fritz449 / srlf

Simple Reinforcement Learning Framework

SRLF

Simple Reinforcement Learning Framework

---

SRLF helps to set up and train agent using one of these five algorithms:

1. Evolution strategies (paper)
2. Deep Deterministic Policy Gradient (paper)
3. TRPO (paper)
4. Rainbow (paper)
5. A3C (paper)

I didn't make very user-friendly interface, but if you know how the algorigthm you want to use works, you can do it more or less easily.

To use, follow these steps:

1. Choose algorithm you want to use
2. Be sure you know how it works
3. Study this implementation and how hyperparameters are used
4. Use prepare_config.py as an example to make your config. Also check helpers/utils.py to know how to set some of non-obvious hyperparameters (like trainer)
5. Launch python3 prepare_config.py to make config-file
6. Launch python3 run_experiment.py to run experiment (you should also check how run_experiment.py works)
7. Enjoy!

I also hope that framework can be helpful for people who want to study how to implement some of 5 algorithms that can be found here.