Currently we just have MVN; but the easiest way forward (IMO) here is to implement some others as "scale mixtures of Normals". Note this only works for distributions with unbounded support.So, the proposal is to allow (for instance) multivariate Student's t with arguments (vector of means, degrees of freedom, covariance matrix), implemented as MVN multiplied by (IIRC) an inverse square root Gamma random variable with the appropriate parameters given the degrees of freedom.
Currently we just have MVN; but the easiest way forward (IMO) here is to implement some others as "scale mixtures of Normals". Note this only works for distributions with unbounded support.So, the proposal is to allow (for instance) multivariate Student's t with arguments (vector of means, degrees of freedom, covariance matrix), implemented as MVN multiplied by (IIRC) an inverse square root Gamma random variable with the appropriate parameters given the degrees of freedom.