A program to train a simple neural net for the Tiger player of the boad game Baghchal.

The neural net training is implemented using PyTorch. A simple GUI is implemented using PySide6.

Trained models can be saved as onnx models using torch.onnx and loaded using onnx web runtime to run the models on web applications. One such saved model is deployed using onnx web runtime at https://sarojadhikari.github.io/TigerNet/.

• PySide6, for GUI
• PyTorch 2.0, for implementing and training the Convolutional Neural Network for AI game play
• torch.onnx: for saving trained models in the onnx format
• numpy, for array and mathematical operations

A screenshot of TigerNet GUI is given below. Red circles are Tiger players (which is moved by the neural net that we call TigerNet) and the blue circles Goat players which is played by a user. Initially the TigerNet has random parameters. After few rounds of training on random games (through the Train button), a noticable improvement in the TigerNet's game play can be observed.

The first 20 moves of the goat player are placing pieces on empty locations on the board. A tiger pice (red) can jump over a goat piece (blue) to an empty board location along a board line and kill the goat. The goat player wins if all the valid movements of the tiger player are blocked.