Hacker News new | comments | show | ask | jobs | submit login

I personally do not understand the need to compare bandit with A/B.

The goal of A/B is to decide as quickly as possible for the best one.

The goal of bandit is to optimize given content.

I wouldn't use bandit to "decide" the button color but as a simple recommendation system. This seems by far more natural to me as it reacts better with changing optima.

Example: Let's say I run a larger "fun content" media webpage. I have "awesome videos", "funny images" and "goofy articles". On the bottom and right side of each content page i show follow up content. I would use bandit here to optimize the mix of images,videos,text i recommend.

To spice it up: I would create cohorts for my typical behaviour of users (e.g. registered male user) and only consider interactions of the last two weeks into my bandit calculations.




Applications are open for YC Summer 2018

Guidelines | FAQ | Support | API | Security | Lists | Bookmarklet | Legal | Apply to YC | Contact

Search: