Efficiently concatenate many sas datasets

2020-07-17 15:53发布

问题:

I have over 200k small datasets with the same variables (n<1000 and usually n<100) that I want to concatenate into a master dataset. I have tried using a macro that uses a data step to just iterate through all of the new datasets and concatenate with the master with "set master new:", but this is taking a really long time. Also, if I try to run at the same time, the call execute data step says that I am out of memory on a huge server box. For reference, all of the small datasets together are just over 5 Gigs. Any suggestions would be greatly appreciated. Here is what I have so far:

%macro catDat(name, nbr) ;
    /*call in new dataset */
    data new ;
    set libin.&name ;
    run ;

    /* reorder names */
    proc sql noprint;
    create table new as 
    select var1, var2, var3
    from new;
    quit;

    %if &nbr = 1 %then %do ;
        data master;
        set new;
        run;
    %end; 
    %if &nbr > 1 %then %do ;
        data master;
        set master new ;
        run;
    %end ;
%mend;

/* concatenate datasets */
data runthis ;
set datasetNames ;
call execute('%catdat('||datasetname||','||_n_||')');
run;

Resolved: see Bob's comments below.

回答1:

Try using PROC APPEND instead of your "new" dataset; that will be much, much faster:

%macro DOIT;

proc sql noprint;
   select count(*) into : num_recs
   from datasetNames;
quit;

%do i=1 %to &num_recs;

data _null_;
  i = &i;
  set datasetNames point=i;
  call symput('ds_name',datasetname);
  stop;
run; /* UPDATE:  added this line */

%if &i = 1 %then %do;
/* Initialize MASTER with variables in the order you wish */
data master(keep=var1 var2 var3);
   retain var1 var2 var3;
   set libin.&ds_name;
   stop;
run;
%end;

proc append base=master data=libin.&ds_name(keep=var1 var2 var3);
run;

%end;

%mend DOIT;

PROC APPEND will add each dataset into your new "master" without rebuilding it each time as you are doing now. This also avoids using CALL EXECUTE, removing that memory issue you were running into (caused by generating so much code into the execution stack).