Minibatch ADVI with Potential for Weighted Observations

Maybe a better approach is to weighted Y first and use the weighted_Y as observed? Is the weighting parameter also part of the free parameters that you want to model?