Easy split testing for Django!
Stop guessing and start testing! With django-easy-split, you'll be able to scientifically prove which aspects of your site convert better than others. Works great for testing copy, designs, styles, layouts, assets, and even pricing and feature sets.
With just a few simple template tags, you'll be able to generate detailed reports about how visitors respond to various experimental aspects of your site.
- Simple installation and usage
- Statistical report generation using chi-squared tests (with optinal support for numypy/scipy)
- Automatic bot exclusion
Getting started with django-easy-split is easy!
Just install it, define your content to be tested and your goals, then sit back, wait for the data to roll in, then finally analyze and view your reports. You can do pretty much everything right from your templates, there's no need for you to ever create your data models manually.
-
Install django-easy-split
pip install django-easy-split
-
Add 'easy_split' to your INSTALLED_APPS setting like this:
INSTALLED_APPS = ( ... 'easy_split', )
-
Sync your DB:
python manage.py syncdb
-
Add easy_split to your URLs:
urlpatterns = patterns('', url(r'^admin/doc/', include('django.contrib.admindocs.urls')), url(r'^admin/split/', include('easy_split.admin_urls')), url(r'^admin/', include(admin.site.urls)), url(r'^split/', include('easy_split.urls')),
-
Assign your visitors to experimental groups by wrapping your views with this decorator:
from easy_split.decorators import set_experiment_user @set_experiment_user def landing_page(request): return render_to_response('split_me.html', {}, context_instance=RequestContext(request))
-
Include the necessary bot-excluding JavaScript in your base HTML. No jQuery is required!:
{% load easy_split %} <head> {% split_js %} </head> -
Define your test content:
{% split try_or_buy control %} <a href="/payment/" class="btn btn-large btn-primary"> Buy now! </a> {% endsplit %} {% split try_or_buy test %} <a href="/payment/" class="btn btn-large btn-primary"> Try it free! </a> {% endsplit %} -
Set up your goals. The easiest way to do it is with this simple tag put on your goal page (ex, on the post-payment page). You can also define them programatically in your views.
{% goal "try_or_buy" %} -
Test it out! Make sure that all of your URLs work and that GoalRecords are being created in your database.
-
Finally, once enough data has been collected over time, run:
python manage.py update_experiment_reports
to generate your reports, then browse over to /admin/split/ to see your reports!
You can also update specific reports by name:
python manage.py update_experiment_reports experiment_name
django-easy-split can also be used directly in your Django views rather than in your templates. This is useful if you're testing whole features, and not just copy or style changes.
There are two important bits here: testing users and recording results.
To split and route your experimental users,
from easy_split.models import Experiment
from easy_split.utils import WebUser
from easy_split.decorators import set_experiment_user
@set_experiment_user
def landing_page(request):
if Experiment.test("my_experiment_name", WebUser(request)):
return render_to_response('new_landing.html', {}, context_instance=RequestContext(request))
else:
return render_to_response('landing.html', {}, context_instance=RequestContext(request))Then, you'll want to record the records of events which pass the test:
from easy_split.models import GoalRecord
from easy_split.utils import WebUser
def conversion_page(request):
GoalRecord.record("my_experiment_name", WebUser(request))
return render_to_response('success.html', {}, context_instance=RequestContext(request))And that's it! django-easy-split is smart enough to create your Goal obejcts for you.
This project is a direct descendant of django-lean. For now, it's probably best that you swim upstream for whatever you are looking for. This package also contains changes mentioned in Alex Kehayias's blog post on the subject.
django-easy-split intends to update and simplify django-lean. This may or may not ever actually come to fruition as a useful product for the community - it may remain simply a mess of undocumented code only really useful for my own purposes. Still, I'll be working in public for the most part, so we shall see.
- Documentation, Packaging, Examples, Tests
- Email reporting?
- Automated, integrated testing queues.