Swab: Simple WSGI A/B testing
Project description
Simple WSGI A/B testing.
Usage
Define an experiment:
from swab import Swab s = Swab('/tmp/.swab-test-data') s.addexperiment('button-size', ['default', 'larger'], 'order-completed') Here we have an experiment called 'button-text', for testing two sizes of button on a form. The two variants the experment are called 'small' and 'large'. The first listed is always taken as the control variant, and results will be presented in terms of how much better or worse the other variants have performed compared with the control The goal associated with the test is called 'order-completed'.Take a WSGI application:
from swab import record_goal, show_variant def get_button_style(environ): if show_variant('button-size', environ) == 'small': return 'font-size: 100%' else: return 'font-size: 150%' def myapp(environ, start_response): if environ['REQUEST_METHOD'] == 'GET': start_response('200 OK', []) return [''' <html> <body> <form method="POST"> <button style='%s' type="submit">Complete your order</button> </form> </body> </html>''' % (get_button_style(environ),) ] if environ['REQUEST_METHOD'] == 'POST': swab.record_goal('order-completed', environ)Wrap it in the swab middleware:
app = s.middleware(myapp)
4. Let it run, and after a while browse to http://<server>/swab/results to see which variant is winning.
Testing
To test your competing pages, start Swab in debug mode:
swab = Swab('/tmp/.swab-test-data', debug=True)
Then append ‘?swab.<experiment-name>=<variant-name>’ to URLs to force one variant or another to be shown.
Disable debug mode in production to avoid people messing about with your figures.
Basic design
Each visitor is assigned an identity which is persisted by means of a cookie. The identity is a base64 encoded randomly generated byte sequence. This identity is used as a seed for a RNG, which is used to switch visitors into test groups.
Every time a test is shown (triggered by the show_variant function), a line is entered into a file at <datadir>/<experiment>/<variant>/__all__
Every time a goal is recorded (triggered by the record_goal function), a line is entered into a file at <datadir>/<experiment>/<variant>/<goal>
Each log line has the format <timestamp>:<identity>\n.
No file locking is used: it is assumed that this will be run on a system where each line is smaller than the fs blocksize, allowing us to avoid this overhead. The lines may become interleaved, but there should be no risk of corruption even with multiple simultaneous writes. See http://www.perlmonks.org/?node_id=486488 for a discussion of the issue.
Statistical analysis
The further reading section contains links to the various articles that helped me implement the algorithms and statistical formulae used in Swab. However I don’t have a background in maths or stats and I would welcome contributions or corrections from anyone with a deeper understanding of these tests.
TODO
Allow a set percentage of users to be tipped into each test version
Filter search engine spiders out from tests
Investigate multivariate testing
Allow users to be excluded from tests after being shown a test variant (this would js-based exclusion of bots, eg http://www.kalzumeus.com/2010/06/07/detecting-bots-in-javascrip/)
Further reading
2010 Oliver Cope. For licensing information please see LICENSE.txt
Version 0.1.1
Bugfix for ZeroDivisionErrors when no data has been collected
Version 0.1
Initial release
Download files
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.
