KraTUZen/dqn-SpaceInvadersNoFrameskip-v4

👾 DQN Agent on SpaceInvadersNoFrameskip-v4

This repository contains a trained Deep Q-Network (DQN) agent that plays the SpaceInvadersNoFrameskip-v4 environment using the Stable-Baselines3 library (github.com in Bing).

---

📊 Model Card

Model Name: dqn-SpaceInvadersNoFrameskip-v4
Environment: SpaceInvadersNoFrameskip-v4
Algorithm: DQN (Deep Q-Network)
Performance Metric:

• Mean Reward: 565.50 ± 114.03
• Verification: Not yet independently verified

---

🚀 Usage

from stable_baselines3 import DQN
from huggingface_sb3 import load_from_hub
import gym

# Load the trained DQN model
model = load_from_hub(
    repo_id="KraTUZen/dqn-SpaceInvadersNoFrameskip-v4",
    filename="dqn.pkl"
)

# Initialize environment
env = gym.make(model["env_id"])

---

🧠 Notes

• The agent is trained using DQN, a value-based deep reinforcement learning algorithm.
• The environment is SpaceInvadersNoFrameskip-v4, a classic Atari game where the agent must shoot down alien invaders.
• The serialized Q-network is stored in dqn.pkl.

---

📂 Repository Structure

• dqn.pkl → Trained Q-network weights
• README.md → Documentation and usage guide

---

✅ Results

• The agent learns to maximize score by shooting invaders while avoiding losing lives.
• Demonstrates stable convergence using DQN, balancing exploration and exploitation.

---

🔎 Environment Overview

• Observation Space: Pixel-based visual input (Atari frames)
• Action Space: Discrete (move left, move right, fire)
• Objective: Survive and maximize score by destroying invaders
• Reward: Positive reward for hitting invaders, penalties for losing lives

---

📚 Learning Highlights

• Algorithm: DQN (Deep Q-Network)
• Update Rule: Q-learning with experience replay and target networks
• Strengths: Handles high-dimensional visual input effectively
• Limitations: Sensitive to hyperparameter tuning and replay buffer size