Re: [griddata]:__delaunay__ function

[Martin Helm]

Am 23.11.2011 09:06, schrieb Petr Gajdos:
> On Wed, Nov 23, 2011 at 08:59:12AM +0100, Thierry Laroche wrote:
>> It's better but there are still some NaN one the edges of the result
>> Matrix. I make some further checks and I will let you know.
> Remember Martin said in his previous post in this thread, that 
> resulting matrix contains some NaNs at the borders.
>
> I can only repeat that I don't have mathematical background.
>
> Petr
>
The NaNs at the border are due to the fact that for this example only a
random region of interior points are defined. Everything outside that
region is undefined and therefor NaN.
This is a feature of that test and is the correct behavior, so it is
obviously correct now.
A better test would be to use something where you know exactly at which
places NaNs have to appear (use a test without random numbers but
something with constants).
(It was of course the wrong behavior when everything was NaN which I
could also reproduce with qhull6)