KraTUZen/q-FrozenLake-v1-no-slippery

🧊 Q-Learning Agent on FrozenLake-v1 (No Slippery)

This repository contains a trained Q-Learning agent that successfully solves the FrozenLake-v1 (4x4, no_slippery) environment.

---

📊 Model Card

Model Name: q-FrozenLake-v1-no-slippery
Environment: FrozenLake-v1-4x4-no_slippery
Algorithm: Q-Learning
Performance Metric:

• Mean Reward: 1.00 ± 0.00
• Verification: Not yet independently verified

---

🚀 Usage

from huggingface_hub import load_from_hub
import gym

# Load the trained Q-learning model
model = load_from_hub(
    repo_id="KraTUZen/q-FrozenLake-v1-no-slippery",
    filename="q-learning.pkl"
)

# Initialize environment (ensure no_slippery flag is set)
env = gym.make(model["env_id"], is_slippery=False)

---

🧠 Notes

• The agent is trained on the deterministic (no_slippery) version of FrozenLake.
• Ensure you set is_slippery=False when creating the environment to reproduce results.
• The model achieves optimal performance on the 4x4 FrozenLake grid.

---

📂 Repository Structure

• q-learning.pkl → Serialized Q-table of the trained agent
• README.md → Documentation and usage guide

---

✅ Results

The agent consistently reaches the goal state with perfect reward in evaluation runs, demonstrating stability and reproducibility in the deterministic environment.