ajsanjoaquin/Shapley_Valuation
PyTorch reimplementation of computing Shapley values via Truncated Monte Carlo sampling from "What is your data worth? Equitable Valuation of Data" by Amirata Ghorbani and James Zou [ICML 2019]
This tool helps data scientists and machine learning engineers understand the individual contribution of each piece of training data to the overall performance of their neural networks. You provide your trained neural network model and your training/test datasets. The output is a ranking of how important each training data point is to the model's classification accuracy, which can be used to identify harmful data or fairly compensate data providers.
No commits in the last 6 months.
Use this if you need to objectively rank the importance of individual training data points for your neural network, perhaps to clean up your dataset or attribute value.
Not ideal if your model isn't a neural network, if you need to use a performance metric other than classification accuracy, or if you require an integrated retraining step.
Stars
27
Forks
6
Language
Python
License
MIT
Category
Last pushed
Jan 21, 2022
Commits (30d)
0
Get this data via API
curl "https://pt-edge.onrender.com/api/v1/quality/ml-frameworks/ajsanjoaquin/Shapley_Valuation"
Open to everyone — 100 requests/day, no key needed. Get a free key for 1,000/day.
Higher-rated alternatives
shap/shap
A game theoretic approach to explain the output of any machine learning model.
mmschlk/shapiq
Shapley Interactions and Shapley Values for Machine Learning
iancovert/sage
For calculating global feature importance using Shapley values.
predict-idlab/powershap
A power-full Shapley feature selection method.
aerdem4/lofo-importance
Leave One Feature Out Importance