Explaining Bayesian Optimization by Shapley Values Facilitates Human-AI Collaboration For Exosuit Personalization
This repository contains ShapleyBO, a modular framework for explaining Bayesian optimization by Shapley Values, as introduced in the paper "Explaining Bayesian Optimization by Shapley Values Facilitates Human-AI Collaboration"
- R contains implementation of ShapleyBO
- main-experiments-R provides code to run experiments (section 5) with adjustable settings
- data contains exosuit personalization data used in experiments
- visualization of results via
- R 4.3.2
- R 4.2.0
- R 4.1.6
- R 4.0.3
on
- Linux Ubuntu 20.04
- Linux Debian 10
- Windows 11 Pro Build 22H2
In order to reproduce the papers' key results (and visualizations thereof) follow these steps:
- First and foremost, clone this repository and install all dependencies
- Note that anonymous.4open.science only allows for individual file downloads, which can be quite tedious. To circumvent manual cloning, please use the zip-folder of this repo that we attached to our submission
- Then download the implementations of ShapleyBO from folder named "R"
- Now run main-experiments-R (defaults to settings reported in paper, estimated runtime: 6 CPU hours
Important: Experimental results will be stored automatically as RDS files in home directory. In addition, you can access them as object after completion of the experiments.
Running experiments triggers automatic visualization of results. For customized visualization, please
- read in the saved RDS files and name them according to viz-results.R
- then run viz-results.R with customzed settings
Additional experimental setups can now easily be created by modifying the setup in main-experiments-R
Find data and files to read in data in folder data. In order to preserve anonymity, we do not include a reference to the study for which the data was collected. Details and data source will be made available after the double-blind reviewing process.