Albert

Hi @erlebach! I am not sure I understand your model. Can you provide the equations for your generative model? Or you can try to explain what are you trying to...

Sorry, I noticed the new topic on `Discussion` earlier than I read your answer @erlebach. As for recognition distribution `q`. This term is also referred to as instrumental distribution. This...

I am not sure I've understood your question. What do you mean by irrelevant distribution? (let's continue this part at the Discussion; it is always helpful to draw a graph...

@erlebach ah, right, I made a typo. Your variant is correct. FL just ignored the last dimensionality specification. We should throw an error or warning when a user does it.

I have a feeling that you haven't specified the dimensionality of your placeholder as I, i.e. `placeholder(s, :s, dims=(2,))`. Are you sure you are using exactly the same code? As...

I understand, no worries. Well, actually you also use "sampling". ForneyLab knows how to combine different message passing techniques. Here you have used hybrid message passing, i.e. SP, VMP, and...

Hi @erlebach! Here's pseudocode for tracking your marginals: ```julia marginals = Dict{Any, Any}(:g => ProbabilityDistribution(Univariate, Categorical, p=one(3)/3)) marginal_storage = Dict{Any, Any}(:g => [], :i => [], :d => []) #...

@erlebach hi! Not sure if this is still relevant, but `ReactiveMP.jl` has released message-passing with a boolean function. You may want to look at this [demo](https://github.com/biaslab/ReactiveMP.jl/blob/master/demo/Assessing%20People%20Skills.ipynb).

Just dropping my thoughts on this issue. When you specify `{allow_missing = true}`, **RxInfer.jl** starts treating the `datavar` as a random variable. Given the constraints, this effectively means that `q_y_x`...

@ismailsenoz @ThijsvdLaar, any thoughts on this one?