This code implements a Two-Armed Bandit task for studying human reinforcement leanring. It was written based on the tutorial of Uri Hertz on Online Experiments Development (http://www.urihertz.net/projects.html). It uses text configuration files, so you don't need to write javascript code (to some extent). Configuration files should be put under the folder "config/TASK-NAME/" Examples of configuration files are provided with the code. You can have many configuration folders. This way, you can configure as many different tasks as you want.
Things that can be modified through the configuration files:
- Global parameters (see below) (task.cfg) - Conditions (task.cond) - Displayed Information text (task.info) - Displayed consent text (task.cons) - Displayed instructions (task.inst) - Displayed postlearning text (task.post) - Questionnaires description (task.quest_info) - Questionnaires (task.quest_items)
Global parameters (task.cfg) include:
- Language (english "en" or french "fr") - Whether or not a completion link is included (0/1) - Number of sessions N - Whether or not to include a postlearning session (0/1) - Whether or not to include a questionnaire (0/1) - Maximum number of training sessions N - Number of training trials N - Number of trials per condition N - Number of different stimuli per condition N - Whether or not trials are interleaved (0/1)
Conditions (task.cond) are specified, with five parameters, one condition per line:
- P1: probability of outcome for option 1 [0 1] - P2: probability of outcome for option 2 [0 1] - Mag: magnitude of the outcome N>0 - Val: valence of the outcome. 1: positive or zero, -1: negative or zero, 0: symmetric +1/-1 - Info: partial or complete outcome (0/1)
Questionnaires can be specified in task.quest_items as following:
#[NAME OF QUESTIONNAIRE 1]-[NUMBER OF ITEMS] 1 $ QUESTION1 $ ANSWER1:SCORE1 $ ANSWER2:SCORE2 $ ANSWER3:SCORE3 2 $ QUESTION2 $ ANSWER1:SCORE1 $ ANSWER2:SCORE2 $ ANSWER3:SCORE3
#[NAME OF QUESTIONNAIRE 2]-[NUMBER OF ITEMS] 1 $ QUESTION1 $ ANSWER1:SCORE1 $ ANSWER2:SCORE2 $ ANSWER3:SCORE3 2 $ QUESTION2 $ ANSWER1:SCORE1 $ ANSWER2:SCORE2 $ ANSWER3:SCORE3
Make sure you don't include these characters in your text '#$:-' as they are used for parsing questionnaires. You can also use "skiptests" to skip some questionnaires. For example, if a person doesn't smoke, it is not necessary to ask her about her smoking habits. Skiptests are defined as following:
#[NAME OF NEXT QUESTIONNAIRE]skiptest-1-[ID OF THE ANSWER FOR THE TEST TO BE TRUE]-[NUMBER OF ITEMS IN THE NEXT QUESTIONNAIRE] 1 $ QUESTION $ ANSWER1:SCORE1 $ ANSWER2:SCORE2
Task images must be put under the folder "images/IMAGE-FOLDER-NAME". Stimuli images under "images/IMAGE-FOLDER-NAME/stim" and outcomes under "images/IMAGE-FOLDER-NAME/outcome".
Once a task is specified through configuration files, it can be generated using the command
python generate.py TASK-NAME IMAGE-FOLDER-NAME
This will generate a javascript file "RLTask.js", which will be the main code for running the task. Example: to generate a task called task1 using the images in cards_gif, you need to run the command "python generate.py task1 cards_gif".
To run the experiment, you need to configure a server (and a mysql database) in which you put the generated code. You will find a template for an empty database in "sql/empty_db.sql". You need to configure your database account in the file "connectDB.php".
In addition to the mysql database, experiment logs are recorded under the "/log" folder. Log files are produced as a backup for eventual missing data from the database (for example due to connexion problems). They contain all task information, including trial by trials data. So, if you don't want to setup a database, you can extract data by parsing log files.
The code is also compatible with mobile devices, so experiments can be run on tablets and smartphones. Here is an example of a task that you can generate http://human-rl.scicog.fr/exp/
Enjoy!