Thank you for your reply. But the data is very sparse and reformat the input file means add a lot of zero, which will affect the performance. To me, it’s better to stick to the sparse representation.
> On Aug 18, 2015, at 10:15 PM, Ted Dunning <[email protected]> wrote: > > Seems like a simple format translation. > > Why not just reformat the input file? > > > > On Tue, Aug 18, 2015 at 7:42 PM, Zhou Jiang <[email protected]> wrote: > >> Hi All, >> >> The Default Random Forests MapReduce works with UCI glass data. >> >> ID f1 f2 f3 … fn L >> >> Is there a way to make it work with image data in libsvm format(Sparse >> Representation) listed as blow: >> >> L index1:feature1 index2:feature2 … indexk:featurek >> >> >> Thanks >> >>
