---
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: -76.70 +/- 25.91
      name: mean_reward
      verified: false
---
# PPO Agent Playing LunarLander-v2

This is my Hugging Face Deep Reinforcement Learning Course **Unit 8 Part 1** submission.
The PPO agent was implemented from scratch with PyTorch and trained on the exact
course environment, `LunarLander-v2`.

## Evaluation

- Mean reward: **-76.70**
- Standard deviation: **25.91**
- Episodes: **20**

## Hyperparameters

```json
{
  "env_id": "LunarLander-v2",
  "seed": 1,
  "total_timesteps": 500000,
  "learning_rate": 0.0003,
  "num_envs": 8,
  "num_steps": 256,
  "num_minibatches": 8,
  "update_epochs": 4,
  "gamma": 0.99,
  "gae_lambda": 0.95,
  "clip_coef": 0.2,
  "ent_coef": 0.01,
  "vf_coef": 0.5,
  "max_grad_norm": 0.5,
  "anneal_lr": true,
  "norm_adv": true,
  "clip_vloss": true,
  "target_kl": 0.03,
  "save_every_updates": 20,
  "resume": true
}
```