Linear regression should be fast enough. Here is a simple case with 1,000,000 obs one variable. It takes 0.34 cpu time on a window 7/I5 laptop.
244 data t1;
245 do i=1 to 1e6;
246 x=rannor(123);
247 Y=1+1*X+rannor(123);
248 OUTPUT;
249 END;
250 RUN;
NOTE: The data set WORK.T1 has 1000000 observations and 3 variables.
NOTE: DATA statement used (Total process time):
real time 0.35 seconds
cpu time 0.34 seconds
251
252 PROC REG DATA=T1;
253 MODEL Y=X;
254 RUN;
255 QUIT;
NOTE: PROCEDURE REG used (Total process time):
real time 0.74 seconds
cpu time 0.15 seconds
|