Simple WSGI A/B testing.

Usage
-----

1. 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'.

2. 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)

3. 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
---------------

    - http://www.slideshare.net/patio11/ab-testing-framework-design-3296257/download
    - http://20bits.com/articles/statistical-analysis-and-ab-testing/.
    - http://elem.com/~btilly/effective-ab-testing/
    - http://www.bingocardcreator.com/abingo/resources
    - http://blog.joshbaker.com/2008/10/20/multivariate-testing-fractional-factorial-full-factorial-taguchi/

(c) 2010 Oliver Cope. For licensing information please see LICENSE.txt
