Stat support in Mahout for dev

[Dmitriy Lyubimov <dl...@gmail.com>]

Hi,

I vaguely remember mention of stat external dependencies in mahout. What
dependencies we would rely on if we wanted to sample say from inverse gamma
or wishart?

Thanks in advance .

Re: Stat support in Mahout for dev

[Dmitriy Lyubimov <dl...@gmail.com>]

Thank you, Ted. no, they don't have inverse Wishart. That's probably the
only one that is interesting enough and they don't have. I am looking again
at those hierarchical implementations along the lines of Agarwal's RLFMs.


On Mon, Feb 10, 2014 at 4:47 AM, Ted Dunning <te...@gmail.com> wrote:

> So far these have been satisfied by commons math.
>
> Not sure if commons has the wishart distribution.  It should be a moderate,
> but reasonable amount of work to implement.
>
> Inverse gamma should be relatively trivial (generate from x ~ Gamma, return
> 1/x)
>
>
>
> On Sun, Feb 9, 2014 at 9:51 PM, Dmitriy Lyubimov <dl...@gmail.com>
> wrote:
>
> > Hi,
> >
> > I vaguely remember mention of stat external dependencies in mahout. What
> > dependencies we would rely on if we wanted to sample say from inverse
> gamma
> > or wishart?
> >
> > Thanks in advance .
> >
>

Re: Stat support in Mahout for dev

[Ted Dunning <te...@gmail.com>]

So far these have been satisfied by commons math.

Not sure if commons has the wishart distribution.  It should be a moderate,
but reasonable amount of work to implement.

Inverse gamma should be relatively trivial (generate from x ~ Gamma, return
1/x)



On Sun, Feb 9, 2014 at 9:51 PM, Dmitriy Lyubimov <dl...@gmail.com> wrote:

> Hi,
>
> I vaguely remember mention of stat external dependencies in mahout. What
> dependencies we would rely on if we wanted to sample say from inverse gamma
> or wishart?
>
> Thanks in advance .
>