Robin, This is a great example of how the problem is a bit harder for Mahout. Mnay of the large datasets of interest will not fit in memory. That means that mixed memory and disk formats are important in this flow which also makes API's more complicated.
On Sat, Feb 22, 2014 at 10:12 AM, Robin East <[email protected]> wrote: > Various existing frameworks like R or python/sklearn already have existing > interfaces so would be good to follow some existing patterns. I personally > find sklearn's transform/fit/predict pattern very easy to use and > understand. > > Sent from my iPhone > > On 22 Feb 2014, at 09:15, Gokhan Capan <[email protected]> wrote: > > > I'm personally positive on this. > > > > Could you give an example code snippet that shows how the usage is going > to be? > > > > Sent from my iPhone > > > >> On Feb 22, 2014, at 5:37, Jay Vyas <[email protected]> wrote: > >> > >> Hi dead. Sure I will take a look. > >> > >> > >>> On Fri, Feb 21, 2014 at 7:51 PM, Ted Dunning <[email protected]> > wrote: > >>> > >>> Great idea. Hard to do well. > >>> > >>> Would it be possible for you to try to build a picture of all the > pieces > >>> that need to be connected before you start building connectors and > >>> converters? > >>> > >>> > >>> > >>>> On Fri, Feb 21, 2014 at 8:01 AM, Jay Vyas <[email protected]> > wrote: > >>>> > >>>> Hi mahout. Was thinking about building adapters so it would be > easier to > >>>> run algorithms on a broader set of input data structures, unless > there is > >>>> already an initiative to do so: > >>>> > >>>> Any thoughts on this JIRA? > >>>> https://issues.apache.org/jira/browse/MAHOUT-1421 > >>>> > >>>> -- > >>>> Jay Vyas > >>>> http://jayunit100.blogspot.com > >> > >> > >> > >> -- > >> Jay Vyas > >> http://jayunit100.blogspot.com >
