(Contextual) multi-armed bandit experiment example

This repository is a simple example of Javascript based experiment for collecting behavioural data. It relies on jsPsych library and a plugin I have developed for that library for multi-armed bandit experiments (can be found at: )

Note, I rely on an older version of jsPsych library (5.03). jsPsych has been changing a lot in recent years, often breaking backward compatibility. I didn't have time to update the plugin yet to align it with the new jsPsych version.

You will need at least some coding skills, basic HTML/Javascript. Even for a basic experiment, most likely you will have to get your hands dirty and do some Javascript coding, adapt CSS etc. I tried to make the plugin fairly general, allowing for options that I did not end up using, but admittedly I have not thoroughly tested everything. If you run into problems, drop me a message, I might be able to help, otherwise feel free to dive into the plugin code.

Since this was a product of my research efforts, I would be grateful if you would cite the article(s) listed below if you end up using the code in your research.

Instructions

There are two examples:

1. Barebones js/exp-simple.js that is automatically loaded if you open index.html in your browser. This is showing a basic version, you would still need to develop a lot of code to make it into a real experiment.
2. A more developed example at js/exp-full.js based on actual contextual bandit experiments I have ran (see references below). It includes some instructions, useful functions for showing performance and earnings reports etc. Running this one is slightly more involved, see below.

To try out exp-full.js first open index.html file, comment line 39 and comment out line 40.

Then you can use a simple Python based server to serve the experiment (this allows loading instructions etc):

With Python 3.x you can use:

python3 -m http.server 8000 --bind 127.0.0.1 

With Python 2.x

python -m SimpleHTTPServer 8000

Then you can navigate your browser to http://localhost:8000/ and it should show the experiment correctly. Read the official documentation for more details on setting up a simple server in Python.

Data storage is the tricky bit and you won't find it here. I deployed the experiment on Amazon Mechanical Turk, hosting the experiment on AWS server and using psiturk library to automatize things. Psiturk has a neat feature that it takes care of the storage, so I didn't have to worry too much about it. Probably it would be useful to provide an example that integrates jsPsych with Psiturk, but it is not a good entry point for someone starting out with online experiments. I might put it in a separate repository in the future, or drop me a message if you really want to jump immediately there.

References

• Hrvoje Stojic, Eric Schulz, Pantelis P. Analytis, & Maarten Speekenbrink. "It's new, but is it good? How generalization and uncertainty guide the exploration of novel options". [PsyArXiv preprint, pdf]
• Hrvoje Stojic, Pantelis P. Analytis & Maarten Speekenbrink (2015). "Human behavior in contextual multi-armed bandit problems". In: Proceedings of the 37th Annual Conference of the Cognitive Science Society. Austin, TX, US: Cognitive Science Society, 2290-2295. [pdf] [data]