Psiturk-Feedback-Learning-with-Contexts

Contains the logic to run a reward feedback-based learning using psiturk

Setup

Files to Modify:

• config.txt: The configuration file for psiturk N.B. there are several variables to edit in this file

  • config_local.txt, config_live.txt: it is convenient to maintain a "live" and a "local" version of the configuration file, to switch between when debugging.
  • The main difference between the the live and local versions is database_url option. In the local version this is set to a SQLite data base (the default is sqlite:///participants.db) while the live version of the file is typically configured with a MySQL database. See the psiturk documentation for more details.

• temp.psiturkconfig: This file will need to be moved to ~/.psiturkconfig with your AMT credentials (See psiturk documentation for more details.)

• /templates/ad.html: Adjust this to advertise your task and explicitly tell the subject how much you will pay them, etc.

• /templates/consent.html: Your online consent form will go here.

Configuration files needed:

• static/game#.json: these are static files that contain predrawn outcomes that control the stimuli and outcomes the subject sees. By Default, exp.html looks for 10 of these files format:
  json file with object: trials, rewards, trial_type

  • trials: The value of trials is an object where the name is a string with the trial number (starting with "0") and the value is array that contains the stimulus, context pair as [S, C], where S is the (integer) stimulus number and C is the (integer) context number (both indexes start with 0)

  for example: the name: value pair
  "23": [0,1]

  would indicate that on the 24th trials, stimulus 0 and context 1 would be shown

  • rewards: The value of rewards is an object where the name is a string with the trial number (starting with "0") and the vaue is an object that contains the response number as a string for the names, and a 0 or 1 for the pre-drawn reward for that action.

  for example: the name: value pair

  "23": {"0": 0, "1": 0, "2": 1, "3": 0}

  would indicate that on the 24th trials, action "2" would be rewarded

  • trial_type: The value of trial_type is an array of strings that indicate the type of the trial for each position in the array (either "test" or "train")

  for example:

  ["train","train","test", (...) ]

• static/images/S#_C#.png: these are static images that are the stimuli. Note! The experiment assumes there are 4 stimuli and 4 contexts (this can be changed in exp.html) (Also, colored shapes are included in this version)

Images needed:

• /static/favicon.ico -- a favicon file
• /static/images/university.png -- logo for the university (displayed in the add and consent form)

Editing the experiment

###Instructions: The experiment is written with the jsPsych toolbox which can be found at http://www.jspsych.org. You will need to edit the experiment file /templates/exp.html to include your instructions (lines 75 through 124). All of the instructions are written in jsPsych compatible variables and need to be formated in html. Each variable (labeled "instruction_survey_block#") corresponds to an indivdual page of instructions. To create another page, create a copy of the variable code:

var game_instructions_4 = {
    type: "text",
    text: "<div id='jspsych-instructions'>" + 
    '<p>Instructions go here in html!".</p>' +
    '<p>[Press ENTER to begin the survey]</p></div>',
    cont_key: 13,
    timing_post_trial: 5
}; 

This new instruction block will have to be added to the experiment. In jsPsych, this can be done by "pushing" the instruction variable to the experiment. The relevant code can be found in line 291-299. If we wanted to add game_instructions_4 to the expriment, we would add the line:

expreriment.push(game_instructions_4);

on line 290, like so:

/* create experiment definition array */
var experiment = [];
experiment.push(welcome_block);
experiment.push(game_instructions);
experiment.push(game_instructions_2); 
experiment.push(game_instructions_3);
experiment.push(game_instructions_4);
experiment.push(banditTask);

Other edits:

There may be other edits you wish to make, including changes to the task timing:

• To change the point-value of wins and loses: change var valueMat = [1, 0] to var valueMat = [W ,L], where W is the point value for correct response and L is the point value for incorrect responses.
• Timing of Presentation: These variables can be found on lines 57-60 and are set in ms, :
  • feedback_display_time: Length of time reward feedback is on screen(default 1500ms),
  • timing_between_response_and_feedback: Length of time delay following response before reward feedback is on screen(default 0ms),
  • timing_iti: Length of inter-trial interval (default 750ms),
  • timing_max_response_time: Maximum response time allowed (default 750ms, set to -1 for no limit),