lihongxun945/gobang

What it solves

This project provides a browser-based AI opponent for the game of Gobang (Five in a Row), allowing users to play against a computer that calculates the best moves in real-time without requiring a server-side backend.

How it works

Unlike modern neural networks, this AI uses a classical game-theory approach based on the Minimax algorithm. It evaluates potential moves by searching through possible future game states to a certain depth. To improve performance and efficiency, it implements several optimizations:

• Alpha-Beta Pruning: Reduces the number of nodes evaluated in the search tree.
• Iterative Deepening: Gradually increases search depth to find the best move within time constraints.
• Zobrist Hashing: Uses a transposition table to cache previously evaluated positions.
• Killer Move Ordering: Prioritizes searching moves that have caused cut-offs in other branches.
• PVS (Principal Variation Search): Further optimizes the search process.

Who it’s for

• Casual players who want to play Gobang against an AI in their browser.
• Developers interested in learning how to implement classical game AI, search algorithms, and heuristic evaluation functions.

Highlights

• Pure Frontend Implementation: The AI runs entirely in JavaScript within the browser, meaning it works offline after the initial page load.
• Comprehensive Testing Suite: Includes a dedicated AI evaluation tool to measure Elo ratings, win rates, and node throughput using independent processes.
• Tactical Problem Set: Features a built-in library of tactical puzzles to detect "leaks" or errors in the AI's move selection.
• Customizable Difficulty: Search depth can be adjusted to change the AI's strength and processing time.