seglir - npm explorer

SeGLiR
======

SeGLiR is a javascript library for rapid A/B-testing with Sequential Generalized Likelihood Ratio Tests.

Sequential GLR tests are a family of sequential hypothesis tests, i.e. tests that stop as soon as a significant result has been detected. Compared to classical fixed samplesize tests, sequential GLR tests may give a significant decrease in the needed samplesize, while keeping the same type-1 and type-2 error guarantees. Below is a comparison of the expected samplesize for fixed samplesize and sequential GLR tests at the same levels. For a more detailed explanation of this family of tests (as well as the graph below), see the reference.

!Expected samplesize comparison

SeGLiR currently contain these tests:
* comparing two bernoulli proportions (for instance for comparing conversion on a website)
* comparing two normal means (with equal, known or unknown variance)
* choosing the best arm in a multi-armed bandit setting (with δ-PAC guarantees)

Improvements and additions are welcome, take a look at issues for suggested improvements.

$3

The recommended way is to install SeGLiR via node package manager (install node.js if you don't already have it):

``



npm install seglir





$3



If you're not running SeGLiR in node, note that SeGLiR requires jStat.js available.

javascript

var glr = require("seglir");

// create an instance of a two-sided test comparing bernoulli proportions, with indifference region with size 0.01, alpha-level = 0.05, beta-level = 0.10

var test = new glr.test("bernoulli", "two-sided", 0.01, 0.05, 0.10);

// add data as it comes in

...

test.addData({x : 0});

test.addData({x : 0, y : 0});

test.addData({y : 1});

...

// when function returns string 'true' or 'false', the test is concluded

test.getResults()

// get bias-adjusted estimates

test.estimate()





For a complete function reference, see the reference.



$3



Make sure you have grunt and node installed.

To install the development dependencies run

`npm install` and to build it run `grunt`` in the root directory.

$3

SeGLiR is distributed under the MIT License.

$3

The recommended way is to install SeGLiR via node package manager (install node.js if you don't already have it):

``



npm install seglir





$3



If you're not running SeGLiR in node, note that SeGLiR requires jStat.js available.

javascript

var glr = require("seglir");

// create an instance of a two-sided test comparing bernoulli proportions, with indifference region with size 0.01, alpha-level = 0.05, beta-level = 0.10

var test = new glr.test("bernoulli", "two-sided", 0.01, 0.05, 0.10);

// add data as it comes in

...

test.addData({x : 0});

test.addData({x : 0, y : 0});

test.addData({y : 1});

...

// when function returns string 'true' or 'false', the test is concluded

test.getResults()

// get bias-adjusted estimates

test.estimate()





For a complete function reference, see the reference.



$3



Make sure you have grunt and node installed.

To install the development dependencies run

`npm install` and to build it run `grunt`` in the root directory.

$3

SeGLiR is distributed under the MIT License.