SESSION

Non-Flink Machine Learning on Flink

It is a truism that when asked which is the best machine learning framework to use, the only correct answer is "all of them". That's terribly cute until we actually have to deploy the models produced by "all of them". Flink has some interesting machine learning capabilities, but it really isn't the center of gravity for the machine learning world and it certainly isn't much in the frame of mind for people developing machine learning frameworks. So if somebody on my team develops a model using Tensorflow or DL4J or R, how can I deploy that model? More importantly, if somebody on my team develops 200 models using a variety of tools, how can I deploy and manage them? I will talk about practical techniques for model development, testing, management and deployment in a stream-first world. This won't just be about how to interface with the code, but will also cover aspects of how to deal with live deployment of models in a real-time world.

Speakers involved
Name

Ted Dunning


Chief Application Architect,
MapR Technologies