On Tue, Oct 28, 2008 at 12:23 AM, Ian S. Worthington
<[EMAIL PROTECTED]> wrote:

> I reckon I can do it with LOOKUP with a DROP n in the "add to" path.  Is there
> an easier way?

The problem with this one (and Sir Mike's suggestion) is that the
"take" or "drop" goes over the entire stream and not the run of
records with the same key. At that point a sipping pipeline can do it
because you could sip "whilelabel.." and then take the n records from
that run. But overhead of that approach is a bit higher, so it's
better to put it in a straight pipe.
I also like "juxtapose count" in this context. If the run is not
identified by a common key but by a separator or so, then you can hold
the start of a run in the juxtapose and the rest will be tagged with
the sequence number within the run. I did something like that here:
http://rvdheij.wordpress.com/2007/11/13/finding-unique-observations/

Rob

Reply via email to