Sorry, I got it. It definitely used 0.20.2 hadoop api. I was wrong.
When I checked kmeans source code, I still have some questions. It seems use
fs to read cluster file from hdfs in every iteration process of map process.
see:
KMeansUtil.java  line 73
reader = new SequenceFile.Reader(fs, path, conf);

Doesn't it a better way to use distributed cache?

Thank you, guys.


2010/8/19 Ted Dunning <[email protected]>

> Are you looking at the current trunk?
>
> On Wed, Aug 18, 2010 at 1:39 AM, Cui tony <[email protected]> wrote:
>
> >  But many algorithms still used 0.19 api, for example k-means.
> >  Is there any plan to re-write this algorithm one by one?
> >
> >
> > 2010/8/17 Drew Farris <[email protected]>
> >
> > > Mahout currently depends on 0.20.2, and the new 0.20.x api is used in
> > > many cases, so 0.19 is no longer an option for 0.4
> > >
> > > On Tue, Aug 17, 2010 at 7:43 AM, Cui tony <[email protected]>
> > wrote:
> > > > Hi,
> > > >
> > > >  Will mahout 0.4 be released used hadoop 0.20.2 or 0.19?
> > > >
> > >
> >
>

Reply via email to