Qitong Gao, Dong Wang, Joshua David Amason, Siyang Yuan, Chenyang Tao, Ricardo Henao, Majda Hadziahmetovic, Lawrence Carin, Miroslav Pajic
Paper can be found at https://openreview.net/forum?id=fXHl76nO2AZ. Accepted to ICLR '22.
Contact: qitong.gao@duke.edu
ATTENTION
Some of the data and checkpoints we uploaded require to be downloaded with Git Large File Storage, i.e., git-lfs.
To install git-lfs, follow the instructions on https://github.com/git-lfs/git-lfs.
Once it is installed, make sure to clone this repository by running
git lfs clone https://github.com/gaoqitong/gradient-importance-learning.git
or
git lfs clone git@github.com:gaoqitong/gradient-importance-learning.git
This code package is tested against the following environmental setup:
Python 3.7
tensorflow 1.15.0
scikit-learn 0.24.2
pandas 1.2.4
numpy 1.20.2
scipy 1.7.0
Here we provided the code for training and evaluating GIL/GIL-D using multivariate tabular and sequential data. Each folder is self-contained and has a seperate readme file introducing how to train, evaluate and load pre-trained checkpoints.
If you find our work and code useful, please consider cite the paper
@inproceedings{
gao2022gradient,
title={Gradient Importance Learning for Incomplete Observations},
author={Qitong Gao and Dong Wang and Joshua David Amason and Siyang Yuan and Chenyang Tao and Ricardo Henao and Majda Hadziahmetovic and Lawrence Carin and Miroslav Pajic},
booktitle={International Conference on Learning Representations},
year={2022},
url={https://openreview.net/forum?id=fXHl76nO2AZ}
}