NaN coefficients

[tseste]

Hi,

I run many models using your library and in one run I got back "nan" coefficients.
I solved that by checking in the glm.py file if the returned values are nan

# line 219 py-glm/glm/glm.py
diff = coef - np.linalg.solve(ddbeta, dbeta)
if np.isnan(diff).any():
    break

I was wondering if it is a valid solution or we could use some different check based on tolerance?

I also checked the julia package which performs a finite check

What are your thoughts on that?
Could we integrate that change in the package?