用宏分割数据集

bobguy 发表于 2013-8-13 10:46
The solution is very inefficient if many data sets needs to be generated.

I give a hint here,
...

Thanks. I didn't consider it. Maybe the following is ok.

proc sql noprint;
    create table vlist as
        select distinct var1 from raw
        order by var1;
quit;

data _null_;
        set vlist end=last;
        call symputx("v"||cats(_n_),var1);
        if last then call symput("N",cats(_n_));
run;

%macro split;
data %do i= 1 %to &N.;
         data_&&v&i
         %end;;
        set raw;
        select (var1);
                %do i= 1 %to &N.;
                when ("&&v&i") output data_&&v&i;
                %end;
        end;
run;
%mend;

%split

Eternal0601 发表于 2013-8-13 21:08
Thanks. I didn't consider it. Maybe the following is ok.

proc sql noprint;

That is the right way. Yoc can improve it by combining 1)+2) as

data raw;
  var1='a'; output; var1='a'; output;
  var1='b'; output; var1='c'; output;
run;

proc sql noprint;
   
        select distinct var1
        into: var1 - :var30000
        from raw
        order by var1;
quit;

%put >>>&sqlobs<<<;
%put >>>&var1 &var2 &&var&sqlobs<<<;

学习了一下程序，很有收获~

bobguy 发表于 2013-8-14 05:23
That is the right way. Yoc can improve it by combining 1)+2) as

data raw;

Hi bob,

I have a question need to require you.

As the code show that:

proc sql noprint;
        select distinct var1
        into: var1 - :var30000
        from raw
        order by var1;
quit;

But mostly, I need to know which variable means what value. For example:

%do n=1 %to &N;
  %put &&var&n..;
%end;

So, how to write _N_ meanwhile in the SQL step as above in case I have to order the dataset raw with another sequence.

Tigflanker 发表于 2013-8-28 13:57
Hi bob,

I have a question need to require you.

In this case,

  _N_=&sqlobs

to order the dataset raw with another sequence= order by YOUR VARIABLE

Eternal0601 发表于 2013-8-13 21:08
Thanks. I didn't consider it. Maybe the following is ok.

proc sql noprint;

真是佩服作者的sas技巧：一个人给出两种截然不同但是都很高超的sas程序。最后的取舍应该根据程序的效率。我运行了两个程序，基于一个数据集1，000，000记录，变量取值大概1000个，取值均匀的随机分布。结果是：where-call execute =140s, select/when-macro=200s.且不说如果数据sorted/indexed，where- 更有效率。也就是，你的第一个程序不但写得好，而且效率也好。
佩服。
京剧

jingju11 发表于 2013-9-6 06:55
真是佩服作者的sas技巧：一个人给出两种截然不同但是都很高超的sas程序。最后的取舍应该根据程序的效率。 ...

Thanks for leting me know that. And your blog is great.

hsaotsud程序比较漂亮。

中间一段proc sql有必要吗？