| agent: | |
| name: sac | |
| class: agent.sac.SACAgent | |
| params: | |
| obs_dim: ??? # to be specified later | |
| action_dim: ??? # to be specified later | |
| action_range: ??? # to be specified later | |
| device: ${device} | |
| critic_cfg: ${double_q_critic} | |
| actor_cfg: ${diag_gaussian_actor} | |
| discount: 0.99 | |
| init_temperature: 0.1 | |
| alpha_lr: 1e-4 | |
| alpha_betas: [0.9, 0.999] | |
| actor_lr: 1e-4 | |
| actor_betas: [0.9, 0.999] | |
| actor_update_frequency: 1 | |
| critic_lr: 1e-4 | |
| critic_betas: [0.9, 0.999] | |
| critic_tau: 0.005 | |
| critic_target_update_frequency: 2 | |
| batch_size: 1024 # 1024 for Walker, 512 for Meta-world | |
| learnable_temperature: true | |
| double_q_critic: | |
| class: agent.critic.DoubleQCritic | |
| params: | |
| obs_dim: ${agent.params.obs_dim} | |
| action_dim: ${agent.params.action_dim} | |
| hidden_dim: 1024 | |
| hidden_depth: 2 | |
| diag_gaussian_actor: | |
| class: agent.actor.DiagGaussianActor | |
| params: | |
| obs_dim: ${agent.params.obs_dim} | |
| action_dim: ${agent.params.action_dim} | |
| hidden_depth: 2 | |
| hidden_dim: 1024 | |
| log_std_bounds: [-5, 2] | |