Preferential Bayesian Optimization of retinal prosthetic stimulation

Requirements:

This software accompanies the paper : Fauvel T. and Chalk M, 2021, Human-in-the-loop optimization of retinal prostheses encoders. If you use this software, please reference it by citing this paper.

• MATLAB

• Psychtoolbox (Optional) http://psychtoolbox.org/download.html

• Python 3

• Pulse2percept, by Beyeler et al (2017):

  • You can either choose to work with the stable version (0.7.1) of pulse2percept or the latest one (0.8.0).
  • To use the stable release, use pip3 install pulse2percept
  • To install the latest release, use pip3 install git+https://github.com/pulse2percept/pulse2percept

• BO_toolbox : Bayesian Optimization toolbox based on the GP_toolbox, https://github.com/TristanFauvel/BO_toolbox

• GP_toolbox : Gaussian Process regression, classification, and preference learning, https://github.com/TristanFauvel/GP_toolbox

• In the pulse2percept package installation folder, replace beyeler2019.py in /envs/pulse2percept-env/lib/python3.8/site-packages/pulse2percept/models with the file beyeler2019 in /p2p_modifications (if you use the stable p2p release) or by beyeler2019_modified_for_latest_p2p (if you use the latest release, and rename the file in beyeler2019).

• Edit the file 'add_modules' according to the locations of the different folders on your computer.

• To run an experiment : open('to_run_BO_experiment')

  • If you want to use Psychtoolbox : set use_ptb3 = 1
  • If you do not want to use Psychtoolbox : set use_ptb3 = 0

Run and analyze experiments:

• Activate the Python environment in which pulse2percept is installed: conda activate pulse2percept_env
• Launch Matlab with Psychtoolbox : ptb3-matlab
• To launch experiments: to_run_BO_experiment.m
• To analyze experiments: analysis_pipeline.m
• Note that the first time you run vision tests, likelihoods used in the QUEST+ procedure will be computed, which can take a while (QPlikelihoods_E_VA.mat and QPlikelihoods_Snellen_VA.mat).

Organization :

* `Experiment`: scripts and functions to run the experiment described in the paper.
* `p2p_modifications` : contains modified p2p files required in order to run the experiment.
* `p2p_analysis`: code to perform the p2p analysis leading to the LN approximation described in the paper. Note that this analysis was performed on pulse2percept 0.5.0.
* `QuestPlus-master` : a Matlab implementation of the QUEST+ adaptive psychometric method, by Pete R Jones.

• Stimuli : stimuli images used in the experiment

pulse2percept analysis:

• The first thing to do is to compute M, the matrix of electrodes' projective fields: to do so, run to_compute_M.py.
• To compare the output of the LN approximation and the original p2p model with random currents, use WN_electrodes.py.
• To perform a visua comparison between the output of p2p and the LN approximation (in particular, for optimized stimuli) use optimized_code_stimuli.py.