alternatives to coding multiple data steps within macro loop to improve efficiency?

[RobF]

I'm working on a SAS program that runs a coordinate descent algorithm which repetitively processes & updates dataset variables inside a macro loop:

%macro test(dataset=, numvars=, numiter=, lambda=, alpha=);

%do i=1 %to &numiter;

     %do j=1 %to &numvars;

           data &dataset (keep=y x1-x&numvars);

           set &dataset end=end_data;

                array x[&numvars] x1-x&numvars;

                                             < ... Code for variable manipulations & calculations ...>

           run;

     %end;

%end;

%mend;

%test(dataset=&dataset, numvars=10, numiter=100, lambda=.1, alpha=1)

I wrote this code out of convenience, realizing that putting a data step inside the macro loop isn't terribly efficient since opening and closing the same dataset eats up CPU resources.

Is there a better way to code this algorithm that will save significant time when running?

[Astounding · Accepted solution]

In part, that depends on the calculations.  It is extremely likely you could recode along these lines:

data &dataset;

set &dataset end=end_data;

array x ...;

do _i_=1 to &numiter;

do _j_=1 to &numvars;

** calculate, possibly exactly as before;

end;

end;

run;