Set of codes for the paper Deep Reinforcement Learning for Portfolio Management: A Simulation Study
- Documentation
- Example usage
- Pre-stored
SimulatorEnv
parameters. - Generalize
SimulatorEnv
to have the same components: a simulator which generates prices, returns, features. => this can then be used to unifystate()
,reset!()
, andenv(action)
. Note: must keep track ofreturns
for rewards such as SR.
- Make agents work for any environment type, not only
GPEnv
-> Solved throught the generalization ofSimulatorEnv
- Add RNG everywhere instead of resetting seeds within functions? -> Unsure
- Gradient clipping for PPO
- CPU/GPU agnosticism
- Recurrent policies
- State space for the environment
- GAIL