As Mahout PMC, I concur with Pat that Samsara backends are flaky excepting for Spark today.
On Fri, Jan 20, 2017 at 1:12 PM, Pat Ferrel <[email protected]> wrote: > The Universal Recommender uses Mahout Samsara so yes, of course. See the > UR template for how to do this “wrapping”. > > The core algorithm for Correlated Cross-Occurrence comes from this bit: > http://mahout.apache.org/users/algorithms/recommender-overview.html > > BTW the backend support is spotty for everything but Spark. Some things > are implemented and some not. > > > On Jan 20, 2017, at 10:07 AM, Gustavo Frederico < > [email protected]> wrote: > > > Is it possible to create an engine using Mahout Samsara? I was looking > at http://predictionio.incubator.apache.org/system/ and comparing with > Samsara's features at http://mahout.apache.org/ . I can see that Samsara > runs on distributed Spark, H2O, and Flink. I can see that Spark is a > binding for Samsara ( http://mahout.apache.org/ > users/sparkbindings/home.html ). I'm not very familiar with the > terminology, but would it be a matter of 'wrapping' the Samsara algorithm > in some class that implements the interfaces described in the DASE page ( > http://predictionio.incubator.apache.org/customize/dase/ ) ? > > Thanks > > Gustavo Frederico > > >
