Abby is a simple, but powerful PHP A/B testing library.
The library lets you easily setup tests (experiments), their control and variation groups, track your users, get detailed statistics (like recommended sample sizes, and determining the confidence of your results), including whether an experiment has achieved statistical significance.
The winner (and confidence) is detemined using Bayesian statistics, calculating the p-value of your results to check if the null hypothesis can be rejected. An accompanying minimum sample size is also calculated using a two-tailed test to control the false discovery rate.
Abby is dependency free, and completely database agnostic, meaning it simply works with data you provide it with, and exposes a variety of methods for you to store the result in your own storage of choice.
- PHP 5.4.0 or higher
composer require andreekeberg/abby
// Setup a new Token instance
$token = new Abby\Token();
// If we can't find an existing token cookie, generate one and set tracking cookie
if (!$token->getValue()) {
$token->generate()->setCookie();
}
// Setup a User instance
$user = new Abby\User();
// Associate the token with our user
$user->setToken($token);// List of experiments associated with a tracking token
$data = [
[
'id' => 1,
'group' => 1,
'viewed' => true,
'converted' => false
]
];
// Loop through users existing experiments and add them to our user instance
foreach ($data as $item) {
// Setup experiment instance based on an existing experiment
$experiment = new Abby\Experiment([
'id' => $item['id']
]);
// Setup a group instance based on stored data
$group = new Abby\Group([
'type' => $item['group']
]);
// Add the experiment (including their group, and whether they have viewed and converted)
$user->addExperiment($experiment, $group, $item['viewed'], $item['converted']);
}// Experiment data
$data = [
'id' => 2
];
// Make sure the experiment isn't already in the users list
if (!$user->hasExperiment($data['id'])) {
// Setup a new experiment instance
$experiment = new Abby\Experiment([
'id' => $data['id']
]);
// Assign the user to either control or variation in the experiment
$group = $user->assignGroup($experiment);
// Add the experiment (including assigned group) to our user instance
$user->addExperiment($experiment, $group);
}
// Getting updated user experiment list
$user->getExperiments();
// Store updated experiment list for our user// Experiment data
$data = [
'id' => 1
];
// Record a view for the experiment in question
$user->setViewed($data['id']);
// If the user is part of the variation group
if ($user->inVariation($data['id'])) {
// Apply a custom class to an element, load a script, etc.
}// Experiment data
$data = [
'id' => 1
];
// On a custom goal, check if user has viewed the experiment and define a conversion
if ($user->hasViewed($data['id'])) {
$user->setConverted($data['id']);
}
// Getting updated user experiment data
$user->getExperiments();
// Store updated data for our user// Setup experiment instance with stored results
$experiment = new Abby\Experiment([
'groups' => [
[
'name' => 'Control',
'views' => 3000,
'conversions' => 300
],
[
'name' => 'Variation',
'views' => 3000,
'conversions' => 364
]
]
]);
// Retrieve the results
$result = $experiment->getResult();
// Get the winner
$winner = $result->getWinner();
/**
* Get whether we can be confident of the result (even if we haven't
* reached the minimum number of views for each variant)
*/
$confident = $result->isConfident();
/**
* Get the minimum sample size (number of views) required for each group to
* reach statistical significance, given the control groups current conversion
* rate (based on the configured minimumDetectableEffect)
*/
$minimum = $result->getMinimumSampleSize();
/**
* Get whether the results are statistically significant
*/
$significant = $result->isSignificant();
/**
* Get complete experiment result
*/
$summary = $result->getAll();Read the contribution guidelines.
Refer to the changelog for a full history of the project.
Abby is licensed under the MIT license.