An Open-Source Framework for Shapley-based value intended for data valuation.

---

Overview • Installation • How To Use

What's New?

$\text{\color{red}{!in progress}}$

• May 2023: Upload dist to PyPI, pip install opensv

Overview

Shapley-based values are prevalent data valuation approaches, which are attractive for its fair properties (axioms).

The approaches are planned to support including Shapley-based values and some other famous values for data valuation.

Shapley-based values:

• Shapley value
• Beta Shapley value
• KNN Shapley value
• Asymmetric Shapley value
• Robust Shapley value
• Cosine gradient Shapley value
• CS-Shapley value
• Banzhaf value
• Volumn-based Shapley value

Others:

• LOO
• DVRL
• Data-OOB

What Can You Do via OpenSV?

• Use the implementations of current Shapley-based values.* We have implemented various of Shapley-based values and corresponding SOTA computation techniques. You can easily call and understand these methods.
• Design your own data valuation work. With the extensibility of OpenSV, you can quickly practice your data valuation ideas.

Installation

Note: Please use Python 3.10+ for OpenSV

Using Pip

Our repo is tested on Python 3.10+, install OpenSV using pip as follows:

pip install opensv

To play with the latest features, you can also install OpenSV from the source.

Using Git

Clone the repository from github:

git clone https://github.com/ZJU-DIVER/OpenSV.git
cd opensv
pip install -r requirements.txt
python setup.py install

Modify the code

python setup.py develop

Use OpenSV

TBD

Base Concepts

TBD