README.md

metadata

tags:
  - reinforcement-learning
  - q-learning
  - gymnasium
  - custom-implementation
model-index:
  - name: q-FrozenLake-v1-4x4-noSlippery
    results:
      - task:
          type: reinforcement-learning
          name: Reinforcement Learning
        dataset:
          type: FrozenLake-v1-4x4-no_slippery
          name: FrozenLake-v1-4x4-no_slippery
        metrics:
          - type: mean_reward
            name: Mean reward
            value: 1
          - type: success_rate
            name: Success rate
            value: 100

Q-Learning Agent playing FrozenLake-v1-4x4-no_slippery

This repository contains a tabular Q-Learning agent trained on FrozenLake-v1-4x4-no_slippery using Gymnasium.

The agent was implemented as part of the Hugging Face Deep Reinforcement Learning Course.

Evaluation results

Metric	Result
Evaluation episodes	1000
Mean reward	1.0000
Reward standard deviation	0.0000
Success rate	100.00%
Mean episode length	6.00

Agent replay

Repository files

• q-learning.pkl: Q-table, environment settings and hyperparameters
• qtable.npy: NumPy Q-table
• results.json: evaluation summary
• evaluation_episodes.csv: individual evaluation episodes
• training_history.csv: training history, when available
• replay.gif: animated agent replay
• replay.mp4: video replay, when available

Load the trained agent

import pickle
import gymnasium as gym

from huggingface_hub import hf_hub_download

model_path = hf_hub_download(
    repo_id="a1914114315/q-FrozenLake-v1-4x4-noSlippery",
    filename="q-learning.pkl"
)

with open(model_path, "rb") as file:
    model = pickle.load(file)

qtable = model["qtable"]

env = gym.make(
    model["env_id"],
    **model.get("env_kwargs", {})
)

print("Environment:", model["env_id"])
print("Q-table shape:", qtable.shape)
print(
    "Mean reward:",
    model["evaluation"]["mean_reward"]
)

Training configuration

{
  "n_training_episodes": 10000,
  "max_steps": 99,
  "learning_rate": 0.7,
  "gamma": 0.95,
  "max_epsilon": 1.0,
  "min_epsilon": 0.05,
  "decay_rate": 0.0005,
  "seed": 42
}