Use machine learning to automatically classify issues/MRs

Description

To be able to handle large amount of issues, I want them categorized and labelled.

Doing this manually is a lot of work. If we'd be able to automate it in a smart way, this would help us a lot.

Proposal

We look to machine learning to automatically classify issues, which would allow us to:

• automatically set labels (easiest target)
• automatically assign something
• etc.

You can have title and description as input and label as output.

A simple implementation would:

1. on creation of an issue, send the issue title and body to a queue to be picked up by a TensorFlow worker
2. a worker picks it up and..
3. if the issue is labelled, uses it to train its network
4. if the issue is not labelled, takes the title and body as input and returns suggested labels (feature proposal or bug, to start
5. the worker sets the label

Links / references

@zj for the idea

cc @sytses