Skip to content

Newbeeer/L_DMI

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

53 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Introduction

Implementation for NeurIPS 2019 paper :

paper link: https://arxiv.org/abs/1909.03388

[Slide]

Fashion MNIST dataset

  • To run experiments of Fashion MNIST dataset in fashion directory:
python3 fashion.py --r noisy_amount --s seed --c case_num --device device_num

noise_amount: the amount of noise amount r of label flipping. (0 <= r <= 1)
seed: random seed
case_num :  1: class-independent; 2: class-dependent (a); 3: class-dependent (b)
device_num: GPU number

CIFAR-10 dataset:

  • To run experiments of CIFAR10 dataset in CIFAR-10 directory, pleases run all the baseline in the following order:

    python3 CE.py --r noise_amount --s seed --device device_num --root path --noise-type noise-type
    python3 FW.py --r noise_amount --s seed --device device_num --root path --noise-type noise-type
    python3 GCE.py --r noise_amount --s seed --device device_num --root path --noise-type noise-type
    python3 LCCN.py --r noise_amount --s seed --device device_num --root path --noise-type noise-type
    python3 DMI.py --r noise_amount --s seed --device device_num --root path --noise-type noise-type
    
    noise_amount: the amount of noise amount r of label flipping. (0 <= r <= 1)
    seed: random seed
    device_num: GPU number
    root: path to the CIFAR-10 dataset
    noise-type: class-dependent or class-independent (default:class-dependent)

Dog & Cat datasete

Link to the dataset: https://www.kaggle.com/c/dogs-vs-cats

  • To run experiments of Dog vs. Cats dataset in dogcat directory, pleases run all the baseline in the following order:

    python3 CE.py --r noise_amount --s seed --device device_num
    python3 FW.py --r noise_amount --s seed --device device_num
    python3 GCE.py --r noise_amount --s seed --device device_num
    python3 LCCN.py --r noise_amount --s seed --device device_num
    python3 DMI.py --r noise_amount --s seed --device device_num
    
    noise_amount: the amount of noise amount r of label flipping. (0 <= r <= 1)
    seed: random seed
    device_num: GPU number

MR dataset

  • To download the dataset:

    python3 download_dataset.py MR
  • To run experiments of MR dataset in MR directory:

    python3 main.py --r noise_amount --s seed --device device_num WordCNN
    
    noise_amount: the amount of noise amount r of label flipping. (0 <= r <= 1)
    seed: random seed
    device_num: GPU number

Clothing1M dataset

Link to the dataset: https://drive.google.com/drive/folders/0B67_d0rLRTQYU2E4aHNHaE1uMTg?resourcekey=0-_FShcGYZwIyESjnz6S6aLQ&usp=sharing . We have split the data into clean_test.txt, clean_train. txt,clean_val.txt and noisy_train.txt in our ./clothing directory.

  • To run experiments of Clothing1M dataset in clothing directory, pleases run all the baseline in the following order:

    python3 main.py --device device_num
    
    device_num: GPU number

TODO:

Combine all the APIs to dataset into one file.

About

Code for NeurIPS 2019 Paper, "L_DMI: An Information-theoretic Noise-robust Loss Function"

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published