q-learning-Taxi-v3

Q-Learning Agent — Taxi-v3 🚕

A Q-Learning agent implemented from scratch with NumPy for the Hugging Face Deep Reinforcement Learning Course Unit 2.

Environment compatibility

The current Gymnasium runtime uses Taxi-v4, which is the current compatible environment.

The project follows the Unit 2 Q-Learning methodology and evaluation protocol. The course target is Taxi-v3.

Evaluation

• Runtime environment: Taxi-v4
• Course target: Taxi-v3
• Mean reward: 7.56
• Standard deviation: 2.71
• Score (mean - std): 4.85
• Training episodes: 25000
• Learning rate: 0.7
• Gamma: 0.95

Files

• q-learning.pkl — trained Q-table and reproducibility metadata
• replay.mp4 — greedy-policy replay
• README.md — model card

Usage

import pickle
from huggingface_hub import hf_hub_download

path = hf_hub_download(
    repo_id="ritzie07/q-learning-Taxi-v3",
    filename="q-learning.pkl"
)

with open(path, "rb") as f:
    model = pickle.load(f)

print(model["qtable"].shape)