topic Re: Replace missing values with 0 in SAS/IML Software and Matrix Computations

Replace missing values with 0

xtc283x — Tue, 20 Oct 2015 14:25:02 GMT

I have a matrix G that contains 2 missing values:

G
0.0909091	0.2	1	0.6666667	0.2857143
0.1428571	0.2857143	0.2	.	0.25
-2	0.1538462	-2	-0.181818	0.6666667
-0.666667	-1	0.1538462	-1	0.5
0.2222222	0.25	2	-1	0.5
0.2222222	2	-0.285714	0.4	-2
1	-0.333333	0.3333333	0.4	2
0.5	0.4	.	-1	1

The CALL CHANGE function only works on text fields. And I don't understand the explanations for the REPLACE function.

What is the logic in SAS/IML to replace those two missing values with 0?

Re: Replace missing values with 0

Rick_SAS — Tue, 20 Oct 2015 14:46:31 GMT

use the LOC function to find the entries that are missing. Then use assignment to replace those values:

idx = loc( G = . );
G[ idx ] = 0;

Re: Replace missing values with 0

xtc283x — Wed, 21 Oct 2015 12:23:00 GMT

Thank you. Using LOC() and idx() worked. Using the same matrix G, I'm getting a new error in applying it to the norm() module. Here's the line of code:

GFrob = norm(G, "Frobenius");

And the error is "ERROR: Invocation of unresolved module NORM."

Based on a post in The Do Loop, this code should work!

http://blogs.sas.com/content/iml/2014/04/07/vector-and-matrix-norms-in-sas.html

Re: Replace missing values with 0

Rick_SAS — Wed, 21 Oct 2015 12:29:04 GMT

The first sentence of the blog says that you need SAS/IML 12.1, which was shipped in August 2012 as part of SAS 9.3m2.

What do you get when you run the following statement?

%put SYSVLONG = &SYSVLONG;

Re: Replace missing values with 0

xtc283x — Wed, 21 Oct 2015 12:37:39 GMT

SYSVLONG = 9.03.01M1P

Re: Replace missing values with 0

Rick_SAS — Wed, 21 Oct 2015 12:57:51 GMT

That answers the question: You are running SAS 9.3.

Now what to do about it? There have been five releases of SAS since SAS 9.3, so you might consider upgrading.

In the meantime, most norms are trivial to program. Here's a SAS/IML function that computes the Frobenius, L1, L2, and "L infinity" norms for matrices:

start matnorm(x, method);
   s = substr(upcase(method), 1, 4);
   isValid = (s="FROB" | s="L1" | s="LINF" | s="L2");
   if ^isValid then return(.);
   
   if (s="FROB") then do;
      return( sqrt(x[##]) );
   end;
   else if (s="L1") then do;
      return( max(x[+,]) );
   end;
   else if (s="LINF") then do;
      return( max(x[,+]) );
   end;
   else if (s="L2") then do;
      v = eigval(x`*x);
      return( sqrt(v[1]) );
   end;
finish;


/* Test the module on the doc example */

x={1 2, 3 4};
mn1 = matnorm(x, "L1");
mnF = matnorm(x, "Frobenius");
mnInf = matnorm(x, "LInf");
print mn1 mnF mnInf;

Re: Replace missing values with 0

xtc283x — Wed, 21 Oct 2015 13:36:35 GMT

That worked. Thank you again.