How to put the dot product of two vector macro variables into data step equations?

[YutongHu]

&effects and &estimate are regression results stored in vector macro variables.

As shown in the pictures above, &effects is the vector variable with the names of effects.

&estimate are the vector of coefficients.

say &effects is "a b c d e f g".

&estimate is "1 2 3 4 5 6 7"

How can I use these two macro variables to create a data step equation that equals to the dot product of the two vectors?

Here's the code that I tried, but it did not produce the correct result.

%do i=1 %to %sysfunc(countw(&effects));
%let effects&i = %scan(&effects, &i, %str( ));
%end;

%do i=1 %to %sysfunc(countw(&estimate));
%let estimate&i = %scan(&estimate, &i, %str( ));
%end;

%let total1=&effects1*estimate1;
%let largeN=%sysfunc(countw(&effects));

%do i=2 %to %sysfunc(countw(&effects));
%let total&i = total%eval(&i-1).+&effects&i*&estimate&i;
%end;

%put &total&largeN;      [<-this should be the yhat definition, it is an expression like a*1 + b*2 + c*3 + d*4 + e*5 + f*6 + g*7]

data option2;
set opion1;
yhat=&total&largeN;   [<-this equation should be the same as "yhat=a*1 + b*2 + c*3 + d*4 + e*5 + f*6 + g*7"]
run;

Can anyone help me to solve this problem? Thank you so much!

[Astounding]

Well, you could cut out a few pieces in this way:

data option2;

set option1;

yhat = 

%do i=1 %to %sysfunc(countw(&effects));

   %scan(&effects, &i) * %scan(&estimate, &i)

   %if &i < %sysfunc(countw(&effects)) %then + ;

%end;

;

run;

This does have to be inside a macro definition, to permit %if and %do.

[YutongHu]

I tried the code, but it still did not work. Thank you anyway.

[Reeza]

I would say, don’t create this problem in the first place. 

There are multiple ways to score new data, this doesn’t seem like a good approach. 

Assuming that’s what you’re doing of course. 

[YutongHu]

The problem was that I predicted the model using sample A, but I also want to use the model to test the accuracy of the model on sample B. However, I only know the step that directly produces predicted yhat for sample A. That is why I need to create a formula by macro variables. 

[ballardw]

Please explain what the overall purpose of the generated data step code.

If your macro variable lists came from a data set it may be that using that data set and some data transformation in a data step will work in a much cleaner manner.

If you are going to use multiple identical calls such as

%sysfunc(countw(&effects))

perhaps you would be better off calling that once and assigning the value to another macro variable. Then you could make the code a little cleaner:

   %do I = 1 to &EffectCount;

[jint83]

Speaking on behalf of the poster. These macros, did not come from a dataset, so the question cannot be avoided, as the macros definitions come from SAS GLM select outputs:

https://support.sas.com/documentation/cdl/en/statug/63347/HTML/default/viewer.htm#statug_glmselect_s...

I don't know the answer to poster's question, but I do think the question is an important one that begs to be addressed within the framework of the poster's question. 

[YutongHu]

Hi, the problem was that I predicted the model from sample A.

Let's say the model is  yhat=a + b*1 + c*2

I want to calculate yhat not only for the observations in sample A, but also for the observations in sample B.

Is there a simpler way that I can use to avoid using macro variables?

Thank you!

[Reeza]

See the various methods here. 

https://blogs.sas.com/content/iml/2014/02/19/scoring-a-regression-model-in-sas.html

[ballardw · Accepted solution]

If @jint83 is correct and this is coming from proc glmselect then help us out by running the model and then posting the result of

%put _user_;

This will give us the actual names and contents of the macro variables.

Or try adding a line something like this to your code (before the run statement)

   ods output parameterestimates= work.parms;

Which will create a data set with the effect and the parameters.

You don't mention if by groups are involved but for a single model

data _null_;
   set work.parms end=eof;
   length longstr $ 500;
   retain longstr;
   longstr = catx(' + ',longstr, catx('*',effect,estimate));
   call symputx('Formula',longstr);
run;

%put &formula;

appears to do what you want.