README.md

Overview

• train.py contains the main training script.
• models contains files for model architectures.
• tensor_mapping contains implementations of TensorMapping which manage the logic of translating between game states + actions and tensors

Getting started

From root folder:

uv run python -m  experiments.connect4.train --name="Test_run"

There are more flags one can pass; see train.py for the details.

Results

Below are some example plots from running the training script on my laptop. The interplay between policy loss and value loss is particularly interesting.