Key hyperparameters for DRL algorithms
| Algorithm | Learning rate | n_steps/batch | Buffer size | Notes |
|---|---|---|---|---|
| A2C | 1 × 10–4 | 10/– | – | GAE 0.95 |
| PPO | 5 × 10–4 | 1,024/64 | – | entropy coef 0.01 |
| SAC | 3 × 10–4 | –/256 | 106 | ent_coef auto 0.2, log std −3 |
| DDPG | 3 × 10–4 | –/256 | 106 | noise σ = 0.1 |
| TD3 | 1 × 10–3 | –/256 | 106 | noise σ = 0.1 |
| TQC | 3 × 10–4 | –/512 | 106 | ent_coef auto 0.1, quantile trunc. 25% |
| RecurrentPPO | 3 × 10–4 | 1,024/128 | – | LSTM size 128, 1 layer |
| Algorithm | Learning rate | n_steps/batch | Buffer size | Notes |
|---|---|---|---|---|
| A2C | 1 × 10–4 | 10/– | – | GAE 0.95 |
| PPO | 5 × 10–4 | 1,024/64 | – | entropy coef 0.01 |
| SAC | 3 × 10–4 | –/256 | 106 | ent_coef auto 0.2, log std −3 |
| DDPG | 3 × 10–4 | –/256 | 106 | noise |
| TD3 | 1 × 10–3 | –/256 | 106 | noise |
| TQC | 3 × 10–4 | –/512 | 106 | ent_coef auto 0.1, quantile trunc. 25% |
| RecurrentPPO | 3 × 10–4 | 1,024/128 | – | LSTM size 128, 1 layer |
Note(s): GAE: generalized advantage estimation
Sharing content requires targeting cookies to be enabled. Please update your cookie preferences to use this feature.