training_config.json
| 1 | { |
| 2 | "experiment_name": "ppo_lunarlander_flip_phase_d_soft_touchdown_01", |
| 3 | "env_id": "LunarLander-v3", |
| 4 | "environment_factory": "make_flip_soft_landing_lander", |
| 5 | "env_kwargs": {}, |
| 6 | "total_timesteps": 2000000, |
| 7 | "n_envs": 16, |
| 8 | "seed": 43, |
| 9 | "actor_layers": [ |
| 10 | 128, |
| 11 | 128 |
| 12 | ], |
| 13 | "critic_layers": [ |
| 14 | 128, |
| 15 | 128 |
| 16 | ], |
| 17 | "learning_rate": 5e-05, |
| 18 | "n_steps": 1024, |
| 19 | "batch_size": 64, |
| 20 | "n_epochs": 4, |
| 21 | "gamma": 0.999, |
| 22 | "gae_lambda": 0.98, |
| 23 | "ent_coef": 0.005, |
| 24 | "clip_range": 0.1, |
| 25 | "eval_every_timesteps": 50000, |
| 26 | "checkpoint_every_timesteps": 100000, |
| 27 | "n_eval_episodes": 30, |
| 28 | "device": "cpu", |
| 29 | "initial_model_path": "/content/rl-experiments/ppo_lunarlander_flip_phase_c_in_zone/checkpoints/ppo_lunarlander_flip_phase_c_in_zone_23700000_steps.zip" |
| 30 | } |