Live data from Hacker News

Simple Python implementation of a Bayesian multi-armed bandit algorithm

peterroelants.github.io

1–7 of 7 posts

Re: Simple Python implementation of a Bayesian multi-armed bandit algorithm

#2
I should open source my cobbled together implementation for react.

I will add that a couple great things for a dashboard using MAB are

1. Pairwise variant comparisons: i.e., P(A>B) for each pair. http://www.evanmiller.org/bayesian-ab-testing.html is an amazing resource on that.

2. Time-to-Confidence estimates: how long until we will have a credible interval of 5%? 1%? 0.5%?

These can help you cull variants and experiments. Sure, it's optimized so you don't have to for the sake of the test, but code cleanliness is nice to have.

Re: Simple Python implementation of a Bayesian multi-armed bandit algorithm

#3
post #2

I should open source my cobbled together implementation for react. I will add that a couple great things for a dashboard using MAB are 1. Pairwise variant comparisons: i.e., P(A>B) for each pair. http://www.evanmiller.org/bayesian-ab-testing.html is an amazing resource on that. 2. Time-to-Confidence estimates: how long until we will have a credible interval of 5%? 1%? 0.5%? These can help you cull variants and experi…

The evanmiller blogpost is indeed a great resource, I should add that that to further readings.

Re: Simple Python implementation of a Bayesian multi-armed bandit algorithm

#4
post #2

I should open source my cobbled together implementation for react. I will add that a couple great things for a dashboard using MAB are 1. Pairwise variant comparisons: i.e., P(A>B) for each pair. http://www.evanmiller.org/bayesian-ab-testing.html is an amazing resource on that. 2. Time-to-Confidence estimates: how long until we will have a credible interval of 5%? 1%? 0.5%? These can help you cull variants and experi…

Please do. Would love to use something like this.

Re: Simple Python implementation of a Bayesian multi-armed bandit algorithm

#7
post #2

I should open source my cobbled together implementation for react. I will add that a couple great things for a dashboard using MAB are 1. Pairwise variant comparisons: i.e., P(A>B) for each pair. http://www.evanmiller.org/bayesian-ab-testing.html is an amazing resource on that. 2. Time-to-Confidence estimates: how long until we will have a credible interval of 5%? 1%? 0.5%? These can help you cull variants and experi…

I'd love to just see a write up and screenshots :)