ray/rllib/tuned_examples/pendulum-td3.yaml

# This configuration can expect to reach -160 reward in 10k-20k timesteps
pendulum-ddpg:
    env: Pendulum-v0
    run: TD3
    stop:
        episode_reward_mean: -130
        time_total_s: 900 # 10 minutes
    config:
        # === Model ===
        actor_hiddens: [64, 64]
        critic_hiddens: [64, 64]

        # === Exploration ===
        learning_starts: 5000
        pure_exploration_steps: 5000

        # === Evaluation ===
        evaluation_interval: 1
        evaluation_num_episodes: 5
Enable Twin Delayed DDPG for RLlib DDPG agent (#3353) 2018-11-22 12:03:20 +08:00			`# This configuration can expect to reach -160 reward in 10k-20k timesteps`
			`pendulum-ddpg:`
			`env: Pendulum-v0`
[rllib] TD3/DDPG improvements and MuJoCo benchmarks (#4694) * [rllib] Separate optimisers for DDPG actor & crit. * [rllib] Better names for DDPG variables & options Config changes: - noise_scale -> exploration_ou_noise_scale - exploration_theta -> exploration_ou_theta - exploration_sigma -> exploration_ou_sigma - act_noise -> exploration_gaussian_sigma - noise_clip -> target_noise_clip * [rllib] Make DDPG less class-y Used functions to replace three classes with only an __init__ method & a handful of unrelated attributes. * [rllib] Refactor DDPG noise * [rllib] Unify DDPG exploration annealing Added option "exploration_should_anneal" to enable linear annealing of exploration noise. By default this is off, for consistency with DDPG & TD3 papers. Also renamed "exploration_final_eps" to "exploration_final_scale" (that name seems to have been carried over from DQN, and doesn't really make sense here). Finally, tried to rename "eps" to "noise_scale" wherever possible. 2019-04-26 17:49:53 -07:00			`run: TD3`
Enable Twin Delayed DDPG for RLlib DDPG agent (#3353) 2018-11-22 12:03:20 +08:00			`stop:`
[rllib] TD3/DDPG improvements and MuJoCo benchmarks (#4694) * [rllib] Separate optimisers for DDPG actor & crit. * [rllib] Better names for DDPG variables & options Config changes: - noise_scale -> exploration_ou_noise_scale - exploration_theta -> exploration_ou_theta - exploration_sigma -> exploration_ou_sigma - act_noise -> exploration_gaussian_sigma - noise_clip -> target_noise_clip * [rllib] Make DDPG less class-y Used functions to replace three classes with only an __init__ method & a handful of unrelated attributes. * [rllib] Refactor DDPG noise * [rllib] Unify DDPG exploration annealing Added option "exploration_should_anneal" to enable linear annealing of exploration noise. By default this is off, for consistency with DDPG & TD3 papers. Also renamed "exploration_final_eps" to "exploration_final_scale" (that name seems to have been carried over from DQN, and doesn't really make sense here). Finally, tried to rename "eps" to "noise_scale" wherever possible. 2019-04-26 17:49:53 -07:00			`episode_reward_mean: -130`
			`time_total_s: 900 # 10 minutes`
Enable Twin Delayed DDPG for RLlib DDPG agent (#3353) 2018-11-22 12:03:20 +08:00			`config:`
			`# === Model ===`
			`actor_hiddens: [64, 64]`
			`critic_hiddens: [64, 64]`

			`# === Exploration ===`
[rllib] TD3/DDPG improvements and MuJoCo benchmarks (#4694) * [rllib] Separate optimisers for DDPG actor & crit. * [rllib] Better names for DDPG variables & options Config changes: - noise_scale -> exploration_ou_noise_scale - exploration_theta -> exploration_ou_theta - exploration_sigma -> exploration_ou_sigma - act_noise -> exploration_gaussian_sigma - noise_clip -> target_noise_clip * [rllib] Make DDPG less class-y Used functions to replace three classes with only an __init__ method & a handful of unrelated attributes. * [rllib] Refactor DDPG noise * [rllib] Unify DDPG exploration annealing Added option "exploration_should_anneal" to enable linear annealing of exploration noise. By default this is off, for consistency with DDPG & TD3 papers. Also renamed "exploration_final_eps" to "exploration_final_scale" (that name seems to have been carried over from DQN, and doesn't really make sense here). Finally, tried to rename "eps" to "noise_scale" wherever possible. 2019-04-26 17:49:53 -07:00			`learning_starts: 5000`
			`pure_exploration_steps: 5000`
Enable Twin Delayed DDPG for RLlib DDPG agent (#3353) 2018-11-22 12:03:20 +08:00
[rllib] TD3/DDPG improvements and MuJoCo benchmarks (#4694) * [rllib] Separate optimisers for DDPG actor & crit. * [rllib] Better names for DDPG variables & options Config changes: - noise_scale -> exploration_ou_noise_scale - exploration_theta -> exploration_ou_theta - exploration_sigma -> exploration_ou_sigma - act_noise -> exploration_gaussian_sigma - noise_clip -> target_noise_clip * [rllib] Make DDPG less class-y Used functions to replace three classes with only an __init__ method & a handful of unrelated attributes. * [rllib] Refactor DDPG noise * [rllib] Unify DDPG exploration annealing Added option "exploration_should_anneal" to enable linear annealing of exploration noise. By default this is off, for consistency with DDPG & TD3 papers. Also renamed "exploration_final_eps" to "exploration_final_scale" (that name seems to have been carried over from DQN, and doesn't really make sense here). Finally, tried to rename "eps" to "noise_scale" wherever possible. 2019-04-26 17:49:53 -07:00			`# === Evaluation ===`
			`evaluation_interval: 1`
			`evaluation_num_episodes: 5`