Column-wise interpolation of an NxM matrix?

Hi!

I'm wondering if it is possible to interpolate column-wise along an NxM matrix.

Previous size were:

x = 1x200
Y = 1x200
xi = 1x200

New values are

x = 1x200
Y = 10x200
xi = 1x200

I have a loop that does it now, but would prefer to vectorize it as long as it is readable.

Thanks!

[Jan]

How do you "interpolate" along a singelton dimension? Could post your loop method - by editing the question, not as comment or answer?

Answer 1

[Sean de Wolski]

Sure:

doc interp1

[D G]

No, It doesn't work. It gives an error in the griddedInterpolant function.

The size of the grid point arrays should match the size of the
value arrays.

Error in interp1>LinearShiftBounds (line 255)
F = griddedInterpolant(X,V(:,1));

Error in interp1 (line 216)
       [X, V] = LinearShiftBounds(X, V, Xq);

[Sean de Wolski]

It works just fine...

y = magic(10);
x = (1:10)';
xi = linspace(1,10,50)';
interp1(x,y,xi)

[D G]

Try this:

      y = magic(10);
      x = (1:20)';
      xi = linspace(1,10,50)';
      interp1(x,y,xi, 'linear', 'extrap');

Why would this fail?

[Teja Muppirala]

That fails because y has 10 values, and your x has 20 values. But as Sean said, INTERP1 works perfectly fine. Just transpose your data first:

Yi = interp1(x',Y',xi','linear','extrap')';

Answer 2

[Doug Hull]

Why do you want to vectorize it?

If it is working, and it is apparently readable, wouldn't your time be better spent on other aspects of the code?

[D G]

The code has a lot of repetitions in our calculations.

Answer 3

[Jan]

A faster implementation than INTERP1: FEX: ScaleTime.