筛选观测值

是 否 +2 论坛币

k人 参与回答

经管之家送您一份

应届毕业生专属福利!

求职就业群

赵安豆老师微信：zhaoandou666

经管之家联合CDA

送您一个全额奖学金名额~ !

立即领取

感谢您参与论坛问题回答

经管之家送您两个论坛币！

+2 论坛币

a        b        c
173.2465        5.641        28375
173.2908        5.645        7161
173.1294        5.642        767098
197.2151        9.998        51203
197.2644        9.999        13350
197.0907        9.999        1193382
197.3505        10.001        6035
173.8311        6.39        6832
197.899        10.34        7456


上述数据中，任意两行观测的变量a和b相减，若-0.5=<变量a的差值=<0.5, -0.3=<变量b的差值=<0.3，则将这些观测输出。
谢谢各位了！

扫码加我 拉你入群

请注明：姓名-公司-职位

以便审核进群资格，未注明则拒绝

分享0 收藏1 回帖

关键词：筛选观测 观测值 51203

任意两行需要全连接下，即
proc sql noprint;
select a.*,b.a as a_b,b.b as b_b,b.c as c_b
from table as a,table as b;
quit;

然后用data步算下a和a_b，b和b_b，c和c_b的差值，再筛选就好了

我这是麻烦的办法，可能还有好办法，等楼下的吧

任意两行需要全连接下，即
proc sql noprint;
select a.*,b.a as a_b,b.b as b_b,b.c as c_b
from table as a,table as b;
quit;

然后用data步算下a和a_b，b和b_b，c和c_b的差值，再筛选就好了

我这是麻烦的办法，可能还有好办法，等楼下的吧

PROC SQL;
SELECT _1.* FROM HAVE AS _1, HAVE AS _2 WHERE _1.A-_2.A BETWEEN -.5 AND .5 OR/AND _1.B-_2.B BETWEEN -.3 AND .3;

jingju11 发表于 2013-5-31 10:05
PROC SQL;
SELECT _1.* FROM HAVE AS _1, HAVE AS _2 WHERE _1.A-_2.A BETWEEN -.5 AND .5 OR/AND _1.B-_2 ...

嗯，这个看起来简单多了
但还是要靠全连接吧
表大的话还是比较费时间的

谢谢！我的表确实很大，有什么更精简的方法吗？

费时间在这里应该不是最大的问题。因为这种程序通常不需要反复执行。问题是是否有足够的内存。从内存使用的方面来看，ＳＱＬ在ＳＡＳ里是最好的选择之一。我的观察是ＳＱＬ在数据变大时，所占内存相对较为稳定，而其他的过程则不然。个人观点。京剧
我对ＳＱＬ和ＦＲＥＱ的运算效率有个比对。
http://blog.sina.com.cn/s/blog_a3a926360101d69b.html

/* try this */

data aaa;
input a  b  c;
datalines;
173.2465  5.641 28375
173.2908  5.645 7161
173.1294  5.642 767098
197.2151  9.998 51203
197.2644  9.999 13350
197.0907  9.999 1193382
197.3505 10.001 6035
173.8311  6.39  6832
197.899  10.34  7456
;run;

proc sql;
   select count(*) into :Nobs    from aaa;
   select a into :a separated by ' ' from aaa;
   select b into :b separated by ' ' from aaa;
quit;

data bbb;
   array aa{&nobs} _temporary_ (&a);
   array bb{&nobs} _temporary_ (&b);
   set aaa end=last;
     do i = 1 to &Nobs;
        a1 = aa(i);
        b1 = bb(i);
      if  -0.5 <= a-a1 <= 0.5 or
          -0.3 <= b-b1 <= 0.3 then output;
     end;
run;

谢谢楼上的大侠但是我运行不了，我的QQ2393695206，能否加你？

if either column A or B has a big deviation, then grouping the observations by their distance to the origin before selecting will have a clear advantage over Cartesian product. For example,

1. data hell;
   
2.         do n=1 to 1000000;
   
3.                 a = 1000*rannor(12345);
   
4.                 b = 10*rannor(12345);
   
5.                 c = rannor(12345);
   
6.                 zone = floor(sqrt(a*a+b*b)/.6);*sqrt(0.5^2+0.3^2)<0.6;
   
7.                 output;
   
8.         end;
   
9.         
   
10. proc freq data=hell;
    
11.         tables zone/nocum nopct;
    
12. run;
    
13. 
    
14. 
    
15. proc sql;
    
16. *create table pairs as
    
17.         select A.n as n_a, B.n as n_b, A.a as a_a, B.a as a_b, A.b as b_a, B.b as b_b, A.c as c_a, B.c as c_b
    
18.                 from hell a, hell b
    
19.                 where abs(A.a-B.a)<=0.5 and abs(A.b-B.b)<=0.3 and a.n < b.n;
    
20. quit;
    
21. proc sql;
    
22. create table pairs1 as
    
23.         select A.n as n_a, B.n as n_b, A.a as a_a, B.a as a_b, A.b as b_a, B.b as b_b, A.c as c_a, B.c as c_b
    
24.                 from hell A join hell B on A.zone=B.zone
    
25.                 where abs(A.a-B.a)<=0.5 and abs(A.b-B.b)<=0.3 and a.n < b.n
    
26.         outer union corr
    
27.         select A.n as n_a, B.n as n_b, A.a as a_a, B.a as a_b, A.b as b_a, B.b as b_b, A.c as c_a, B.c as c_b
    
28.                 from hell A join hell B on A.zone+1=B.zone
    
29.                 where abs(A.a-B.a)<=0.5 and abs(A.b-B.b)<=0.3;
    
30. quit;

复制代码

However if you data looks like the following, then I have to say that may the Lord bless you!

1. data hell;
   
2.         do n=1 to 1000000;
   
3.                 a = rannor(12345);
   
4.                 b = ranuni(12345);
   
5.                 c = rannor(12345);
   
6.                 zone = floor(sqrt(a*a+b*b)/.6);*sqrt(0.5^2+0.3^2)<0.6;
   
7.                 output;
   
8.         end;

复制代码