Minesweeper-AI

💣 Minesweeper AI: CSP vs Probabilistic Reasoning

📌 Overview

This project implements an intelligent agent to solve the Minesweeper game using two independent AI approaches:

• Constraint Satisfaction Problem (CSP)
• Probabilistic reasoning (exact inference + Monte Carlo)

The goal is not only to solve the game, but to compare deterministic vs uncertainty-based decision making in AI systems.

---

🧠 AI Approaches

1. CSP-based Agent

• Knowledge-based reasoning
• Logical inference using constraints (sentences)
• Identifies safe cells and mines with certainty

✔ Strength:

• Guaranteed correctness when inference is possible

❌ Limitation:

• Fails when uncertainty appears (no safe moves)

---

2. Probabilistic Agent

• Builds constraint models from the board

• Uses:

  • Exact inference (backtracking)
  • Monte Carlo sampling (for large states)

• Computes probability of each cell being a mine

✔ Strength:

• Can act under uncertainty

❌ Limitation:

• Slower and not always optimal

---

⚖️ Comparison

Feature	CSP	Probabilistic
Deterministic	✅	❌
Handles uncertainty	❌	✅
Speed	Fast	Slower
Accuracy	High (when solvable)	Risk-based

---

🏗️ System Architecture

Game Engine (Minesweeper)
        ↓
AI Agent
 ├── CSP Reasoning
 └── Probabilistic Inference
        ↓
Decision Making

---

🎮 Features

• Playable Minesweeper (Pygame)
• AI autoplay mode
• Safe vs mine visualization
• Performance stats (win rate, speed)
• Two independent AI agents

---

▶️ Installation & Run

Install dependencies

pip install pygame

Run CSP agent

python game/runner_csp.py

Run Probabilistic agent

python game/runner_probabilistic.py

---

📊 Example Behavior

• CSP:

  • Plays perfectly when logic is sufficient
  • Stops when no safe move is known

• Probabilistic:

  • Chooses lowest-risk move
  • Continues even under uncertainty

---

⚠️ Limitations

• No learning (non-adaptive)
• Performance drops on large boards
• Probabilistic agent depends on sampling

---

🚀 Future Improvements

• Hybrid AI (CSP + Probability)
• Reinforcement Learning agent
• Performance benchmarking system
• Visualization dashboard

---

🧠 Key Takeaways

• Difference between symbolic AI and probabilistic AI
• Trade-offs between certainty and risk
• Constraint modeling and inference techniques

---

📸 Demo

-CSP

-Probability model

---

👨‍💻 Author

Pham Tan Minh

---

📄 License

This project is licensed under the MIT License.