Skip to content

Running experiments

Andrew Mao edited this page Jun 10, 2013 · 1 revision

Running your experiments is a relatively painless process, but it takes some effort to get the hang of it the first time. The more tricky task is to manage the meta-environment of participation on MTurk.

Config Files

TurkServer uses a standard Java config file for global settings in your experiment that don't change dynamically. An example is included in the code base; specifically, you'll want to put your Amazon API keys in this file, named "turkserver.properties".

You can create a second config file (named anything) to load specific settings for an experiment. This file takes the same settings as the main config file, but overrides anything with the same value.

For all possible settings that can be in these file, check out the source.

Configurator

The Configurator is a class that can assign treatments and can otherwise make dynamic decisions right as experiments are started. It can be used to set up instances of experiments with different settings.

The most common use case of the Configurator is to dynamically assign different treatments to different experiments that start.

Running the Experiment

Once you have all the pieces, you can run your experiment. The MTurk Sandbox is a good place to test your experiment without actually spending money.

To be completed...

Clone this wiki locally