Skip to content

vinkels/RL_visual_task

Repository files navigation

RFL task for image complexity preference

Visual decision task with reinforcement learning using images categorized by SC/CE scores. task consists of three experimental phases:

  • control phase
  • learning phase
  • testing phase Control and testing phase are the same but with different image sets.

To Do

  • clean up code
  • data analysis

Getting Started

All code is designed to run and tested in python 2.7 and 3.6. To be able to run the code you need all libraries mentioned in prerequisites.To run the code navigate to the directory in your terminal and run main.py (python main.py).

Prerequisites

All code is tested in python3.6, but experiment scripts are also python 2.7 compatible. All external libraries needed can be found in requirements.txt. To install required libraries run the following line in terminal/command line. Read more about pip here.

pip install requirements-txt

project structure

main.py is the execution script for the experiment. Here you can adjust parameters like participant number, reward scheme and keybinding for the learning phase. All scripts used for preprocessing of images are located in the img_prep folder. These scripts are not always python2.7 compatible and are not needed to run the experiment. In the experiment folder you can find scripts session.py that contains the actual experiment and img_sets.py that creates randomized images lists for every session.

Running

Download complete repository and navigate to repository in terminal. Run main.py in python 2 or 3 to run the experiment. Make sure capslock is off, otherwise globalKeys will not work

python main.py

Authors

  • Yannick Vinkesteijn - Initial work - Github

License

This project is licensed under the MIT License - see the LICENSE.md file for details

Acknowledgments

  • Sara Jahfari
  • Noor Seijdel

About

Visual decision task with reinforcement learning

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published