Downloads · 30 days
0
ppo-course-LunarLander-v2
bestdive/ppo-course-LunarLander-v2
ppo-course-LunarLander-v2 is a reinforcement learning model from bestdive. Use it for the reinforcement learning task on the model card, and read the license before you ship it in a product. The card lists the license as mit.
Kay Zheng's Unit 8 coursework, built with AI coding assistance. Trained from scratch using standalone PyTorch PPO with clipped policy objective, GAE and separate actor/critic networks. No pretrained weights.
Access
Public
Updated Sep 10, 2026
Repo size
43.2 KB
Likes
0
Public
Repo makeup
Click a slice to open those files.
.pt43.2 KB · 83%
At a glance
- Task
- Reinforcement Learning
- License
- mit
- Access
- Public
- Created
- Sep 10, 2026
- Updated
- Sep 10, 2026
- SHA
- 9bb5354f
Model card
From the Hugging Face model README
Standalone PPO — LunarLander-v2
Kay Zheng's Unit 8 coursework, built with AI coding assistance. Trained from scratch using standalone PyTorch PPO with clipped policy objective, GAE and separate actor/critic networks. No pretrained weights.
Training: 51,200 initial steps, then 204,800 further steps, seed 42. The optimizer was reset between phases. Total 256000 steps. Evaluation: 100 independent episodes, seeds 100000–100099. Mean -66.76560896596166, standard deviation 54.44184922283729, mean minus standard deviation -121.207458. This is an early-stage policy, not a solved LunarLander model. It meets the Unit 8 introductory threshold (-500); it does not meet the Unit 1 threshold (200).
The script supports resuming from policy.pt and evaluation.json. Install gymnasium==0.29.1, box2d-py==2.3.8, numpy<2, torch. For the original two phases use 100 updates first and 400 updates on resume.
- Task
- Reinforcement Learning
- License
- mit
- Created
- Sep 10, 2026
- Updated
- Sep 10, 2026