In the em namespace a lot of classes are redundant or poorly named. I propose the following changes in order to clarify the steps and make it easier to associate them with the appropriate research papers (⇒ means rename, this means deleted).
ApproximateSupervisedEStep ⇒ FastSupervisedEStep
SecondOrderSupervisedMStep ⇒ SupervisedMStep
SupervisedMStep ⇒ FastSupervisedMStep
OnlineSupervisedMStep ⇒ OnlineFastSupervisedMStep
FastSupervisedEStep
FastUnsupervisedEStep
@paschalidoud what should we do with CorrespondenceSupervised*Step, MultinomialSupervised*Step ?
All steps shall check for convergence using the mean variational parameter change instead of the likelihood.
In the em namespace a lot of classes are redundant or poorly named. I propose the following changes in order to clarify the steps and make it easier to associate them with the appropriate research papers (⇒ means rename,
thismeans deleted).ApproximateSupervisedEStep
⇒FastSupervisedEStep
SecondOrderSupervisedMStep
⇒SupervisedMStep
SupervisedMStep
⇒FastSupervisedMStep
OnlineSupervisedMStep
⇒OnlineFastSupervisedMStep
FastSupervisedEStep
FastUnsupervisedEStep
@paschalidoud what should we do with
CorrespondenceSupervised*Step
,MultinomialSupervised*Step
?All steps shall check for convergence using the mean variational parameter change instead of the likelihood.
See #8 for likelihood estimation.