Re: FIXED (again): INTERP2 problem

[Paul Kienzle]

On Nov 27, 2005, at 8:49 AM, Pierre Baldensperger wrote:

> Paul Kienzle wrote:
>
> > I'm happy to support XI and YI arbitrary, however according to the 
> > docs
> > on the web, interp2(z,[1:5],[1:6]) should return a 5x6 matrix. Do you 
> > have
> > reason to believe otherwise?
>
>
> Well, yes, sorry... Which web docs are you referring to?
> Matlab behaves differently depending whether YI is a row or column 
> vector:
>
> - interp2(Z,[1:5],[1:6]) should not work
> - interp2(Z,[1:5],[1:6]') should indeed auto-grid and return a 5x6 
> matrix

Ick.  I've posted a change which does this.

I missed the significance of "...you can pass in the row and column
vectors xi and yi..." meaning that one had to be a row and the other
a column.

BTW, the docs mention that interp2 can accept an extrapolation value
as the last parameter rather than always using NaN.  You may wish
to implement this feature for your problem.

> I have no Matlab handy to check that, but that is how I understood the
> Matlab help for interp2, and that is also IMO the way it should 
> ultimately
> behave. Why would we forbid anybody to perform a 2D-to-1D interpolation
> with that function (that is: entering a 1D-list of XI and YI and 
> returning
> a 1D vector ZI)?

Agreed.  That way you can take a slice through a surface.

> In the present case, I'm concerned with limiting the field of 
> application
> of that routine, just because we want to avoid "lazy" users a 
> (_trivial_)
> preliminary call to meshgrid.
>
> So if you really want to restore this "auto-meshgridding" behaviour, 
> I'd
> suggest you trigger it the "Matlab way" (or at least the way I 
> understand
> it ;-), that is:
>
> - if XI and YI have the same size and orientation, do _not_ meshgrid
> - if XI and YI have the same orientation but different sizes, _error_
> - if XI and YI have different orientations, auto-meshgrid
>
> Of course, that may cause programs relying on the "auto-meshgridding" 
> to
> fail when they provide XI and YI with the same orientation (even with a
> "silent" failure when they have the same size).
>
> Personally, I always feel uncomfortable with routines that "think" for
> the user (same for programs, e.g. MSWord...), so I would entirely 
> remove
> that "auto-meshgridding" thing.

Yeah, treating x and x.' identically in routines that accept
vectors probably seemed like a good idea at the time, but it
does lead to a lot of ugliness.  We are stuck with it.

> Does it provide any performance or memory gain?

It could but we don't have the code in place to take
advantage of it.  If anyone were using interp2 to rescale
lots of images they might want to improve it.

> Here's a test for non-gridded XI and YI based on my initial 2D polar to
> rectangular interpolation example (simplify if necessary):

Okay, I simplified.

> Thank you for your help and explanations,
> -Pierre.

Thank you for your contributions,

        - Paul



-------------------------------------------------------------
Octave is freely available under the terms of the GNU GPL.

Octave's home on the web:  http://www.octave.org
How to fund new projects:  http://www.octave.org/funding.html
Subscription information:  http://www.octave.org/archive.html
-------------------------------------------------------------