selected-models/phase_a_training_config.json
| 1 | { |
| 2 | "experiment_name": "ppo_lunarlander_flip_phase_a", |
| 3 | "env_id": "LunarLander-v3", |
| 4 | "environment_factory": "make_flip_acquisition_lander", |
| 5 | "env_kwargs": {}, |
| 6 | "total_timesteps": 10000000, |
| 7 | "n_envs": 16, |
| 8 | "seed": 42, |
| 9 | "actor_layers": [ |
| 10 | 128, |
| 11 | 128 |
| 12 | ], |
| 13 | "critic_layers": [ |
| 14 | 128, |
| 15 | 128 |
| 16 | ], |
| 17 | "learning_rate": 0.0003, |
| 18 | "n_steps": 1024, |
| 19 | "batch_size": 64, |
| 20 | "n_epochs": 4, |
| 21 | "gamma": 0.999, |
| 22 | "gae_lambda": 0.98, |
| 23 | "ent_coef": 0.03, |
| 24 | "clip_range": 0.2, |
| 25 | "eval_every_timesteps": 50000, |
| 26 | "checkpoint_every_timesteps": 100000, |
| 27 | "n_eval_episodes": 20, |
| 28 | "device": "cpu", |
| 29 | "initial_model_path": null |
| 30 | } |