Great alternate approaches ...

Symmetrically padding appears to me a most valid direction, and fairly
simple to implement. Difficult to apply "n/a" in a plot but zero values
would be at least comprehensible to the viewer.

Bo's idea of shorter edge samples is also a nice touch.

Thanks all;

David

On Thu, 2010-06-10 at 11:18 -0400, Raul Miller wrote:
> Here's a moving average implementation which
> uses shorter samples near the edges:
> 
> mean=: +/ % #
> noNaN=: #~ 1 - 128!:5
> avg=: m...@nonan
> nanPad=: ] (],[,]) _. #~ <:@[
> movingAvg=: [ avg\  nanPad
> 
> Example use:
> 
>   5 movingAvg i.10
> 
> Note that this particular result is longer than the right,
> but a different amount of padding would change that.
> 
> For example
> 
> nanPad2=:  <....@-:@[ |. ] ,~ _. #~ <:@[
> movingAvg2=:[ avg\  nanPad2
> 
> FYI,
> 


----------------------------------------------------------------------
For information about J forums see http://www.jsoftware.com/forums.htm

Reply via email to