Optimistic Posterior Sampling for Reinforcement Learning with Few Samples and Tight Guarantees

Official implementation of OPSRL algorithm and baselines from the paper D.Tiapkin et al. "Optimistic Posterior Sampling for Reinforcement Learning with Few Samples and Tight Guarantees". The algorithms are implemented in the folder algorithms/, the parameters are contained in the folder config\.

Requirements:

Python 3.8
rlberry 0.2.1

Running experiment opsrl_vs_baselines and generate the plots

    python run.py config/experiments/opsrl_vs_baselines.yaml
    python plot_opsrl_vs_baselines.py

Running experiment opsrl_samples and generate the plots

    python run.py config/experiments/opsrl_samples.yaml
    python plot_opsrl_samples.py

Running experiment opsrl_prior and generate the plots

    python run.py config/experiments/opsrl_prior.yaml
    python plot_opsrl_prior.py

d-tiapkin/optimistic-psrl-experiments

Optimistic Posterior Sampling for Reinforcement Learning with Few Samples and Tight Guarantees