Using Schrodinger Bridge for Model-based Offline RL

• train_stochastic.py can be viewed as Main.py

• train_eval2, train_schrodinger_bridge.py are decrepited

• the environment is in requirements.txt, you also have to download and set up mujoco

• you can directly debug train_stochastic to see how the experiment run, you don't need any inputs from command line while debugging

• the whole learned-model part is done by conditional schrodinger bridge implemented in [CSBI github](MSML/papers/Conditional_Schrodinger_Bridge_Imputation/configs/default_pm25_config.py at main · morganstanley/MSML (github.com)) from paper Provably Convergent Schrödinger Bridge with Applications to Probabilistic Time Series Imputation (arxiv.org)