805points

 

Here too the conversion back uses the midpoint as the rating equivalent of the rank.

 

5) Finally calculate ELO rating difference using diff = k * (1 - res ^ ... ) formula.

Is this correct?

 

     Except as noted above, I believe it is correct.

@George: Thank you for table update I'll use that one instead of FESA suggested one.

 

     Note that George's table is not a table of "fair" handicaps but of suggested ones. If you use his table the stronger player will still be expected to win well over 50% of the time, maybe around 60% or more.

Once again thank you for help.

Cheers,
Adam

 

Larry