matching problem

[drfg2008]

I have a large number of football-matches where a home team plays against an away team (see following example).

Home / Away

Bayern München - Hertha BSC
Augsburg - Hamburger SV
1. FC Köln - Bremen
Hertha - Augsburg
Bremen - Bayern München

...

Now I try to find a way of 'clustering' those teams who ever played against each other.
(Just using the league -as an example- is not possible, since most teams also play in different leagues, some are newcommers, ... .) Each team has a unique Team ID.

My solution with clustering / factor analysis is heavily complex and need python. Isn't there an easy solution to the problem?

[Art Kendall]

Are there multiple matches between a given pair of teams?�

Roughly how many teams are there altogether?

How is the input data now arranged? Are they in an SPSS system file, a spread sheet, a text file, etc.

Are there a number of seasons?

Why do you think you need PYTHON?

Once you have a matrix with (all teams as Home) by (all teams as away) what else do you want to know?

Are there occasions where two teams would play at a third location?

Are the character strings the unique IDs?� Are you very sure that there are no variations like spacing, diacriticals, etc. in the ID for a specific team?

On 5/3/2012 5:28 AM, drfg2008 wrote:

I have a large number of football-matches where a home team plays against an
away team (see following example).

Home / Away

Bayern München - Hertha BSC
Augsburg - Hamburger SV
1. FC Köln - Bremen
Hertha - Augsburg
Bremen - Bayern München

...

Now I try to find a way of 'clustering' those teams who ever played against
each other.
(Just using the league -as an example- is not possible, since most teams
also play in different leagues, some are newcommers, ... .) Each team has a
unique Team ID.

My solution with clustering / factor analysis is heavily complex and need
python. Isn't there an easy solution to the problem?



-----
Dr. Frank Gaeth
FU-Berlin

--
View this message in context: http://spssx-discussion.1045642.n5.nabble.com/matching-problem-tp5682738.html
Sent from the SPSSX Discussion mailing list archive at Nabble.com.

=====================
To manage your subscription to SPSSX-L, send a message to
[hidden email] (not to SPSSX-L), with no body text except the
command. To leave the list, send the command
SIGNOFF SPSSX-L
For a list of commands to manage subscriptions, send the command
INFO REFCARD

===================== To manage your subscription to SPSSX-L, send a message to [hidden email] (not to SPSSX-L), with no body text except the command. To leave the list, send the command SIGNOFF SPSSX-L For a list of commands to manage subscriptions, send the command INFO REFCARD

[David Marso]

Please define your question with greater specificity and try again Frank!
I have NO idea of what your problem is and I am NOT going to try to guess what you mean by 'clustering' in this case or try to make *ANY* sense of your 'solution with clustering / factor analysis is heavily complex and need python.' .  Perhaps you need to outsource your ongoing convoluted project!

drfg2008 wrote

I have a large number of football-matches where a home team plays against an away team (see following example).

Home / Away

Bayern München - Hertha BSC
Augsburg - Hamburger SV
1. FC Köln - Bremen
Hertha - Augsburg
Bremen - Bayern München

...

Now I try to find a way of 'clustering' those teams who ever played against each other.
(Just using the league -as an example- is not possible, since most teams also play in different leagues, some are newcommers, ... .) Each team has a unique Team ID.

My solution with clustering / factor analysis is heavily complex and need python. Isn't there an easy solution to the problem?

[Mike]

For starters, what is the unit of analysis:  a team or a match?

If a team, then you want a matrix that ist all teams by all teams
to see how often a team has played another.  I assume that
such a matrix will be somewhat sparse but you might be
able to determine whether there is a tendency for certain
teams to play other teams (that is the point of your analysis,
right?).  You might be able to "reduce" it if you want.

If a match and the first entry is home team, then you want
to see often a home team played an opponent.  This seems
to me to be a issue of nesting and whether the degree of
nesting is the same for all home teams.

But, as Daivd says, what exactly are you asking for?
I have no idea what you could be clustering/factor analyzing.

