Operations on objects with different element type

[nathanaelbosch]

The concrete example I have is a multiplication

[1 0; 0 1] * Normal(rand(2), Symmetric(rand(2,2)))

This does not work right now. I think similar considerations apply to many other functions that are implemented, and they often assume that the element type is exactly the same. Maybe it would be better to let julia handle type promotion etc?

[filtron]

I think the problem is that e.g.

*(C::Abstractmatrix{T}, D::AbstractDistribution{T})

is really defined as

marginalize(D, DiracKernel(C))

so the following should not really be allowed (and vice versa):

*(C::Abstractmatrix{<:Real}, D::AbstractDistribution{<:Complex})

I guess an alternative is to simply do:

*(C::AbstractMatrix{<:Real}, D::AbstractDistribution{<:Real}) = marginalize(D, DiracKernel(C))
*(C::AbstractMatrix{<:Complex}, D::AbstractDistribution{<:Complex}) = marginalize(D, DiracKernel(C))

Unless there is a more clever way?

[filtron]

The above is too complicated I think just treating e.g. Normal{<:Real} as a Normal{<:Complex} with zero imaginary variance appears the simplest. This would make the following legal:

kldivergence(::Normal{<:Complex}, ::Normal{<:Real})

which is a bit awkward but it is perhaps still fine.