pendulum-sac:
    env: Pendulum-v0
    run: SAC
    stop:
        episode_reward_mean: -300  # note that evaluation perf is higher
        timesteps_total: 10000
    config:
        evaluation_interval: 1  # logged under evaluation/* metric keys
        soft_horizon: True
        clip_actions: False
        normalize_actions: True
        metrics_smoothing_episodes: 5