griddedInterpolant (via e.g. interp1) bug with NaNs?

I hope I've done due diligence searching for the source of this behavior; I haven't seen this question asked or answered.

Is interp1 with method 'linear' correct in its sensitivity to NaN order when basis and interpolant points are equal?

The plot generated by the following code demonstrates that, although the value for x=2 is present in the source data y, it is returned as NaN by the interpolant for xi=2.

x = 1:5;
y = [2,1,NaN,1,2];
xi = 1:0.5:5;
yi = interp1(x,y,xi,'linear');

plot(x,y,'.-k');hold on;
plot(xi,yi,'s-r'); hold off

Is this the expected behavior? In looking at the interp1 code, it seems that the fault, if it exists, may lie within griddedInterpolant.

[Walter Roberson]

Behavior with a NaN is not defined in the documentation.

Answer 1

[dominik]

Hey, I have the same problem: I define a griddedinterpolant IP over a grid G and the values V. whenever I evaluate IP at a gridpoint x that has an adjacent gridpoint y for which V(y) = NaN, I get IP(x)=NaN, even though V(x) is not NaN. Can that be avoided? I Use matlab 2012a.

Answer 2

[Trevor Harris]

Seems as if this answer still isn't solved. I'm in 2017b and am having the same problem. As you can see by the attached screenshot, the value for 300 does indeed exist, but when I reference it, I get a NaN. Any idea why this behaviour exists?

Trevor

Answer 3

[Vladimir Kazei]

You can use fillmissing(y, 'linear') since 2016b version

https://www.mathworks.com/help/matlab/ref/fillmissing.html