I have a large sas dataset (1.5m obs, ~250 variables) that I need to split into several smaller sas datasets of equal size for batch processing. Each dataset needs to contain all the variables but only a fraction of the obs. What is the fastest way of doing this?
可以将文章内容翻译成中文,广告屏蔽插件可能会导致该功能失效(如失效,请关闭广告屏蔽插件后再试):
问题:
回答1:
You could do something like the following:
%macro splitds(inlib=,inds=,splitnum=,outid=);
proc sql noprint;
select nobs into :nobs
from sashelp.vtable
where libname=upcase("&inlib") and memname=upcase("&inds");
quit;
%put Number of observations in &inlib..&inds.: &nobs;
data %do i=1 %to &splitnum.;
&outid.&i
%end;;
set &inds.;
%do j=1 %to (&splitnum.-1);
%if &j.=1 %then %do;
if
%end;
%else %do;
else if
%end;
_n_<=((&nobs./&splitnum.)*&j.) then output &outid.&j.;
%end;
else output &outid.&splitnum.;
run;
%mend;
An example call to split MYLIB.MYDATA into 10 data sets named NEWDATA1 - NEWDATA10 would be:
%splitds(inlib=mylib,inds=mydata,splitnum=10,outid=newdata);
回答2:
Try this. I haven't tested yet, so expect a bug somewhere. You will need to edit the macro call to BATCH_PROCESS to include the names of the datasets, number of new data sets, etc.
%macro nobs (dsn);
%local nobs dsid rc;
%let nobs=0;
%let dsid = %sysfunc(open(&dsn));
%if &dsid %then %do;
%let nobs = %sysfunc(attrn(&dsid,NOBS));
%end;
%else %put Open for dataset &dsn failed - %sysfunc(sysmsg());
%let rc = %sysfunc(close(&dsid));
&nobs
%mend nobs;
%macro batch_process(dsn_in,dsn_out_prefix,number_of_dsns);
%let dsn_obs = &nobs(&dsn_in);
%let obs_per_dsn = %sysevalf(&dsn_obs / &number_of_dsns);
data
%do i = 1 %to &number_of_dsns;
&dsn_out_prefix.&i
%end;
;
set &dsn_in;
drop _count;
retain _count 0;
_count = _count + 1;
%do i = 1 %to &number_of_dsns;
if (1 + ((&i - 1) * &obs_per_dsn)) <= _count <= (&i * &obs_per_dsn) then do;
output &dsn_out_prefix.&i;
end;
%end;
run;
%mend batch_process;
%batch_process( dsn_in=DSN_NAME , dsn_out_prefix = PREFIX_ , number_of_dsns = 5 );