1234567891011121314151617181920212223242526272829303132333435363738 |
- # Runs on a single g3.4xl node
- # See https://github.com/ray-project/rl-experiments for results
- dueling-ddqn:
- env:
- grid_search:
- - ALE/Breakout-v5
- - ALE/BeamRider-v5
- - ALE/Qbert-v5
- - ALE/SpaceInvaders-v5
- run: DQN
- config:
- # Works for both torch and tf.
- framework: torch
- # Make analogous to old v4 + NoFrameskip.
- env_config:
- frameskip: 1
- full_action_space: false
- repeat_action_probability: 0.0
- double_q: true
- dueling: true
- num_atoms: 1
- noisy: false
- replay_buffer_config:
- type: MultiAgentReplayBuffer
- capacity: 1000000
- num_steps_sampled_before_learning_starts: 20000
- n_step: 1
- target_network_update_freq: 8000
- lr: .0000625
- adam_epsilon: .00015
- hiddens: [512]
- rollout_fragment_length: 4
- train_batch_size: 32
- exploration_config:
- epsilon_timesteps: 200000
- final_epsilon: 0.01
- num_gpus: 0.2
- min_sample_timesteps_per_iteration: 10000
|