Vimal wrote: > Thanks David for explanation, > > Can I say that: > 1) For EM, E-step: \int p(x) log(q(x)) dx > and in > Maximization wrt some parameter, we can take p(x) as fixed ie > M : \int p(x) (log(q(x))' dx >
The earlier discussion was really just for "ordinary" maximum likelihood ... if you want to think about the EM algorithm specifically, then I think you need to extend the notation to explicitly distinguish between the observed and "missing" data, since the "expectation" step applies only to the missing part of the data. > and > > 2) Entropy = - \int p(x) log(p(x)) dx > and in > maximization, we use differentiation (of multiples rule) for some > parameter, we get M : -\int p(x)' log(p(x)) dx > > Cheers, > Vimal . . ================================================================= Instructions for joining and leaving this list, remarks about the problem of INAPPROPRIATE MESSAGES, and archives are available at: . http://jse.stat.ncsu.edu/ . =================================================================
