Dan Bron wrote:
> yours. However, I was hoping for a solution that would complete in
> about half the time of the solutions I provided. That is, I'd like
> a 2x or 100% speedup; drag is only about 20% faster than drag0
> on my dataset.
Hmm... taking this apart:
dragR=: {~ [EMAIL PROTECTED] >./\@:* ] [EMAIL PROTECTED] 1{.0{.]
We have several full scans of the data set:
e.
-.
i.
*
>./\
{
That's a total of six operations.
Ideally, we'd like to eliminate at least two more of those, maybe
three. And, frankly, I can't think of any approach that folds
several of these together or takes advantage of special code.
That said, perhaps replacing { with } would be useful.
In other words, something like:
b=. (e. 1{.0{.]) DATA
r=. DATA {~ b # >./\(-.b)*i.#b
r (I.b)} DATA
This adds a few operations, but if DATA has fat rows
and relatively few blank rows, this should perform better.
--
Raul
----------------------------------------------------------------------
For information about J forums see http://www.jsoftware.com/forums.htm
