Implementation of Schmidhuber's Upside Down Reinforcement Learning paper

Link to paper with theory: https://arxiv.org/pdf/1912.02875.pdf

Link to paper with implementation details and results: https://arxiv.org/pdf/1912.02877.pdf

Use as you wish. Tweet(@mfharoon)/email([email protected]) me any interesting results you find and sets of hyperparameters that work for particular environments. I will share here. Thanks!

replay_size = 600
last_few = 50
batch_size = 64
n_warm_up_episodes = 50
n_episodes_per_iter = 50
n_updates_per_iter = 100
command_scale = 0.02
lr = 0.001