Prev: Size of the picture
Next: convert 3D matrix into 2D
From: Pinpress on 4 Feb 2010 22:57
Hi,

I need to interpolate multiple YI values using:

YI(:,i) =griddata(X,Y(:,i), XI);

Instead of performing griddata N times in a for loop, is there a better/faster way? It seems that internally "dsearchn" would be unnecessarily executed multiple times. Any input is appreciated!
From: John D'Errico on 5 Feb 2010 00:13
"Pinpress" <nospam__(a)yahoo.com> wrote in message <hkg4uh$otc$1(a)fred.mathworks.com>...
> Hi,
>
> I need to interpolate multiple YI values using:
>
> YI(:,i) =griddata(X,Y(:,i), XI);
>
> Instead of performing griddata N times in a for loop, is there a better/faster way? It seems that internally "dsearchn" would be unnecessarily executed multiple times. Any input is appreciated!

Easiest is just to do the interpolation yourself.
Something like this:

% 2-d data (independent variables)
n = 100;
X = rand(n,2);

% 10 dependent variables (I'm not very creative
% here about the actual function. Too bad.)
m = 10;
Y = rand(n,m);

% do a triangulation of your own
tri = delaunayn(X);

% some points to interpolate
Xint = rand(500,2);

[Tin,bcc] = tsearchn(X,tri,Xint);
K = ~isnan(Tin);
Yint = zeros(size(Xint,1),m);
for i = 1:3
Yint(K,:) = Yint(K,:) + bsxfun(@times,bcc(K,i),Y(tri(Tin(K),i),:));
end
Yint(~K,:) = NaN;

HTH,
John
From: Pinpress on 5 Feb 2010 11:11
Thanks, John. I ended up editing the Matlab built-in griddatan.m, and made some very minor changes and now it should work for multiple columns of Y.

"John D'Errico" <woodchips(a)rochester.rr.com> wrote in message <hkg9cu$nq3$1(a)fred.mathworks.com>...
> "Pinpress" <nospam__(a)yahoo.com> wrote in message <hkg4uh$otc$1(a)fred.mathworks.com>...
> > Hi,
> >
> > I need to interpolate multiple YI values using:
> >
> > YI(:,i) =griddata(X,Y(:,i), XI);
> >
> > Instead of performing griddata N times in a for loop, is there a better/faster way? It seems that internally "dsearchn" would be unnecessarily executed multiple times. Any input is appreciated!
>
> Easiest is just to do the interpolation yourself.
> Something like this:
>
> % 2-d data (independent variables)
> n = 100;
> X = rand(n,2);
>
> % 10 dependent variables (I'm not very creative
> % here about the actual function. Too bad.)
> m = 10;
> Y = rand(n,m);
>
> % do a triangulation of your own
> tri = delaunayn(X);
>
> % some points to interpolate
> Xint = rand(500,2);
>
> [Tin,bcc] = tsearchn(X,tri,Xint);
> K = ~isnan(Tin);
> Yint = zeros(size(Xint,1),m);
> for i = 1:3
> Yint(K,:) = Yint(K,:) + bsxfun(@times,bcc(K,i),Y(tri(Tin(K),i),:));
> end
> Yint(~K,:) = NaN;
>
> HTH,
> John
From: Loren Shure on 8 Feb 2010 08:55
In article <hkhfuo$ec4$1(a)fred.mathworks.com>, nospam__(a)yahoo.com says...
> Thanks, John. I ended up editing the Matlab built-in griddatan.m, and made some very minor changes and now it should work for multiple columns of Y.
>
You would be better off editing the file and saving it elsewhere under a
new name. That way, if you update MATLAB, you won't lose your
modifications.

--
Loren
http://blogs.mathworks.com/loren
 | 
Pages: 1
Prev: Size of the picture
Next: convert 3D matrix into 2D