Production Grade Machine Learning Service

Stack

● Flask as the web framework.

● Redis for a fast loading of the trained model and other data between the workers.

● NGINX as a web server and reverse proxy.

● Gunicorn automatically creates parallel workers/threads according to the capacity of the machine it is running on.

● Celery to support asynchronous time-consuming requests as training and initializing the ML model.

Important Info

● Made to help you scale from a basic Machine Learning project for research purposes to a production grade Machine Learning web service.
● General purpose project, so it assumes that your service needs initialization, training, saving models to the databases for further usage in estimation.
● Based on Docker, so it could be scalable and OS-agnostic.

For the detailed API, use the file ml-service.yml on any swagger editor, and you will see the API definition.

You can find a postman collection of this service in the file MLServiceStructure.postman_collection.json, use it to validate your deployment.

Don't forget to create the file ./src/config.properties , use the following template to add the auth-related configuration:
NOTE: expiry_time_unit MUST BE ONE OF THE FOLLOWING:
(days | seconds | microseconds | milliseconds | minutes | hours | weeks)

[auth_info]
expiry=XXXX
expiry_time_unit=XXXX  

expiry is basically the amount of time in expiry_time_unit for the generated bearer tokens to expire. example:

[auth_info]
expiry=120
expiry_time_unit=seconds  

Also Don't forget to create the file ./redis/config.properties , use the following template to add the redis information:

MASTER_USER=XXXXX
REDIS_MASTER_PW=XXXXX
REDIS_CELERY_PW=XXXXX
HOST=redis
END_FILE=true

There are no restrictions about the values of XXXX in this file, you can use your own or use the following example:

MASTER_USER=master_user
REDIS_MASTER_PW=1234pw!@$
REDIS_CELERY_PW=4321wp!@$
HOST=redis
END_FILE=true