Hello All,
with the recent work on ML one thing i noticed was the time taken to load a model at the start for any ML program, in my case it was python.
As with traditional python integration approaches, i started off with integration with BP through run process and then getting the output in a file. Also as some time is taken to load the model, it was more of a batch approach.
One of my colleagues later came up with the solution of using DjangoRestFramework package to wrap the python code as a Rest service, and keep the model loaded in memory, there making this real time and we obtained the response of the python code in milliseconds as a json web api response.
Another option is to load the model in memory always through pickle etc.
Coming from a web service background, i know that as this matures and into the field of more real time responses, the technology being used to wrap the python program would get more configurable and "integratable" (i doubt thats a word) similar to a web server providing a web service.
Wanted to check if anyone has used other tools in their projects - like flask, Django etc..
and is there a tool available which supports standards like WS reliability, transactions, memory configuration etc...
I mean tools which have the required configurations and reliability to run in production which exposing the python ML programs as Rest APi services.
thanks
Ashish
------------------------------
Ashish Easow
Senior Consultant - Professional Services
Blueprism
Asia/Kolkata
------------------------------
