On Monday 07 April 2008, Ted Dunning wrote:
> > That is the simplest explanation of Gibb's sampling I have read so far :)
>
> Of *one*kind* of Gibb's sampling anyway.  :)

So that is why it sounded almost to easy to be all there is about Gibb's 
sampling ;)


> Yes.  You still pass some parameters in via the prior and there is one
> parameter that expressive how aggressive the algorithm is about thinking
> there is another cluster it doesn't yet know about.

Right. I remember that. I think that is still easier than fixing the number of 
clusters in advance.


> The jury is still out on how well Gibb's sampling works for infinite
> mixtures with text, but there are some very exciting early developments.
>
> See DCA and LDA and the later adaptations to non-parametric form in the
> following references.
>
> http://cosco.hiit.fi/Articles/buntineBohinj.pdf
> http://www.cs.princeton.edu/~blei/papers/BleiNgJordan2003.pdf

I do remember these two.


> http://citeseer.ist.psu.edu/756482.html
> http://citeseer.ist.psu.edu/736109.html
> http://citeseer.ist.psu.edu/756901.html

Thanks for the additional pointers.


> > I wonder whether one should split this thread into at least four threads:
>
> +10.

Done.


> The first and last might be the same unless somebody is interested in doing
> a general purpose EM implementation.

I have at least created an additional issue for a general purpose EM 
implementation so discussion on whether we want that or not can be tracked 
within JIRA.

Isabel


-- 
Flying is the second greatest feeling you can have.  The greatest 
feeling?Landing...  Landing is the greatest feeling you can have.
  |\      _,,,---,,_       Web:   <http://www.isabel-drost.de>
  /,`.-'`'    -.  ;-;;,_
 |,4-  ) )-,_..;\ (  `'-'
'---''(_/--'  `-'\_) (fL)  IM:  <xmpp://[EMAIL PROTECTED]>

Attachment: signature.asc
Description: This is a digitally signed message part.

Reply via email to