Multivariate normal-inverse-gamma distribution.
This represents the joint distribution:
which may be denoted:
and is a conjugate prior of a Gaussian distribution with both unknown mean and variance. The variance scaling is independent and identical in the sense that all components of share the same .
In model code, it is not usual to use this final class directly. Instead, establish the conjugate relationship via code such as the following:
σ2 ~ InverseGamma(α, β); x ~ Gaussian(μ, Σ*σ2); y ~ Gaussian(x, σ2);
where the last argument in the distribution of
y must appear in the
last argument of the distribution of
x. The operation of
be multiplication on the left (as above) or the right, or division on the