[SciPy-user] Arrayfns in numpy?

[Travis Oliphant]

Alan Isaac wrote:

>Matthieu Brucher wrote:
>  
>
>>I'm opposed to this as well, interpolation is not linear algebra, it is
>>signal processing, and as Souheil said, why not nd-interpolation with
>>B-Splines then ?
>>    
>>
>
>I am not taking sides on this issue, but I do
>take issue with trying to settle it with this
>kind of "argument".  A rhetorical question does
>not set out the issues.  For example, treating
>a rhetorical question as a real question for a
>moment, one might respond that a simple 1-d
>interpolation would
>- find wide use among people who need nothing more
>- enhance backward compatibility
>- impose minimum maintenance requirements
>
>Again, I am not taking sides on the issue, so I
>am in no way suggesting that such points are
>decisive.  However opposition that fails to
>address such points is hardly decisive either.
>
>My point of reference would be: if it is likely
>to drain even modest developer effort away from
>core NumPy issues, then it is harmful. Otherwise
>it is harmless as long as it introduces no new
>dependencies.
>  
>
This is an impoirtant point for me.  I'm only considering adding interp 
because it's already done and just a matter of cut-and-paste.   

My take on this, is that I would rather use developer time to make scipy 
more modular and easier to install (this usually translates to binary 
builds for multiple platforms).   That way you could grab just the 
modules of interest.  Ideally, grabbing these modules is a matter of 
using setuptools (or some other package management concept) to do it 
automatically as needed.

-Travis