You model is really just a function that you wrap in a REST service in order to deploy in MaaS. In the case of something like spark, you would just wrap it in a udf instead of wrapping it in a REST service, at that point, applying it in batch is just a case of a simple dataframe query.
On Fri, 16 Nov 2018 at 15:51, deepak kumar <kdq...@gmail.com> wrote: > Simon, > Can you elaborate more on this: > ' > > *wrapped up in a batch engine like Spark to takeadvantage of more > efficient "mass" scoring.* > ' > How the mass model wrapped in spark can take advantage of mass scoring? > > Thanks > Deepak > > On Fri, Nov 16, 2018 at 9:15 PM Otto Fowler <ottobackwa...@gmail.com> > wrote: > >> That may be the best MAAS explanation I’ve seen Simon. >> >> >> On November 16, 2018 at 10:28:57, Simon Elliston Ball ( >> si...@simonellistonball.com) wrote: >> >> MaaS is designed to wrap model inference (scoring) an event at a time, >> via a REST api. As such, running it batch doesn't make a lot of sense, >> since each message would be processed individually. Most of the models >> you're likely to run in MaaS however, are also likely to be easily >> batchable, and are probable better wrapped up in a batch engine like Spark >> to take advantage of more efficient "mass" scoring. >> >> Simon >> >> On Fri, 16 Nov 2018 at 15:18, deepak kumar <kdq...@gmail.com> wrote: >> >>> Hi All >>> Right now MAAS supports running the model against real time events being >>> streamed into metron platform. >>> Is there any way to run the models deployed in MAAS on the batch events >>> / data that have been indexed into hdfs ? >>> If anyone have tried this batch model , please share some insights. >>> Thanks >>> Deepak. >>> >>> >> >> -- >> -- >> simon elliston ball >> @sireb >> >> -- -- simon elliston ball @sireb
