This repository is the official PyTorch implementation of Meta-Balance. Find the paper on arxiv

Cifar10 dataset is downloaded by the code itself. Both the Severe and Moderate Class Imbalance is simulated by the code as well.

cd MetaCifar

Severely Imbalanced Cifar10 data

python3 train.py 
--dataset_create 
--dataset_type 'severe_imbalance'  

Moderately Imbalanced Cifar10 data

python3 train.py 
--dataset_create 
--dataset_type 'imbalance' 

Need to download the CelebA dataset from this link The Training and Testing splits are further explained in the paper.

cd MetaFace
python3 train.py 
--modify_data 
--modify_gender 'women' 
--proportion 0.1 
--data_train_root '/loc/to/training/data' 
--data_test_root '/loc/to/testing/data' 
--comet_key 'key'

Download the Loan Default datset from this link inside MetaCC.

cd MetaCC 
python3 Meta_credit_card_fraud.py 

Download the Loan Default datset from this link inside MetaLD.

cd MetaLD 
python3 Meta_loan_default.py