When does the kernel fitdist function returns nan?

Dear all,

I am using the fitdist function in my code and notice that for some inputs I have a vector with nan entries.

This is the way I implement the code: pdf_x = fitdist(x,'Kernel','Kernel','epanechnikov');

So I wonder why? When does the a fitdistr kernel code return missing vallues? Can I have some feedback on this point?

thank you in advance.

Clarisha

I have the impression that it return a vector with missing values. Is that possible, pdLong = fitdist(NoAddLong,'Kernel','Kernel','epanechnikov');

[the cyclist]

Can you upload MAT file with an example of the vector x that exhibits this behavior?

[Clarisha Nijman]

Finally I was able to find one of the problematic situation: sum(f) equals zero, such that I cannot devide by zero. The sum of that probability distribution vector should equal one. So the question is now, why do I get a zero vector for pdf?

The code reads partially:

x=[0.294117647058824;0.352941176470588;0.529411764705882;0.588235294117647;0.588235294117647;0.470588235294118;0.588235294117647;0.529411764705882;0.125000000000000;0.375000000000000;0.562500000000000;0.625000000000000;0.625000000000000;0.500000000000000;0.125000000000000]

M=1000;
U = -M:1:M; %lijst met alle overgangen
NoAdd=x;

pd = fitdist(x,'Kernel','Kernel','epanechnikov');
f = pdf(pd,U);
prob=f/sum(f);