|
--- |
|
tags: |
|
- LunarLander-v2 |
|
- ppo |
|
- deep-reinforcement-learning |
|
- reinforcement-learning |
|
- custom-implementation |
|
- deep-rl-course |
|
model-index: |
|
- name: PPO |
|
results: |
|
- task: |
|
type: reinforcement-learning |
|
name: reinforcement-learning |
|
dataset: |
|
name: LunarLander-v2 |
|
type: LunarLander-v2 |
|
metrics: |
|
- type: mean_reward |
|
value: 301.97 +/- 19.65 |
|
name: mean_reward |
|
verified: false |
|
--- |
|
|
|
# PPO Agent Playing LunarLander-v2 |
|
|
|
This is a trained model of a PPO agent playing LunarLander-v2. |
|
The agent has been trained with a custom PPO implementation inspired to |
|
[a tutorial by Costa Huang](https://www.youtube.com/watch?v=MEt6rrxH8W4). |
|
|
|
This work is related to Unit 8, part 1 of the Hugging Face Deep RL course. I had to slightly modify |
|
some pieces of the provided notebook, because I used gymnasium and not gym. |
|
Furthermore, the PPO implementation is available on GitHub, here: |
|
[https://github.com/micdestefano/micppo](https://github.com/micdestefano/micppo). |
|
|
|
# Hyperparameters |
|
```python |
|
{ |
|
'exp_name': 'micppo' |
|
'gym_id': 'LunarLander-v2' |
|
'learning_rate': 0.00025 |
|
'min_learning_rate_ratio': 0.01 |
|
'seed': 1 |
|
'total_timesteps': 10000000 |
|
'torch_not_deterministic': False |
|
'no_cuda': False |
|
'capture_video': True |
|
'hidden_size': 256 |
|
'num_hidden_layers': 3 |
|
'activation': 'leaky-relu' |
|
'num_checkpoints': 4 |
|
'num_envs': 8 |
|
'num_steps': 2048 |
|
'no_lr_annealing': False |
|
'no_gae': False |
|
'gamma': 0.99 |
|
'gae_lambda': 0.95 |
|
'num_minibatches': 16 |
|
'num_update_epochs': 32 |
|
'no_advantage_normalization': False |
|
'clip_coef': 0.2 |
|
'no_value_loss_clip': False |
|
'ent_coef': 0.01 |
|
'vf_coef': 0.5 |
|
'max_grad_norm': 0.5 |
|
'target_kl': None |
|
'batch_size': 16384 |
|
'minibatch_size': 1024 |
|
} |
|
``` |
|
|
