What you have outlined is exactly how PIO is designed to work. The EventServer is multi-tenant in that separate datasets and entities can store to it and can even be granted different permissions. But the PredictionServer serves one engine per process on a single port. This scales in a heavyweight manner, one PredictionServer per engine per port.
What template are you using and how many datasets? On May 25, 2017, at 2:40 AM, Ravi Kiran <[email protected]> wrote: I have a template code ready. Lets say I have multiple datasets D1, D2, ...,Dn How to train model M1, M2, ...Mn corresponding to Di dataset and deploy each of these model in one engine. (I am assuming running multiple engines is not efficient way) Only solution I found so far was to create different apps and add one dataset to one app. Then train and deploy individual model on different ports. But it wont be scalable. As no. of models is limited by no. of ports. Can someone suggest better solution?