-Mike Palij
New York University
[hidden email]

On Thu, May 3, 2012 at 7:50 AM, David Marso <[hidden email]> wrote:

> Please define your question with greater specificity and try again Frank!
> I have NO idea of what your problem is and I am NOT going to try to guess
> what you mean by 'clustering' in this case or try to make *ANY* sense of
> your 'solution with clustering / factor analysis is heavily complex and need
> python.' .  Perhaps you need to outsource your ongoing convoluted project!
>
>
>
> drfg2008 wrote
>>
>> I have a large number of football-matches where a home team plays against
>> an away team (see following example).
>>
>> Home / Away
>>
>> Bayern München - Hertha BSC
>> Augsburg - Hamburger SV
>> 1. FC Köln - Bremen
>> Hertha - Augsburg
>> Bremen - Bayern München
>>
>> ...
>>
>> Now I try to find a way of 'clustering' those teams who ever played
>> against each other.
>> (Just using the league -as an example- is not possible, since most teams
>> also play in different leagues, some are newcommers, ... .) Each team has
>> a unique Team ID.
>>
>> My solution with clustering / factor analysis is heavily complex and need
>> python. Isn't there an easy solution to the problem?

=====================
To manage your subscription to SPSSX-L, send a message to
[hidden email] (not to SPSSX-L), with no body text except the
command. To leave the list, send the command
SIGNOFF SPSSX-L
For a list of commands to manage subscriptions, send the command
INFO REFCARD

[drfg2008]

Are there multiple matches between a given pair of teams?�


Yes

Roughly how many teams are there altogether?


A few hundred. However, the number may vary.

How is the input data now arranged? Are they in an SPSS system file, a spread sheet, a text file, etc.

SPSS sav

Are there a number of seasons?

Yes, but this would not have to be considered.


Why do you think you need PYTHON?

Because the number of teams vary and I used a part of my 'dummy-coding' program in Python do create a matrix.


Once you have a matrix with (all teams as Home) by (all teams as away) what else do you want to know?

I do not want to differentiate between home and away. I just want to get some sort of group membership for each team. In that group teams have played against each other (no matter if home or away) at least one time.


Are there occasions where two teams would play at a third location?

Location does not matter. Usually the teams play either at home in their own stadium or away at the home-stadium of the away team.

Are the character strings the unique IDs?� Are you very sure that there are no variations like spacing, diacriticals, etc. in the ID for a specific team?

There are unique IDs for each team. No variations.

Here is an example file (probably this file has no unique ID, but the normal files do have one)

www.frag-einen-statistiker.de/ibm_spss/Football_example.xls

[David Marso]

Frank,
  The 'example' file you referenced has *OVER* 5000 Team-ids!!!!  Many teams occur only once, many up to 50 times.  That is *MUCH* more than a few hundred.
Still nor sure what you are up to.  You should look into the MATRIX language, maybe that will be useful!
D
--

drfg2008 wrote

Are there multiple matches between a given pair of teams?�


Yes

Roughly how many teams are there altogether?


A few hundred. However, the number may vary.

How is the input data now arranged? Are they in an SPSS system file, a spread sheet, a text file, etc.

SPSS sav

Are there a number of seasons?

Yes, but this would not have to be considered.


Why do you think you need PYTHON?

Because the number of teams vary and I used a part of my 'dummy-coding' program in Python do create a matrix.


Once you have a matrix with (all teams as Home) by (all teams as away) what else do you want to know?

I do not want to differentiate between home and away. I just want to get some sort of group membership for each team. In that group teams have played against each other (no matter if home or away) at least one time.


Are there occasions where two teams would play at a third location?

Location does not matter. Usually the teams play either at home in their own stadium or away at the home-stadium of the away team.

Are the character strings the unique IDs?� Are you very sure that there are no variations like spacing, diacriticals, etc. in the ID for a specific team?

There are unique IDs for each team. No variations.

Here is an example file (probably this file has no unique ID, but the normal files do have one)

www.frag-einen-statistiker.de/ibm_spss/Football_example.xls

[drfg2008]

The 'example' file you referenced has *OVER* 5000 Team-ids!!!!

oh, such much.

