The rules toward variance in Gaussian node produce improper InverseGamma distributions that is incompatible with projection like methods

[bvdmitri]

I guess, this is intentional, but the rules defined in normal_mean_variance/var.jl produce GammaInverse with α explicitly set to -1/2. This renders the distribution as improper and does not allow to use it with our available approximation methods.

[abpolym]

Is this still an issue? If yes can you provide some simple example code that throws an error but should not throw an error when the issue is resolved? Btw nice issue number.

[bvdmitri]

Btw nice issue number.

:D

Is this still an issue?

yeah, I think we have to live with that or wait if someone actually having issues in some models with that, for which we can either re-open this one or open a new one