search

dattalab / pyhsmm-library-models

library models built on top of pyhsmm
0 stars 1 forks source link

E_step not implemented in HSMM #19

Open alexbw opened 11 years ago

alexbw commented 11 years ago 
/home/alexbw/Code/pyhsmm_library_models/pyhsmm/internals/states.pyc in E_step(self)
    811 
    812     def E_step(self):
--> 813         raise NotImplementedError
    814 
    815     ### structure-exploiting methods

NotImplementedError:
mattjj commented 11 years ago

Right, because it'd be a bit complicated to do. Why do you want it?

mattjj commented 11 years ago

But we could totally do it. I've done it for HSMMs before, and actually for our intnegbin models it may be even easier (in the usual they're-like-HMMs way).

It might be best practice to use EM for comparing held-out likelihoods, since it'll find a local maximum of training likelihood and it's the notion of "model fitting" people are familiar with.

alexbw commented 11 years ago

Do we still care about this?