Date:         Mon, 14 Mar 2005 17:13:19 -0500
Reply-To:     "Howard Schreier <hs AT dc-sug DOT org>" <nospam@HOWLES.COM>
Sender:       "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From:         "Howard Schreier <hs AT dc-sug DOT org>" <nospam@HOWLES.COM>
Subject:      Re: Lookahead [was: output to two different datasets based on a
              pattern]

On Sat, 12 Mar 2005 00:15:16 -0800, Jack Hamilton <jfh@STANFORDALUMNI.ORG>
wrote:

>Here's a version which doesn't use BY.  I think the code is simpler, but it
>might not be as flexible as yours:
>
>=====
>data lookahead1 (drop=nextk);
>
>   set demo end=endmain;
>
>   if not endnext then
>      set demo (firstobs=2
>                rename=(k=nextk val=nextval))

I think I would always include KEEP= coding here to avoid introducing
incorrect values for other variables.

>          end=endnext;
>
>   if (k ne nextk) or endmain then

If there are multiple variables defining groups, each one must be tested,
with "OR" links.

>      nextval = .;
>
>run;
>=====
>
>Results are the same using your sample data.  Neither solution actually
>answers the original question, but that's life on the internet.
>
>
>> -----Original Message-----
>> From: SAS(r) Discussion [mailto:SAS-L@LISTSERV.UGA.EDU] On
>> Behalf Of Howard Schreier <hs AT dc-sug DOT org>
>> Sent: Friday, March 11, 2005 6:19 am
>> To: SAS-L@LISTSERV.UGA.EDU
>> Subject: [SAS-L] Lookahead [was: output to two different
>> datasets based on a pattern]
>>
>> As Jack implies, the technique using FIRSTOBS=2 gets messy
>> when there are
>> BY groups.
>>
>> Here is a model for a lookahead technique which works within
>> BY groups.
>>
>>    data demo;
>>    input k $ val @@ ;
>>    cards;
>>    a 11 a 12 a 13
>>    b 21
>>    c 31 c 32
>>    d 41 d 42 d 43 d 44
>>    ;
>>
>>    data lookahead (drop=i);
>>    set demo;
>>    by k;
>>    if               (first.k or  not last.k) then
>>     do i = 1 to 1 + (first.k and not last.k);
>>       set demo(keep=val rename=(val=nextval) );
>>       end;
>>    if last.k then do;
>>       nextval = .;
>>       end;
>>    run;
>>
>> The second SET statement does the lookahead. It must read two
>> observations
>> at the start of a group but none at the end. The case where
>> there is only
>> one observation in the group must also be handled. So there is one
>> expression to decide whether the SET executes at all, and another
>> expression to determine how many times.
>>
>> Result:
>>
>>    Obs    k    val    nextval
>>
>>      1    a     11       12
>>      2    a     12       13
>>      3    a     13        .
>>      4    b     21        .
>>      5    c     31       32
>>      6    c     32        .
>>      7    d     41       42
>>      8    d     42       43
>>      9    d     43       44
>>     10    d     44        .
>>
>> It should be possible to generalize this for deeper
>> lookaheads, but it's
>> not trivial. LAST.K is not useful, because it's necessary to
>> detect second-
>> from-last etc. I think it has to be done using a Double DoW
>> so that the
>> size of each group can be determined and used in the controlling
>> expressions.
[snip]