Thanks, Tim. Yes, of course the dispatch mechanism is a much better way to
find out the type parameter.
Cheers,
---david
On Tuesday, October 28, 2014 11:10:57 AM UTC+1, Tim Holy wrote:
>
> 2nd is much better. But you can even write your `kind` function using
> dispatch:
>
> kind{T}(g::GMM{T,Matrix{T}}) = :diag
> kind{T}(g::GMM{T,Vector{Matrix{T}}}) = :full
>
> See the FAQ for more info about containers with abstract fields
> (especially the
> part about inner constructors).
>
> --Tim
>
> On Tuesday, October 28, 2014 02:03:57 AM David van Leeuwen wrote:
> > Hello,
> >
> > For a long lime I've been having a bad feeling for having a Union in a
> > composite type, as in
> >
> > type GMM{T:<FloatingPoint}
> > ...
> > kind::Symbol
> > μ::Matrix{T}
> > Σ::Union(Matrix{T},Vector{Matrix{T}})
> > ...
> > end
> >
> > where the choice in `Σ` depends on whether `kind==:diag` or
> `kind==:full`.
> > I have the feeling it is better to have defined something along the
> lines
> > of
> >
> > typealias DiagCov{T} Matrix{T}
> > typealias FullCov{T} Vector{Matrix{T}}
> >
> > type GMM{T:<FloatingPoint,CT}
> > ...
> > μ::Matrix{T}
> > Σ::CT
> > end
> >
> > function kind(g::GMM)
> > T = eltype(g.μ)
> > if isa(g.Σ, FullCov{T})
> > return :full
> > else
> > return :diag
> > end
> > end
> >
> >
> > In the second case, the `kind` is inferred from the type instance,
> rather
> > than stored as a type variable.
> >
> > What is in your expert's opinion a better way of defining such a type?
> >
> > Thanks,
> >
> > ---david
>
>
