Run the EM algorithm from a given starting point. This is for ML (not REML), but it seems to be good enough to use for REML starting values.
Returns: | fe_params : 1d ndarray
cov_re : 2d ndarray
scale : float
|
---|
Notes
This uses the parameterization of the likelihood scale*I + Z'*V*Z, note that this differs from the profile likelihood used in the gradient calculations.