# Why is scatteredInterpolant slower than TriScatteredInterp ?

41 views (last 30 days)
Natan on 25 Sep 2013
Commented: Matt J on 27 Sep 2013
I've written a code that uses TriScatteredInterp, but I read in Matlab's documentation that this will not be supported in future release and that I should instead use scatteredInterpolant. So I did, and found to be twice slower for a 512 by 512 matrix. The only difference in my code was just using:
F = scatteredInterpolant(double(x'),double(y'),double(z'),'natural');
F = TriScatteredInterp(double(x'),double(y'),double(z'),'natural');
Why is that and what can I do to make it faster?

Matt J on 26 Sep 2013
Your calls to double() and ctranspose() could have affected the speed comparison.
Natan on 26 Sep 2013
I doubt it. I use the exact same input for TriScatteredInterp.
Matt J on 27 Sep 2013
Even if the construction time does matter to you, I cannot reproduce the slow behavior. I even see slightly faster behavior in scatteredInterpolant for a 512^2 data,
x = rand(512)*4-2;
y = rand(512)*4-2;
z = x.*exp(-x.^2-y.^2);
tic;
obj1 = TriScatteredInterp(x(:),y(:),z(:));
toc
%Elapsed time is 1.475620 seconds.
tic
obj2 = scatteredInterpolant(x(:),y(:),z(:));
toc
%Elapsed time is 1.303282 seconds.

Matt J on 26 Sep 2013
Edited: Matt J on 26 Sep 2013
You're not supposed to care about the time to construct F. That's only a one-time set-up computation. You're supposed to care about the speed of the actual interpolation operations, which presumably you would be doing repeatedly and on many more points.
In my tests below, I do find scatteredInterpolant to be a bit slower, but certainly not by a factor of 2, and the difference varies with the data sizes. There might be robustness improvements that explain the slower interpolation speed of the new version (better error checking, for example...)
x = rand(100,1)*4-2;
y = rand(100,1)*4-2;
z = x.*exp(-x.^2-y.^2);
obj1 = TriScatteredInterp(x,y,z);
obj2 = scatteredInterpolant(x,y,z);
ti = linspace(-2,2,2e3);
[qx,qy] = meshgrid(ti,ti);
tic;
qz1 = obj1(qx,qy);
toc;
%Elapsed time is 4.324792 seconds.
tic;
qz2 = obj2(qx,qy);
toc
%Elapsed time is 4.990118 seconds.

### Community Treasure Hunt

Find the treasures in MATLAB Central and discover how the community can help you!

Start Hunting!

Translated by