Split testing for Django
Python
Pull request Compare This branch is 144 commits ahead, 1 commit behind shimon:master.
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Failed to load latest commit information.
docs
example
splango
tests
.gitignore
AUTHORS
CHANGES
COPYING
LICENSE
MANIFEST.in
README.rst
requirements.txt
requirements_test.txt
runtests.py
setup.py

README.rst

Splango: Drop-in Split Testing for Django

Splango is designed to help you take the first steps with split (A/B) testing with minimal friction. It allows you to instantly declare and run a split test experiment in your templates or in python code, and provides an admin UI for viewing simple funnel reports on the results.

Template Example

{% load splangotags %}

{# first declare the experiment and its variants #} {% experiment "signup_text" variants "control,free,trial" %}

Welcome to my site! Please <a href="/signup">

{# change what is rendered based on which experimental variant you're in #} {% hyp "signup_text" "control" %}

sign up

{% endhyp %}

{% hyp "signup_text" "free" %}
sign up for free

{% endhyp %}

{% hyp "signup_text" "trial" %}
sign up for a trial

{% endhyp %} </a>

Python View Example

def mypage(request):

exp_manager = request.experiments_manager

exp_variant = exp_manager.declare_and_enroll("call_to_action", ["a","b"])

if exp_variant == "a":
call_to_action_label = "try it"
elif exp_variant == "b":
call_to_action_label = "this might not suck"
if request.method == "POST":

form = PleaseDoThisForm(request.POST)

if form.is_valid():
exp_manager.log_goal("pleasedoform.completed") return HttpResponseRedirect(...)
else:
form = PleaseDoThisForm() exp_manager.log_goal("pleasedoform.seen")
return render_to_response("mytemplate.html", {
"call_to_action_label": call_to_action_label }, RequestContext(request))

Things to Note

  • In order to filter out bots, Splango injects a javascript fragment into your HTTP response. Only clients that have a Django session and can run javascript will be tracked in experiments.
  • When a user logs in or registers, any experiment enrollments created while the user was an anonymous Subject will be merged into a Subject associated with the User. In case of conflict, enrollments previously associated with a logged-in Subject will override anonymous enrollments. In other words, Splango tries to be consistent as to what it presented to a particular human, as long as we can identify them.

Installation

  • Ensure you have the dependencies: * django's session package * django's admin for viewing results * jQuery

  • Put the splango directory somewhere in your PYTHON_PATH.

  • In your project's settings.py:

    • add "splango" to INSTALLED_APPS

    • add this to your MIDDLEWARE_CLASSES after the session and auth middleware:

      'splango.middleware.ExperimentsMiddleware'

    • optionally, define a goal to be logged when the first visit to your site is made:

      SPLANGO_FIRST_VISIT_GOAL = "firstvisit"

      If this is defined, splango will automatically log the goal "firstvisit" as being completed on the user's first request.

  • In your urls.py, include the splango urls and admin_urls modules:

    (r'^splango/', include('splango.urls')),

  • Ensure jQuery is available on all text/html responses. Otherwise splango will not work. Splango will remind you of this by putting annoying javascript alert() messages on such pages if settings.DEBUG is true.

  • Finally, go to /splango/admin to create and view experiments.

Usage Notes

  • The names of experiments and goals are their sole identifier. This keeps things simple, but also means that typos can mess things up.
  • Hypotheses within an experiment must have unique names, but you can reuse a hypothesis name (e.g. "control") in multiple experiments if you wish.