An upper confidence bounds algorithm for multi-armed bandit problems

This implementation is based on Bandit Algorithms for Website Optimization and related empirical research in "Algorithms for the multi-armed bandit problem". In addition, this module conforms to the BanditLab/2.0 specification.

• Node.js 6.x+ (LTS track)
• npm

Install with npm (or yarn):

npm install ucb --save

This implementation often encounters extended floating point numbers. Arm selection is therefore subject to JavaScript's floating point precision limitations. For general information about floating point issues see the floating point guide.

1. Create an optimizer with 3 arms:

   const Algorithm = require('ucb');
   
   const algorithm = new Algorithm({
     arms: 3
   });

2. Select an arm (exploits or explores, determined by the algorithm):

   algorithm.select().then((arm) => {
     // do something based on the chosen arm
   });

3. Report the reward earned from a chosen arm:

   algorithm.reward(arm, value);

Creates a new optimization algorithm.

• config (Object): algorithm instance parameters

The config object supports two optional parameters:

• arms (Number, Integer): The number of arms over which the optimization will operate; defaults to 2

Alternatively, the state object resolved from Algorithm#serialize can be passed as config.

An instance of the ucb optimization algorithm.

const Algorithm = require('ucb');
const algorithm = new Algorithm();

assert.equal(algorithm.arms, 2);

Or, with a passed config:

const Algorithm = require('ucb');
const algorithm = new Algorithm({ arms: 4 });

assert.equal(algorithm.arms, 4);

Choose an arm to play, according to the optimization algorithm.

None

A Promise that resolves to a Number corresponding to the associated arm index.

const Algorithm = require('ucb');
const algorithm = new Algorithm();

algorithm.select().then(arm => console.log(arm));

Inform the algorithm about the payoff from a given arm.

• arm (Number, Integer): the arm index (provided from Algorithm#select())
• reward (Number): the observed reward value (which can be 0 to indicate no reward)

A Promise that resolves to an updated instance of the algorithm. (The original instance is mutated as well.)

const Algorithm = require('ucb');
const algorithm = new Algorithm();

algorithm.reward(0, 1).then(updatedAlgorithm => console.log(updatedAlgorithm));

Obtain a plain object representing the internal state of the algorithm.

None

A Promise that resolves to a stringify-able Object with parameters needed to reconstruct algorithm state.

const Algorithm = require('ucb');
const algorithm = new Algorithm();

algorithm.serialize().then(state => console.log(state));

PRs are welcome! For bugs, please include a failing test which passes when your PR is applied. Travis CI provides on-demand testing for commits and pull requests.

1. Feature development and bug fixing should occur on a non-master branch.
2. Changes should be submitted to master via a Pull Request.
3. Pull Requests should be merged via a merge commit. Local "in-process" commits may be squashed prior to pushing to the remote feature branch.

To enable a git hook that runs npm test prior to pushing, cd into the local repo and run:

touch .git/hooks/pre-push
chmod +x .git/hooks/pre-push
echo "npm test" > .git/hooks/pre-push

To run the unit test suite:

npm test

Or, to run the test suite and view test coverage:

npm run coverage