Re: [OctDev] corrected version of interp2.m

[Paul K. <pki...@us...>]

On Mar 3, 2005, at 3:21 AM, Thomas Weber wrote:

> Paul,
>> Please let me know if you have any problems with it.
> Actually yes, I have problems with it. They are not due to your (or my)
> changes, but I consider this function flawed.
>
> I'm currently in the following situation:
> x, y:	vectors of length ~200
> Z: 	matrix of 200^2
> xi, yi: vectors of length ~16000
>
> Now, interp2 tries a meshgrid(xi,yi), which results in 'out of memory'.
> However, I'm only interested in the interpolated values at
> [xi(i), yi(i)], not at the complete grid one can construct with these
> vectors.
>
> Therefore, I'm thinking about changing the function to return an
> interpolated  vector at the positions of (xi,yi) if they are vectors; 
> if
> the user wants a complete grid, he can still do an
> [XI, YI] = meshgrid(xi,yi) before and hand the matrices to interp2.
>
> Please note that this will break compatibility with Matlab (which shows
> the same 'out of memory' behaviour).
>
> Any suggestions?
>

Please don't break compatibility.

You can make a new function which doesn't do an implicit
meshgrid on the xi,yi arguments and call it from interp2.

I don't really like that solution, but I can't think of a
nice way to mark which form of xi,yi you want in the parameter
list.

- Paul