请高手过招

shiyiming

pf! Using hash searching is absolutely right solution to sovle this sort of problem!


[quote="hopewell":znd11mcn][code:znd11mcn]data regular;
    set sample;
    rename child=c parent=highest_parent;
run;
data out(drop=rc c);
    length c highest_parent $8;
    if _n_=1 then do;
        declare hash h(dataset:'regular');
        rc=h.defineKey('c');
        rc=h.defineData('highest_parent');
        rc=h.defineDone();
        call missing(c,highest_parent);
    end;
    set sample;
    highest_parent=parent;
    do while(h.find(key:highest_parent)=0);
    end;
run;[/code:znd11mcn][/quote:znd11mcn]

shiyiming

深度优先～～跟Oracle的start with ... connect by相似的思路。
精妙的hash,学习了。

我原来都是用广度优先，感觉数据量大的时候不受内存影响，且速度较快，loop的次数有限。看来我的思路都限制在大数据上了。

[quote="hopewell":2rw35q2y][code:2rw35q2y]data regular;
    set sample;
    rename child=c parent=highest_parent;
run;
data out(drop=rc c);
    length c highest_parent $8;
    if _n_=1 then do;
        declare hash h(dataset:'regular');
        rc=h.defineKey('c');
        rc=h.defineData('highest_parent');
        rc=h.defineDone();
        call missing(c,highest_parent);
    end;
    set sample;
    highest_parent=parent;
    do while(h.find(key:highest_parent)=0);
    end;
run;[/code:2rw35q2y][/quote:2rw35q2y]