dabble

Simple A/B testing framework


Keywords
python, web, abtest, split, ab, a/b, test
License
BSD-3-Clause
Install
pip install dabble==0.1

Documentation

Dabble

Dabble is a simple A/B testing framework for Python. Using dabble, you configure your tests in code, collect results, and analyze them later to make informed decisions about design changes, feature implementations, etc.

You define an A/B test in dabble with class ABTest, which describes the test name, the names of each of the alternatives, and the set of steps the users will progress through during the test (in the simplest case, this is just two steps). You then define one or more ABParameters, which contain the values you wish to vary for each alternative in the test. Each test can have one or more alternatives, though the most common case is to have 2 (hence "A/B testing").

Example

import dabble
dabble.configure(
    CookieIdentityProvider('dabble_id'),
    FSResultStorage('/path/to/results.data')
)

class Signup(page):
    path = '/signup'

    signup_button = ABTest('signup button',
                           alternatives=['red', 'green'],
                           steps=['show', 'signup'])
    button_color = ABParameter('signup button', ['#ff0000', '#00ff00'])

    def GET(self):
        self.signup_button.record('show')
        return render('index.html', button_color=self.button_color)

    def POST(self):
        self.signup_button.record('signup')
        return redirect('/account')

Behind the scenes, dabble has used a cookie for each user on your site to assigne them each an identity, so that each user always ever sees the same alternative. Users may visit the homepage many times over many browsing sessions, but as long as they have the same cookie present in their browser, they will always see either the red or the green button, depending on which was chosen the first time the viewed the page.

When a user signs up, the record() method of ABTest is called, to track the user's action. Later on, reports can be generated to determine whether the red or the green button induced more users to sign up.

Configuring Dabble

In addition to ABTest and ABParameter, dabble also needs an IdentityProvider and a ResultsStorage. Dabble provides several alternatives for each of these out of the box, and it is also straightforward to write your own.

IdentityProviders should make their best possible effort to always identify individuals, rather than browsing sessions (particularly if cookies are set to expire when the user closes his/her browser). If you are testing a feature that requires users to be logged in, then their username is a good choice for identity.

ResultsStorage stores configuration and results of A/B tests, and provides some facilities for generating reports based on the stored results. Dabble provides several backends, including MongoResultsStorage, and FSResultsStorage.

At this time it is not possible to configure different IdentityProviders or ResultsStorages for different tests within the same application.

Reporting

Dabble will also produce reports on all users who have taken part in an A/B test, by way of the report() method. The report is a dictionary which describes, for each alternative, how many users attempted and converted at each of the defined steps. For the above example, a report might look like:

>>> storage = FSResultStorage('/path/to/results.data')
>>> storage.report('signup button')
{
    'test_name': 'signup button',
    'results': [
        {
            'alternative': 'red',
            'funnel': [{
                'stage': ('show', 'signup'),
                'attempted': 187,
                'converted': 22,
            }],
        },
        {
            'alternative': 'green',
            'funnel': [{
                'stage': ('show', 'signup'),
                'attempted': 195
                'converted': 18,
            }],
        }
    ],
}

The funnel key in each of the results entries will have one element fewer than the number of steps, since each entry describes the progression of users from one step to the next.