Wouter Nuijten
Wouter Nuijten
When defining rules around a node where a `PointMass` distribution comes in, we default to `@rule .... (q_out :: PointMass,)` even if the prior is conjugate. Therefore when defining a...
If we want to do streaming inference where we don't want to do state estimation, we might want to use `rxinference` without having to use `autoupdates`.
Should be very easy (parameterizing the backend) since all machinery is already there in `GraphPPL`, but we should expose (and document) the API to users
If we add a vector/tensor of inputs in streaming inference and create a (data)variable for every entry, we get the following error message: ```julia MethodError: no method matching is_data(::Vector{RxInfer.GraphVariableRef}) Closest...
We should be able to access posteriors of variables inside nested models (also such that we can do`@autoupdates` for variables inside nested models). Now we can only access inference results...
This only works when we have 4 or more interfaces, since we need SVMP with 3 or more clusters. For example, in the GCV node, we have $q(x, y)q(w)q(k)q(z)$ as...
Whenever we try to put a `PointMassFormConstraint` on a variable where the marginal is a `MatrixDirichlet` we run into an error: MWE: ```julia @model function foo(y) p ~ MatrixDirichlet(ones(5, 5))...
MWE: ```julia θ = 1.0f0 d = Exponential(θ) typeof(rand(Exponential(θ))) == typeof(θ) ``` returns `false` Because Exponential sampling is also used when sampling `Gamma` with shape 1, the problem also occurs...