Code for the experiment in the paper "Distributionally Robust Coreset Selection under Covariate Shift"
Authors: Tomonari Tanaka, Hiroyuki Hanada, Hanting Yang, Aoyama Tatsuya, Yu Inatsu, Akahane Satoshi, Yoshito Okura, Noriaki Hashimoto, Taro Murayama, Hanju Lee, Shinya Kojima, Ichiro Takeuchi
Transactions on Machine Learning Research, to appear
URL: https://openreview.net/forum?id=Eu7XMLJqsC
Programs are implemented in Python.
The following packages can be installed via pip or conda:
numpyscipyscikit-learnmatplotlibcvxpypytorch(required only when we specifycifar10as a dataset)torchvision(required only when we specifycifar10as a dataset)
The following package can be installed by compiling with C++ compiler (gcc, g++ and make are required) and then via pip:
liblinear-weighted-kernelized(a modification of LIBLINEAR with weights for data instances for kernel models); seesourcefolder and the fileREADME.mdin it for installation.
Run the followings:
./experiment_drcs.sh: Model performance of DRCS
All programs are written by the authors.
Dataset files are retrieved from "LIBSVM Data" by Rong-En Fan (National Taiwan University). https://www.csie.ntu.edu.tw/~cjlin/libsvmtools/datasets/