弱问如何取某sas数据集的前a%个和后a%个的观测值？

pobel 发表于 2011-8-17 15:00
另一种方法：
data _null_;
        if 0 then set sashelp.class nobs=nobs;

data objet;
   do obsnum=91 to 3731 by 91;
      set origine point=obsnum;
      output;
   end;
   stop;
run;

Does it work?

zhangzachary 发表于 2011-8-17 22:19
data objet;
   do obsnum=91 to 3731 by 91;
      set origine point=obsnum;

It may not work as he/she wanted.
This data step creates only one dataset: OBJET. It doesn't split the dataset ORIGINE into several datasets.

But I think it's a good one when we want to subset a dataset.

pobel 发表于 2011-8-18 06:54
It may not work as he/she wanted.
This data step creates only one dataset: OBJET. It doesn't spli ...

额，是我看错问题了

楼上和楼楼上，请问下
一个data步遍历一次输出到n张表快
还是分n个data步用point输出快啊？

soporaeternus 发表于 2011-8-18 16:16
楼上和楼楼上，请问下
一个data步遍历一次输出到n张表快
还是分n个data步用point输出快啊？

这个问题不懂，应该和数据集大小，和要分成多少块有关系。

做了个小测试：

740  data test;
741       do i=1 to 1000000;
742          varlue=ranuni(0);
743          output;
744       end;
745  run;
746
747  %macro split(origin=test, dsnum=, times=100);
748
749  %put %str( );
750  %put %str(If we want to split a dataset into &dsnum datasets and do it &times times:);
751    *** Split in one data step;
752    %let totaltime=0;
753    %do j=1 %to ×
754          %let timestart=%sysfunc(time());
755          data %do i=1 %to &dsnum;
756                  test&i
757               %end;
758              ;
759              set &origin nobs=n;
760              obs=ceil(n/&dsnum);
761              %do i=1 %to &dsnum;
762                  %if &i ne 1 %then %do;   else   %end;
763                  if %eval(&i-1)*obs<_n_<=&i*obs then output test&i;
764              %end;
765          run;
766          %let timestop=%sysfunc(time());
767          %let timeused=%sysevalf(&timestop-&timestart);
768          %let totaltime=%sysevalf(&totaltime+&timeused);
769     %end;
770
771      %let totaltime=%sysfunc(round(&totaltime,0.01));
772      %put %str(   when spliting in one data step, we need: &totaltime seconds.);
773
774
775    *** Split in separate data steps;
776    %let totaltime=0;
777
778    %do j=1 %to &times;
779        %do i=1 %to &dsnum;
780                data _null_;
781                    if 0 then set &origin nobs=n;
782                    obs=ceil(n/&dsnum);
783                    call symputx("start",(&i-1)*obs+1);
784                    call symputx("end",min(&i*obs,n));
785                run;
786
787             %let timestart=%sysfunc(time());
788              data test&i;
789                 do obs=&start to &end;
790                     set &origin nobs=n point=obs;
791                     output;
792                 end;
793                 stop;
794              run;
795              %let timestop=%sysfunc(time());
796              %let timeused=%sysevalf(&timestop-&timestart);
797              %let totaltime=%sysevalf(&totaltime+&timeused);
798         %end;
799     %end;
800
801      %let totaltime=%sysfunc(round(&totaltime,0.01));
802      %put %str(   when spliting in separate data steps, we need: &totaltime seconds.);
803      %put %str( );
804      %put %Str( );
805  %mend;
806
807  options nonotes;
808  %split(dsnum=10,times=10)

If we want to split a dataset into 10 datasets and do it 10 times:
   when spliting in one data step, we need: 4.11 seconds.
   when spliting in separate data steps, we need: 6.31 seconds.


809
810
811  %split(dsnum=10,times=100)

If we want to split a dataset into 10 datasets and do it 100 times:
   when spliting in one data step, we need: 41.36 seconds.
   when spliting in separate data steps, we need: 54.25 seconds.


812
813  %split(dsnum=100,times=10)

If we want to split a dataset into 100 datasets and do it 10 times:
   when spliting in one data step, we need: 28.47 seconds.
   when spliting in separate data steps, we need: 14.59 seconds.


814  %split(dsnum=1000,times=10)

If we want to split a dataset into 1000 datasets and do it 10 times:
   when spliting in one data step, we need: 371.59 seconds.
   when spliting in separate data steps, we need: 104.97 seconds.


815  %split(dsnum=4,times=1000)

If we want to split a dataset into 4 datasets and do it 1000 times:
   when spliting in one data step, we need: 349.94 seconds.
   when spliting in separate data steps, we need: 492.72 seconds.

pobel 发表于 2010-8-16 17:25
data a;
    set sashelp.class nobs=nobs;
        if _N_

请问这里的sashelp.class 是数据集的名字吗？