Skip to content
This is a collection of multi-armed-bandit examples written in javascript.
Branch: master
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Permalink
Type Name Latest commit message Commit time
Failed to load latest commit information.
js
README.md
bandit.htm

README.md

js-multi-armed-bandits

This is a collection of multi-armed-bandit examples written in javascript.

Multi-armed-bandits might be useful for determining the effectiveness of multiple variations of a web design simultaneously, when the website in question has high enough traffic that significant results can be achieved quickly.

How it works

The MAB is provided with a list of 'levers' each with their expected conversion rate, which the bandit will discover over repeated attempts.

The "Auto" button on the page starts the test for each module. A random number is generated and compared to the expected conversion rate for the current lever, rewarding the lever or starting the next iteration depending on the value.

A variety of different strategies are present on the page, as are statistical significance calculations and a regret delta for each strategy. The aim is to achieve significance for all levers with minimum regret in the shortest amount of time possible.

Note

If you're looking for a really great set of MAB implementations, this may be a better resource: http://bandit.sourceforge.net/

You can’t perform that action at this time.