Statistical programming, matrix languages, and more

convert 3 column dataset into matrix

Reply
Highlighted
Contributor
Posts: 32

convert 3 column dataset into matrix

[ Edited ]

I have the following input:

 

Ticker1Ticker2Edge
ABCEFG3
ABCNOM10
ABCXYZ50
EFGNOM2
EFGXYZ0
EFGQQQ2
XYZQQQ6
XYZNOM99

 

Which I would like to convert to an n x n matrix:

 

 

 ABCEFGNOMXYZQQQ
ABC 31050 
EFG3 202
NOM102 996
XYZ50099  
QQQ 26  

 

How should I do this? the above is just to give an idea. My input dataset is: 

data WORK.CLASS;
  infile datalines dsd truncover;
  input ticker1:$10. ticker2:$10. edge:32.;
datalines4;
ABC,EFG,3
ABC,NOM,10
ABC,XYZ,50
EFG,NOM,2
EFG,XYZ,0
EFG,QQQ,2
XYZ,QQQ,6
XYZ,NOM,99 ;;;;

I cut this down to 15 obs but my dataset has 500 and the ticker set are the same for ticker1 and ticker2. Thus I would want a n x n matrix. 

 

SAS Super FREQ
Posts: 3,801

Re: convert 3 column dataset into matrix

[ Edited ]
Posted in reply to ducman1611

To clarify:

1. You say "the ticker symbols are the same for ticker1 and ticker2." Does that mean that you have a long data set that has (n x n) rows and you want to convert it to an (n x n) matrix? Or do you only have the upper triangular portion of the matrix, as shown in the little 3x3 example, which has  3 observation?

2. In the second data set, do the cell counts from the Position variable? Is the Edge variable used at all?

 

In your example, Ticker1 and Ticker2 do not have the same symbols. Please provide sample data that fits the problem you want to solve (or adjust your description of the problem)

Contributor
Posts: 32

Re: convert 3 column dataset into matrix

[ Edited ]

1. What I meant was ticker1 and ticker2 contains tickers from a set of tickers. If you are familiar with networks, this dataset is an edge list for a network and each observation specifies the weighted connection between 2 nodes (vertices). 

2. Sorry there should not be a position variable. I have removed it in the question. Regarding the second question, the Edge variable is used for the values of the matrix.  So its similar to correlation between 2 tickers in the correlation matrix.

I have updated the question.

SAS Super FREQ
Posts: 3,801

Re: convert 3 column dataset into matrix

Posted in reply to ducman1611

Another question:

Look at the output of the following call to PROC FREQ. Is this what you want, except you want it in a symmetric matrix in SAS/IML?

 

proc freq data=class;
   tables ticker1*ticker2 / norow nocol nocum nopercent;
   weight Position;
run;
SAS Super FREQ
Posts: 3,801

Re: convert 3 column dataset into matrix

[ Edited ]
Posted in reply to ducman1611

The way to solve this problem depends on properties of the input data:

1. Are the data sorted by Ticker1 Ticker2 ?

2. Are all joint levels represented? That is, do you have an observation for every possible pair?

 

If the answers to (1) and (2) are both yes, then I think the following DATA step produces sample data that meets the conditions. The SAS/IML statement reads the data into the upper triangular portion of a matrix and then makes the matrix symmetric. The program uses the ROW and COL function calls, which were introduced in SAS/IML 13.1 (SAS 9.4M1).

 

 

data Have;
array c[5] $ ('A','B','C','D','E');
do i = 1 to dim(c);
   Var1 = c[i];
   do j = i+1 to dim(c);
      Var2 = c[j];
      Count = floor(10*uniform(1));
      output;
   end;
end;
keep Var1 Var2 Count;
run;

proc iml;
/* assume Have is sorted by (Var1, Var2) and that 
   all joint levels are represented (even zero counts) */
use Have;
read all var {Var1 Var2 Count};
close;

u = union(var1, var2); 
n = ncol(u);
M = j(n, n, 0);                   /* allocate matrix of zeros */
upperTriIdx = loc(row(M) < col(M)); /* upper triangular indices */
M[ upperTriIdx ] = Count; /* assign upper triangular values */
M = M + M`;               /* make symmeetric */
print M[r=u c=u];
Trusted Advisor
Posts: 1,068

Re: convert 3 column dataset into matrix

@Rick_SAS - A minor quibble.  Does not

 

              loc(row(M) > col(M));

 

generate indices for lower (rather than upper) triangle?

SAS Super FREQ
Posts: 3,801

Re: convert 3 column dataset into matrix

[ Edited ]

Thanks, Mark. I have changed the inequality to match the comment.

SAS Super FREQ
Posts: 3,801

Re: convert 3 column dataset into matrix

Posted in reply to ducman1611

Your example contains QQQ symbols, but your output does not contain those symbols. Please clarify.

Contributor
Posts: 32

Re: convert 3 column dataset into matrix

Sorry, I edited it.
Ask a Question
Discussion stats
  • 8 replies
  • 140 views
  • 0 likes
  • 3 in conversation