As I said afraid. It's a little complex.


(usually I do not take more than a few hundreds from European leagues)

[David Marso]

Well Frank,
Maybe you need to be a little more clear about the nature of this 'complexity' and how 'a few hundred' becomes over 5000?  You expect people on this group to help you but you always seem to be playing games.  Maybe as a start get RID of all the teams which have only 1,2 or 3 games.  Really not sure what you expect people in this group to do for you.  Maybe you need to do a little thinking for yourself for a change?

drfg2008 wrote

The 'example' file you referenced has *OVER* 5000 Team-ids!!!!

oh, such much.

As I said afraid. It's a little complex.

[drfg2008]

David: I am absolutely serious about that.

This could be the solution: a matrix that is all teams by all teams
to see how often a team has played another.

If the great number of teams should be the problem, it might be possible to reduce it according to the country code.

But how to create such a Matrix?

Frank

[Jerabek Jindrich]

Hello,

id like to suggest following data restructure:

Original
ID1                                ID2
Bayern München - Hertha BSC
Augsburg - Hamburger SV

Restructured
ID1                                ID2        HOME
Bayern München - Hertha BSC 1
Hertha BSC - Bayern München 2
Augsburg - Hamburger SV 1
Hamburger SV - Augsburg 2

Every match is recorded twice, indicator 1/2 says which team played home. Change results so that win = team ID1 won.
Now Aggregate by ID1 ID2 (or crosstabs) shows all matches played between ID1 and ID2 teams, it is easy to find how many times played, number of win/loose, results home/away etc.

Regards
jindra

> ------------ Původní zpráva ------------
> Od: drfg2008 <[hidden email]>
> Předmět: Re: matching problem
> Datum: 04.5.2012 10:52:59
> ----------------------------------------
> David: I am absolutely serious about that.
>
> This could be the solution: a matrix that is all teams by all teams
> to see how often a team has played another.
>
> If the great number of teams should be the problem, it might be possible to
> reduce it according to the country code.
>
> But how to create such a Matrix?
>
> Frank
>
>
>
> -----
> Dr. Frank Gaeth
> FU-Berlin
>
> --
> View this message in context:
> http://spssx-discussion.1045642.n5.nabble.com/matching-problem-tp5682738p5685166.html
> Sent from the SPSSX Discussion mailing list archive at Nabble.com.
>
> =====================
> To manage your subscription to SPSSX-L, send a message to
> [hidden email] (not to SPSSX-L), with no body text except the
> command. To leave the list, send the command
> SIGNOFF SPSSX-L
> For a list of commands to manage subscriptions, send the command
> INFO REFCARD
>
>
>

=====================
To manage your subscription to SPSSX-L, send a message to
[hidden email] (not to SPSSX-L), with no body text except the
command. To leave the list, send the command
SIGNOFF SPSSX-L
For a list of commands to manage subscriptions, send the command
INFO REFCARD

[David Marso]

".....But how to create such a Matrix?"
-----------
GET DATA /TYPE=XLS   /FILE='C:\Users\David Marso\Desktop\Football_example.xls'.
AGGREGATE OUTFILE * / BREAK team1 team2 / N=N.
AUTORECODE team1 team2 /INTO t_id.1 t_id.2 /GROUP.
PRESERVE.
SET MXLOOPS=100000.

MATRIX.
GET DATA / FILE * / VAR n t_id.1 t_id.2 .
COMPUTE SIZEDAT=RMAX(CMAX(DATA(:,2:3))).
COMPUTE PAIRED=MAKE(SIZEDAT,SIZEDAT,0).
LOOP #=1 TO NROW(DATA).
+  COMPUTE PAIRS(DATA(#,2),DATA(#,3))=PAIRS(DATA(#,2),DATA(#,3))+DATA(#,1).
END LOOP.
COMPUTE PAIRS=PAIRS+T(PAIRS).
SAVE PAIRS / OUTFILE *.
END MATRIX.
RESTORE.

drfg2008 wrote

David: I am absolutely serious about that.

