Ref: https://github.com/JuliaReinforcementLearning/ReinforcementLearning.jl/issues/257#issuecomment-831710404