Generalize the concept of what is happening in the Brainiac model where we have
Qprior = h * someMatrix
and a gaussian likelihood with
y \sim N(eta, 1-h)
i.e. the hyperparameter h is shared between both and this is not standardly supported. Add Gaussian likelihood where this concept is generalized. I'm not yet exactly sure how to best specify this.
The submodel dictionary will take priors for its hyperparameters and the gaussian likelihood as well. so somehow there would have to be a way to link them (and define them not to be the same but allow for e.g. 1-h).
This is currently handled for the Brainiac using a lot of if(Brainiac) ... statements in the model class which is not very clean.
Generalize the concept of what is happening in the Brainiac model where we have
Qprior = h * someMatrix
and a gaussian likelihood with
y \sim N(eta, 1-h)
i.e. the hyperparameter h is shared between both and this is not standardly supported. Add Gaussian likelihood where this concept is generalized. I'm not yet exactly sure how to best specify this.
The submodel dictionary will take priors for its hyperparameters and the gaussian likelihood as well. so somehow there would have to be a way to link them (and define them not to be the same but allow for e.g. 1-h).
This is currently handled for the Brainiac using a lot of if(Brainiac) ... statements in the model class which is not very clean.