#A/B Testing Tool

Motivation

Harvard is a leader in both research and higher education. As Harvard changes academic platforms, there are new opportunities to allow course staff and education researchers access to data about how these platforms are used. Data about learning can inform teaching to improve learning outcomes and other success metrics. This enables a cycle of continuous, iterative improvement.

Faculty asked, and we listened. A/B testing can help identify changes in courses that impact the success of students.

About the tool

The A/B Testing Tool complements existing features within Canvas that provide data about how the platform is used. An LTI optimized for Canvas, the A/B Testing Tool creates experiments that:

• divide course enrollment into sets of users that are part of a track unique to the course

• apply interventions by displaying different content to each track at intervention points throughout the course

For more information, take a look at these slides

---

In order to run locally:

Note: In order to run a command from the ab-testing-tool directory, open terminal and go to the ab-testing tool using something like cd ab-testing-tool

• Run sudo pip install -r ab_testing_tool/requirements/local.txt from the ab-testing-tool directory (don't use sudo if using cygwin or virtualenv). If installing on OSX and you encountering the error clang error: unknown argument: '-mno-fused-madd' when installing lxml, refer to this: http://stackoverflow.com/a/22322645/2812260

• Create an access token in Canvas (click on Settings in top right of screen, look for 'New Access Token' button under 'Approved Integrations' header) and copy this value (token needs to be a string i.e. in quotes)

• Run cp ab_testing_tool/settings/secure.py.example ab_testing_tool/settings/secure.py from the ab-testing-tool directory

• Edit the new ab_testing_tool/settings/secure.py and fill in values requested there. The minimum needed for local setup is as follows: Modify the line referring to the course_oauth_token, adding the access token from Canvas like this: "course_oauth_token": "asdlkjf234aADKUEJskjdf2l3a6k7sjdf",.

• Run python manage.py syncdb from the ab-testing-tool directory

• Run python manage.py runserver from the ab-testing-tool directory

• Install app in canvas by adding an external tool by XML. The key and secret will be "test" and "secret" respectively. Either copy the config from http://localhost:8000/ab-testing/lti/tool_config or use the following:

<?xml version="1.0" encoding="UTF-8"?>
<cartridge_basiclti_link xmlns:lticm="http://www.imsglobal.org/xsd/imslticm_v1p0"
    xmlns:blti="http://www.imsglobal.org/xsd/imsbasiclti_v1p0"
    xmlns:lticp="http://www.imsglobal.org/xsd/imslticp_v1p0"
    xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance"
    xmlns="http://www.imsglobal.org/xsd/imslticc_v1p0"
    xsi:schemaLocation="http://www.imsglobal.org/xsd/imslticc_v1p0 http://www.imsglobal.org/xsd/lti/ltiv1p0/imslticc_v1p0.xsd
        http://www.imsglobal.org/xsd/imsbasiclti_v1p0
        http://www.imsglobal.org/xsd/lti/ltiv1p0/imsbasiclti_v1p0p1.xsd
        http://www.imsglobal.org/xsd/imslticm_v1p0
        http://www.imsglobal.org/xsd/lti/ltiv1p0/imslticm_v1p0.xsd
        http://www.imsglobal.org/xsd/imslticp_v1p0
        http://www.imsglobal.org/xsd/lti/ltiv1p0/imslticp_v1p0.xsd">
    <blti:title>A/B Testing Tool</blti:title>
    <blti:description>Tool to allow students in a course to get different content in a module item.</blti:description>
    <blti:launch_url>http://localhost:8000/ab-testing/lti/</blti:launch_url>
    <blti:secure_launch_url>http://localhost:8000/ab-testing/lti/</blti:secure_launch_url>
    <blti:vendor/>
    <blti:extensions platform="canvas.instructure.com">
        <lticm:property name="tool_id">ab_testing_tool</lticm:property>
        <lticm:property name="privacy_level">public</lticm:property>
        <lticm:options name="resource_selection">
            <lticm:property name="url">http://localhost:8000/ab-testing/lti/resource_selection</lticm:property>
            <lticm:property name="enabled">true</lticm:property>
        </lticm:options>
        <lticm:property name="selection_width">800</lticm:property>
        <lticm:property name="selection_height">800</lticm:property>
        <lticm:options name="course_navigation">
            <lticm:property name="text">A/B Testing Tool</lticm:property>
            <lticm:property name="enabled">true</lticm:property>
            <lticm:property name="visibility">admins</lticm:property>
        </lticm:options>
    </blti:extensions>
</cartridge_basiclti_link>

Updating locally:

All commands here should be run from the ab-testing-tool directory

• Update the code via git pull.

• Update the database with python manage.py migrate

• If you get an error during migrations, delete your local database and create a new one with rm ab_testing_tool/db.sqlite3 and then python manage.py migrate.

• Run the tool via python manage.py runserver.

• If you get an import error, update requirements with sudo pip install -r ab_testing_tool/requirements/local.txt (don't use sudo if using cygwin or virtualenv).

• If a library has changed versions and secure.py needs updating, the library update can be done via [sudo] pip install -r ab_testing_tool/requirements/base.txt --upgrade

Running Tests:

• Backend (Python) tests can be ran with command: python manage.py test

• Frontend (Javascript) tests are written in Jasmine syntax and ran by Karma as a test runner to interact with Phantom.js, a headless webkit, as our browser mock. In order to be able to run these tests, you need to:

  • install Node.js and npm
  • install Node dependencies located in package.json with command sudo npm install,
  • install Karma Command Line interface globally with command sudo npm install -g karma-cli, and finally
  • if necessary, install a hidden dependency for Phantom.js with sudo apt-get install libfontconfig (e.g. on Ubuntu)

  You can now run frontend tests with command: karma start karma.conf.js (configuration file is karma.conf.js)