humnrdble/DeepRL-unit1-optuna

PPO Agent playing LunarLander-v2

This is a trained model of a PPO agent playing LunarLander-v2 using the stable-baselines3 library.

Made as part of the Deep RL course: https://huggingface.co/learn/deep-rl-course. Tuned with Optuna, as introduced in the course. This is my first successful attempt of using Optuna, so do not expect the code or parameters to be ideal!

I was able to improve upon my result from Unit1, https://huggingface.co/humnrdble/DeepRL-unit1. Both models were trained for 1500000 steps. The video of my first attempt certainly looks smoother, but scores worse.

The code is available in unit1-notebook-tuned.ipynb, but no attempt was made to make it particularly legible.

Hyperparameters deviating from the Stable-baselines3 baseline:

• gamma: 1-0.006075594024321983
• max_grad_norm: 1.8559426752164974
• exponent_n_steps: 9 (i.e. 2**9 steps)
• learning_rate: 0.0011176199638550707

Usage (with Stable-baselines3)

TODO: Add your code

from stable_baselines3 import ...
from huggingface_sb3 import load_from_hub

...