OCAN： One-Class Adversarial Nets fo Fraud Detection [pytorch Implementation]

Reference:

official implementation
paper link

Running Environment

python 3.7.1
pytorch 1.0.1

Guideline

• For help: python main.py -h

• For wiki data:

  1.modify net_cfg['g_cfg']['output_dim']=200; net_cfg['g_cfg']['d_cfg']['input_dim']=200;in config.json

  2.run python main.py ./data/hidden/wiki/ben_hid_emd_4_50_8_200_r0.npy ./data/hidden/wiki/val_hid_emd_4_50_8_200_r0.npy

• For credit data: just modify the same configuration consistent with the data dimension. credit_card: 50, raw_credit_card: 30

• For your own data: You can customize the net structure and training hyperparameters of OCAN in config.json

Welcome for bugs and issues reporting