kmeans inital cluster selection

[Mark Snow <ma...@yahoo.com>]

I was looking through the kmeans code. As I recall, a good way to pick the inital cluster positions is to choose random data points. Is there an easy way to do 'randomly select N records' in map reduce?



      

Re: kmeans inital cluster selection

[Miles Osborne <mi...@inf.ed.ac.uk>]

why not just have N distinct keys and within the mapper, assign each item
one of these keys (chosen randomly)

Miles

2008/7/3 Mark Snow <ma...@yahoo.com>:

> I was looking through the kmeans code. As I recall, a good way to pick the
> inital cluster positions is to choose random data points. Is there an easy
> way to do 'randomly select N records' in map reduce?
>
>
>
>




-- 
The University of Edinburgh is a charitable body, registered in Scotland,
with registration number SC005336.