Code for "R-divergence for Estimating Model-oriented Distribution Discrepancy"

HIGGS

Get the HIGGS data from here - https://drive.google.com/file/d/1sHIIFCoHbauk6Mkb6e8a_tp1qnvuUOCc/view

To run HIGGS experiment, choose --n = 500,1000,1500,2500,4000,5000 and run

python higgs.py

MNIST

Get the fake MNIST data from here - https://drive.google.com/file/d/13JpGbp7PEm4PfZ6VeqpFiy0lHfVpy5Z5/view

To run MNIST experiment, choose --n = 100,200,300,400,500 and run

python mnist.py

Blob

To run Blob experiment with KDE

python blob_kde.py --exptype power

To run Blob experiment with GMM

python blob_gmm.py --exptype power

HDGM

To run HDGM experiment, choose --vtype = vjs, --n = 100,1000,1500,2500, and --d = 3,5,10,15,20 and run

python hdgm.py --exptype power

Case Study: learning with noisy labels

To run CIFAR10 experiment with symmetry flipping labels

python CIFAR10.py --noise_type 'symmetric'

To run CIFAR10 experiment with pair flipping labels

python CIFAR10.py --noise_type 'pairflip'