This could be the solution: a matrix that is all teams by all teams
to see how often a team has played another.

If the great number of teams should be the problem, it might be possible to reduce it according to the country code.

But how to create such a Matrix?

Frank

[David Marso]

AGGREGATE OUTFILE * / BREAK team1 team2 / N=SUM(N).

But don't get too excited Frank:
N
  Frequency Percent Valid Percent Cumulative Percent
Valid 1.00 16562 91.9 91.9 91.9
        2.00 1368 7.6 7.6 99.5
        3.00 92 .5 .5 100.0
        4.00 5 .0 .0 100.0
        5.00 1 .0 .0 100.0
        6.00 1 .0 .0 100.0
        Total 18029 100.0 100.0

The resulting matrix is so sparse as to be virtually useless for any sort of analytical purpose.
with the 5000 teams you have 25,000,000 entries.

[David Marso]

Even simpler after a little bit of thought.
POST AGGREGATE/AUTORECODE...
--------------------------------------
MATRIX.
-  GET DATA / FILE * / VAR n t_id.1 t_id.2 .
-  COMPUTE PAIRED=MAKE(MMAX(DATA),MMAX(DATA),0).
-  LOOP #=1 TO NROW(DATA).
+    COMPUTE PAIRS(DATA(#,2),DATA(#,3))= DATA(#,1).
-  END LOOP.
-  COMPUTE PAIRS=PAIRS+T(PAIRS).
-  SAVE PAIRS / OUTFILE *.
END MATRIX.
RESTORE.

David Marso wrote

".....But how to create such a Matrix?"
-----------
GET DATA /TYPE=XLS   /FILE='C:\Users\David Marso\Desktop\Football_example.xls'.
AGGREGATE OUTFILE * / BREAK team1 team2 / N=N.
AUTORECODE team1 team2 /INTO t_id.1 t_id.2 /GROUP.
PRESERVE.
SET MXLOOPS=100000.

MATRIX.
GET DATA / FILE * / VAR n t_id.1 t_id.2 .
COMPUTE SIZEDAT=RMAX(CMAX(DATA(:,2:3))).
COMPUTE PAIRED=MAKE(SIZEDAT,SIZEDAT,0).
LOOP #=1 TO NROW(DATA).
+  COMPUTE PAIRS(DATA(#,2),DATA(#,3))=PAIRS(DATA(#,2),DATA(#,3))+DATA(#,1).
END LOOP.
COMPUTE PAIRS=PAIRS+T(PAIRS).
SAVE PAIRS / OUTFILE *.
END MATRIX.
RESTORE.

drfg2008 wrote

David: I am absolutely serious about that.

This could be the solution: a matrix that is all teams by all teams
to see how often a team has played another.

If the great number of teams should be the problem, it might be possible to reduce it according to the country code.

But how to create such a Matrix?

Frank

[drfg2008]

Thanks Jerabek Jindrich


Thanks David





Error Message:

Run MATRIX procedure:
>Error encountered in source line #    61

>Error # 12492
>An attempt has been made to use previously undefined matrix (or scalar).
>Execution of this command stops.
Matrix - 'PAIRS' is undefined
>Error encountered in source line #    61

>Error # 12331
>Left hand side is undefined for subscripted COMPUTE.
>Error encountered in source line #    63

>Error # 12492
>An attempt has been made to use previously undefined matrix (or scalar).
>Execution of this command stops.
Matrix - 'PAIRS' is undefined
>Error encountered in source line #    63

>Error # 12359
>Undefined operand for transpose.
>Error encountered in source line #    64

>Error # 12492
>An attempt has been made to use previously undefined matrix (or scalar).
>Execution of this command stops.
Matrix - 'PAIRS' is undefined
>Error encountered in source line #    64

>Error # 12459
>The matrix to save is undefined.

------ END MATRIX -----

[David Marso]

Change PAIRED to PAIRS.
* I changed name after I pasted ;-(
This is something I should think you could have easily fixed by simply reading the code?
---