basnetsoyuj/alphabaghchal

A deep reinforcement learning AI agent inspired by Alpha Zero that learns to master the traditional Nepali Board Game of Bagh Chal through self-play.

AlphaBaghChal

Bagh Chal is one of the many variants of the tiger hunting board games played locally in South East Asia. The strategic, two-player board game is played on a 5x5 grid.

Inspired by AlphaZero, the project tries to use state-of-the-art methods in the deep reinforcement learning paradigm to master the traditional Nepali board game of Bagh Chal through self-play. It uses a single deep residual convolutional neural network which takes in a multilayered binary board state and outputs both the game policy and value, along with Monte Carlo Tree Search.

Check out the accompanying blog where I write about my experience developing this self-learning AI agent: Mastering Bagh Chal with self-learning AI

Todos

• Tabula Rasa Learning
• Better training
• Pytorch version

alphabaghchal's Issues

Action Space

Your action space size is only 217.
Adding 2 No Moves for each.
And not treating bagh and goat's normal moves as separate makes it confusing for the policy network to find patterns.
Accounting that as well, the total Action Space size could be 331.