I am leaning twards Suneel's position on this.. for different reasons and three orders of magnitude less -ve..
Re: GPU gpu and streams it's difficult .. there is alot of overhead in copying from main memory to gpu memory, and so it makes sense to wait for a bat h of data to come in anyway with our framework... There could be some interesting window g applications for this, eg stream a 5 sec window to main memory, copy to gpu, and perform operation, copy back, etc.. It would be very cool to stream directly to gpu memory would be a cool engineering problem that I wish I had the time to work on. All in all I'd say I'm -.9 on this, since we have a good amount of technical debt that we have to pony up for at the moment. I would think it is something that we may want to revisit wither after 0.14 0 (algos) or after 1.0. Sent from my Verizon Wireless 4G LTE smartphone -------- Original message -------- From: Suneel Marthi <[email protected]> Date: 12/09/2016 9:13 AM (GMT-08:00) To: mahout <[email protected]> Subject: Re: [DISCUSS] Mahout Streaming Bindings -100 It doesn't make sense adding support for all of the batch and streaming engines that are available. We presently have a H2O binding which has never seen any use and I have long been thinking of trashing out. It would be more productive utilization of the very limited resource time by porting the legacy Recommender algorithms to Samsara as opposed to adding anymore of these streaming/batch frameworks. I have had folks asking me if we support Akka Streams, Apex etc... but most folks don't know as to how they are gonna be using it or what is it they are trying to accomplish. How many Recommender Systems are out there today that need Streaming? Amazon doesn't do Realtime streaming recommendations yet. Lambda frameworks like Oryx 2.0, Summingbird etc.. support that and Mahout-Spark should be pluggable as the ML engine into those frameworks. My 2รง. On Fri, Dec 9, 2016 at 11:04 AM, Trevor Grant <[email protected]> wrote: > Wanted to kick off a discussion about if we're ready to start thinking > about coming up with some bindings for streaming engines. > > Had a question at a meetup tuesday in seattle, and it got me thinking about > it. > > In my mind they would be a discreet set of Bindings, Flink Streaming > /Spark Streaming / (Beam?) / etc > > I have some fleeting thoughts, but didn't write any of the bindings and > have only attempted to grok them in passing. > > Or maybe not something we want to pursue at this time? However, it is > something I'd be interested in tinkering with on a branch of my own... > > tg > > Trevor Grant > Data Scientist > https://github.com/rawkintrevo > http://stackexchange.com/users/3002022/rawkintrevo > http://trevorgrant.org > > *"Fortunate is he, who is able to know the causes of things." -Virgil* >
