Weekly Digest #47

3 min readOct 25, 2021

Articles

from many individual training data generation pipelines and models with different feature-set

to using unified precomputed embedded feature with single model

result in improvement in engagement, diversity, infra cost.

when creating a pull request, we think about:

This translates to 7 types of pull-request:

No Pull Request/Ship (I’m so confident in my changes (e.g. documentation changes), I don’t need to see if they’ll pass CI. Ship it now!)
Never Merge (This is probably a proof-of-concept, and I want to see what the CI thinks of my approach. It may never leave draft)
CI Review (I’m confident of my changes, but I want the CI to check them. If they pass CI, they can be merged (e.g. using Github’s “Enable auto-merge”))
Self-Review (This is a CI review, but I want to review my own work as if I was one of the official reviewers.)
Look At This! (I want my colleagues to be able to take a look, as there is something new and interesting in this pull requests they’ll want to know about.)
Optimistic Merge (I want my colleagues to be able to take a look. I’m going to merge this as soon as it passes CI, but they’ve been notified and can review it if they want to.)
Pessimistic (I want my colleagues to take a good look. I’m not confident in my changes, and I know that at least one of my colleagues will be able to give good feedback on it.)