pipeline: create a pipeline to help us labeling

[yorkie]

Alternative bot is here: https://github.com/marketplace/issue-label-bot, which uses machine learning for labeling like feature_request and question, unfortunately the bot doesn't support more categories.

So that's why we want to create yet another one, moreover we will make it:

• open source 100%, so that anyone could build own bot by Pipcook.
• managed at GitHub and deployed to Heroku automatically, so that every commit will change the bot's behavior, that's cool, right?.

And it must be developed community-driven, so if you are interested in having a try? Just look at the following plugins that I listed for this pipeline:

type	plugin
Data Collect	data-collect-github, selects the issue list by a repo, and outputs csv files
Data Access	csv-data-access
Model Define	bayesian-model-define
Model Train	bayesian-model-train
Model Evaluate	bayesian-model-evaluate

Steps to implement:

• [ ] write data-collect-github to output csv files just like: text-bayes-classification.json.
• [ ] run the new pipeline and make sure its result is green.
• [ ] deploy the model service at Heroku.
• [ ] add GitHub event trigger and call the above model service to add labels or comments automatically.

[Txiaozhe]

Do you mean that create a pipeline of pipcook to label all of prs and issues automatic? It means that this pipeline will become a use case for pipcook for github?

[yorkie]

Exactly.

[Txiaozhe]

Great, it may also be a good case for pipcook:)

[yorkie]

A possible rule-based solution is use GitHub's actions/labeler, and an example is here: https://github.com/bytecodealliance/wasmtime/blob/master/.github/labeler.yml :)

[Txiaozhe]

It looks like label all of things by regexp:(

[yorkie]

It looks like label all of things by regexp:(

The labeler action could reduce our works for labeling on PR partially, see #176. It's a good start to address this issue :)