If you find our research helpful, please cite our paper:

@article{xu2025mining,
  title={Mining Intraday Risk Factor Collections via Hierarchical Reinforcement Learning based on Transferred Options},
  author={Xu, Wenyan and Chen, Jiayu and Li, Chen and Hu, Yonghong and Lu, Zhonghua},
  journal={arXiv preprint arXiv:2501.07274},
  year={2025}
}

HRL-HF_risk_factor_set

1 How to config the environments:

• on Ubuntu 20.04.5
• python 3.8
• numpy 1.24.2
• pandas 1.1.5
• pytorch 1.7.1
• tensorboard 2.14.0
• scipy 1.10.1

2 Prepare you training data:

• For the above HRL/RL model use .csv file which contains open,high,low,close,volume,vwap,target as input. The dataset is of two types i.e. arg.add_arg("env_name", "hs300", "Environment name, can be [hs300, sp500,nifty100,csi800]").
• There are three ways to calculate rewards (a measure of the correlation between mathematical expressions of generated High Frequency Risk Factors and targets) in class Risk_Factor_Env: IC*, RankIC*, and IR*.

3 How to train model:

3.1 HPPO-TO:

python run_main_hppo_to.py

3.2 HPPO/DAC:

python run_main_hrl.py

3.3 PPO:

python run_main_ppo.py